
## Stats

- Not sure how many hours worked
- Streamed 0 hours on Twitch (0% M/M)
- Stripe payouts $2,694.53 (+11% M/M)
- 295 Twitter followers (-5% M/M)
- 171 Twitch followers (0% M/M)

## FB and IG Ads

We still haven't had a breakout success with ads despite working with an agency. I am optimistic that we will get there but it may require a lot of iteration.

I am trying to build a system for producing organic content and I may try to produce simple videos of myself using and explaining the product. I want to hire a designer to help produce graphics, but I am also exploring talking to agencies to get this done.

## Harvesting reviews

Super curious about this one. We have a thumbs up button to "rate" the minutes. If someone clicks it, I want to ask them for a review in the website. Probably using trustpilot to start.

## Goals for Q1

1. 10x sales with an “insanely profitable” FB ads strategy
2. Extreme improvement to QSR product pillars

## Goals for February

1. Hire a Designer/Social media manager
2. Setup and integrate ManyChat
3. Post organic content
4. Automatically collect reviews
5. Ship a metrics-neutral (or positive) landing page redesign
6. Integrate the speaker system with minutes
7. Streamline minutes generation UX
8. Improve QSR


Monthly Retrospective - January 2025


## Stats

- Not sure how many hours worked
- Streamed 0 hours on Twitch (0% M/M)
- Stripe payouts $2,430.28 (+9% M/M)
- 312 Twitter followers (-2% M/M)
- 171 Twitch followers (0% M/M)

## QSR

I am adopting a framework for product development that I'm calling QSR, which is short for Quality, Speed, and Reliability.

### Quality

1. Product quality, like generating minutes with good content and formatting. Output that looks similar to what the customer really wants.
2. UI/UX that makes sense, feels natural, easy to use, simple.

Possible KPIs here:

1. Minutes thumbs up
2. Minutes thumbs down
3. % thumbs up
4. % thumbs down
5. % exported
6. % edited
7. Average edit distance, e.g. length of the diff
8. Rageclicks
9. "Bug report" intercom tickets

(7) is definitely an interesting one now that we have editable minutes.

### Speed

Pretty self explanatory, generally all perf should be fast.

Possible KPIs here:

1. Time from file upload -> preview minutes
2. Time from file upload -> full minutes (for non-paywalled)
3. File upload throughput, file_size / upload_duration
4. Sidebar navigation duration
5. Landing page -> dashboard navigation duration
6. Landing page load duration

### Reliability

Minimize crashes and avoidable failure modes. Inform the user on how to proceed.

Possible KPIs here:

1. SAD users, i.e. users with a negative event like transcribe_failed or on_drop_errored
2. % failed, i.e. the percentage of uploads resulting in neither transcribe_finished, or transcribe_paused

## Goals for Q1

1. 10x sales with an “insanely profitable” FB ads strategy
2. Extreme improvement to QSR product pillars

## Goals for January

1. Hire an FB ads specialist
2. Overhaul dashboard UI to be beautiful, Minutes, Paywall, Progress indicator
3. Create and track KPIs for QSR in Posthog
4. Run A/B test for redesigned landing page
5. Improve QSR


Monthly Retrospective - December 2024


## Stats

- Not sure how many hours worked
- Streamed 0 hours on Twitch (0% M/M)
- Stripe payouts $2,222.24 (-9% M/M)
- 318 Twitter followers (-46% M/M)
- 171 Twitch followers (0% M/M)

## FB Ads A/B test

We ran an A/B test with some new creative and the same targeting settings. Our initial testing showed a $0.35 cost per sign up. Although in December it's averaged out to $0.56. This is still better than our old rate by a factor of 2. Huge success!

We've started sending InitiateCheckout events in preparation to run a retargeting campaign for likely purchasers. We can also run a campaign like this for all website visitors period.

## Improving performance

We're close to bringing all ffmpeg work in-process which I think will have a huge effect on perf.

## Speaker labeling

Our biggest new feature to ship is using voiceprint embeddings to remember speakers across recordings. This should be a huge improvement for existing customers, and I'm hoping it reduces the time to get completed minutes.

Soon I think we should move to a model where labeling speakers can be done instantly, rather than requiring a re-generation of minutes. If we do this, it will mean we can show meeting minutes to users with zero button clicks. Right now we require the user to click a "Get your minutes" button after labeling speakers. We should just eliminate this piece of friction, and let then rename speakers cheaply and quickly. Also they can regenerate their minutes with feedback if they need to change something major.

## Goals for December

- Run a retargeting or lookalike ad campaign
- Improve server performance a lot
- Ship 2 major features


Monthly Retrospective - November 2024


## Stats

- Not sure how many hours worked
- Streamed 0 hours on Twitch (0% M/M)
- Stripe payouts $2,429.04 (+32% M/M)
- 590 Twitter followers (+5% M/M)
- 171 Twitch followers (0% M/M)

## FB Ads A/A test

We ran an A/A test with our FB ad to test if doubling ad spend doubles conversions. The test ran for one week, and the results were 25% worse on link clicks and cost per conversion (sign ups not purchases).

This is a pretty large difference but is potentially explained by the "control" A group having been running for months, so maybe more optimized by the FB system.

My takeaway is that when we run an A/B test, we need to beat the control on these metrics by 25% or more. Otherwise it may just be noise.

## SEO

An SEO course from an entrepreneur I follow on Twitter released this last week. I want to take a look at it this month and see if I can scope out some work to invest in SEO, as this is a high ROI channel for us.

## Improving performance

We shipped a huge upgrade to our diarization (speaker segmentation) ML model. There's a lot more work to do, but I think we can realize enough perf improvement to show the fast preview meeting minutes almost instantly. This may involve some tweaks to the open source whisper model as well.

I think we can get a lot of wins by using ffmpeg in process to hold a buffer of the decoded (as WAV) file. Then only serving ML models with this buffer. There's a lot of room for improvement.

## Shipped new features

- Sidebar and transcript redesign
- Upgraded diarization pipeline

## Goals for November

- Run one FB ads A/B test
- Scope and ship an SEO project
- Improve server performance a lot
- Ship 2 major features


Monthly Retrospective - October 2024


## Stats

- Not sure how many hours worked
- Streamed 0 hours on Twitch (0% M/M)
- Stripe payouts $1,840.35 (-14% M/M)
- 560 Twitter followers (+10% M/M)
- 171 Twitch followers (0% M/M)

## Hired another engineer!

We hired a new engineer, growing the team by 50%. In their first week they were able to propose and implement a better solution to a problem that we had around proxying requests to our server to avoid corporate firewalls from breaking our app. I'm super excited about it.

The bottleneck of the business is speed of engineering. We need to iterate faster to make the product something people want, faster, more reliable, and useful. Putting more energy into engineering is going to accelerate our growth.

## Prioritization

As we grow the team, I want to make sure we are executing on only high priority tasks. What does it mean to be high priority? The highest priority tasks are the ones which, when done, yield the largest increase in conversion rate, and also require the minimum amount of time to complete.

That is a working definition but I think it's ok. It optimizes for quickly increasing the conversion rate.

One issue is that it's not always obvious which tasks will increase conversion rates, or how long those tasks will take. In practice, you have to make educated guesses. Part of me feels like prioritization is a kind of boogey man, when the real answer is to just do work as fast as possible and learn along the way.

To try and uncover more information around what should be prioritized, I would like to build out an automated way of collecting feedback from users. I think eventually we can convert this into a system to automatically get testimonials.

Some known areas of improvement are still reliablity and performance. Improvements there will always translate to a better user experience, and I believe a higher conversion rate.

## Paywall A/B tests

We ran a couple small A/B tests with the paywall, though nothing has had a major effect. Low samples makes it hard to get reliable signal, you have to wait longer to collect more data. I think it would be worthwhile to A/B the headline and subheadline copy on top of the paywall.

## Shipped new features

- Adaptive concurrency (an intelligent multipart upload parallel upload).
- Streaming back minutes from the server (shortens time to value).
- Fast preview (increases perceived likelihood of success because users see preview of minutes)

Some iteration is needed on all of these, but super happy to be shipping these.

## Goals for October

- Email customers to ask for feedback
- Ship three major features


Monthly Retrospective - September 2024


## Stats

- Not sure how many hours worked
- Streamed 0 hours on Twitch (-100% M/M)
- Stripe payouts $2,134.02 (+23% M/M)
- 510 Twitter followers (+7% M/M)
- 171 Twitch followers (0% M/M)

## Building an engineering team

Things are going well. I spent some time this month thinking about how to set goals. There's a meme in the software engineering world about how it's bad to evaluate performance based purely on lines of code written. I agree with this, but ultimately I think you do need to have a framework to think about how to evaluate performance, and right now (at least for the phase the business is in) this is the right metric to look at.

I can envision a future where the (linear) commit history of the repo has commits added orders of magnitude more frequently than we do now. This is a leading indicator for product development, features shipped, performance and reliability improvements, and so on.

It's much more useful to set a goal for engineering around this metric because it is controllable. For example, "create one PR per day" is simple to operationalize. All you need to do is have a reliable process for creating work (e.g. find an existing Github issue), and a reliable process for doing that work (e.g. reproduce a bug locally and fix it).

Compare that to the goal of "doubling revenue". It's much harder to operationalize this because it requires experimentation and time. There is a degree of luck involved, and there's no clear roadmap that I know which will produce this outcome directly.

I do believe that if we consistently ship new features, bugfixes, perf and reliability improvements to production, then we will guarantee an increase in revenue over time, so number of PRs created is the way to operationalize more revenue. It's just not going to be a linear path probably.

As we ship, we will learn more, and this will compound over time.

I'm still interviewing for a second engineer. The goal is to find someone who raises the bar of the team.

## One time payment

We shipped a new payment option to pay for credits as a one time payment instead of a subscription. There were some issues here and our paywall conversion rate cratered. We've reverted the change, and the paywall conversion rate seems to be recovering.

As the team and product grows, a muscle we need to build is responding quickly to drops in metrics (like paywall conversion rate). This won't be the last time it happens, and we will have more opportunities to get reps in. Over time, we need to get better at shortening time-to-realizing-the-problem, and time-to-fixing-the-problem.

Two takeaways I have are:

1. When large features ship, we should check Posthog recordings and analytics a few times in the days following. This will help us understand what follow up work is needed and identify problems. A great place to do this is in 1:1's
2. Changes to the paywall are extremely high leverage. Two of the largest moves in conversion rate are from paywall changes: lowering prices from $27 -> $15 (increased CR), and adding OTP (decreased CR). An extra degree of caution is needed when we make changes to the paywall. We should get into the habit of running A/B tests for paywall changes.

## Optimizing time to paywall

We need to focus on performance so that number of users who make it to the paywall increase, and also these users will have a heightened perception of our product delivering them a result fast. Fast delivery is a core value prop, because a large part of the pain of writing minutes comes from how long it take to do manually. Other core value props are that it's boring/tedious/not fun, and that we can also do a much better job than someone who does it manually.

## Increasing Perceived Liklihood of Success

I think a big reason why people do not pay is that they are not sure they will get the desired outcome. We offer a partial transcript, but we do not show partial meeting minutes. There are some technical challenges to overcome here, especially if we want to do this fast, but I think it would go a long way to increasing users' confidence that paying will guarantee them a great outcome.

This is a priority.

## Offering Hosting

In the future, we may want to expand into B2B territory to upsell organizations on dedicated website integration for hosting past meeting minutes for public (or internal) consumption. We're not there yet, but it could be an interesting direction to expand the business.

## Increasing Import and Export Capability

We want to increase the number of ways to import and export inputs and outputs. For example, being able to import meeting recordings and transcripts directly from Google drive, Zoom, and MS Teams. This would also eliminate any slow network issues we face from customers on slow internet connections.

Similarly for exports, we can make the product more valuable by making it easier to use. For example, an option that allows users to export meeting minutes by emailing them.

## Paywall Abandonment Emails

We shipped an automated email sequence for paywall abandoners. Next steps here are to find ways to optimize the campaign, likely by A/B testing copy, and also generating meeting minutes in the background, and attaching them to emails. Then seeing if we can convince more people to come back because they like what they see.

## Goals for September

- Hire and onboard another engineer
- Ship two major features


Monthly Retrospective - August 2024


## Stats

- Not sure how many hours worked
- Streamed 0 hours on Twitch (-100% M/M)
- Stripe payouts $1,738.06 (+49% M/M)
- 475 Twitter followers (+8% M/M)
- 171 Twitch followers (0% M/M)

I'm going to experiment with a new format for these retrospectives. My goal is to be slightly more expository so it's easier to write them. I also will likely ditch the grading template I've been following.

## Hired an engineer

Last month, I wrote 'I'm optimistic that having more help on the engineering is going to accelerate the growth of the business dramatically.'

I hired an engineer on July 8, and it's been a good experience so far. Work doesn't happen as fast as it needs to yet, but I can see a world where I hire more, build out an engineering organization, and then things will go much faster. It's been very gratifying to see that work gets done even when I don't do it. It also helps motivate me to work when I can act in reaction to something, for example having code reviews requested.

I've also noticed that I need to spend more time thinking about developer experience/process type tasks, but these have been slowly getting ironed out, and I think will be mostly one-time-costs that subsequent employees won't need to worry about.

## What else happened

- We successfully migrated Stripe accounts after I set up a new one with Stripe Atlas.
- Added a feature to regenerate minutes with feedback
- Lowered the USD price to $15/$25.
- Grew by $400 MRR, 2-4x'ing our normal M/M MRR growth
- Set up Github, started tracking work in Issues
- Started working on a paywall abandonment email campaign

## Pay as you go pricing

I think we can increase paywall conversions by offering a pay as you go option that is not a subscription. I want to test this soon. Another nice element of this would be that we can offer preferred rates for subscribers.

For example, $7/hr on the free tier, $4/hr on Basic, and $1/hr on Pro.

We can also add a calculator to the landing page where users put in their meeting frequency, and we tell them the most economical plan.

## Other things I want to do

- Optimize time to paywall
- Collect feedback from regenerate minutes prompts
- Build a referral system
- Ship paywall abandonment emails
- Automatically collect testimonials
- Let users "try the product on the landing page"
- Hire another engineer

## Goals for August

- Hire and onboard another engineer
- Ship two major features


Monthly Retrospective - July 2024


## Stats

- Worked 53 hours (+67% M/M)
- Streamed 28.5 hours on Twitch (+54% M/M)
- Stripe payouts $1,162.78 (+16% M/M)
- 441 Twitter followers (+16% M/M)
- 171 Twitch followers (+2% M/M)

## June Goals

### Drive down failed transcripts to 0

Grade: F

The failure rate stayed flat around 20%. There's still more work to do here. I made a change mid-way through the month to stop a "banzai" effect where all parts of a multipart upload would be uploaded at once. This is counter-productive because it saturates the network's bandwidth, and it was also unintentional. Changing it to upload parts in serial didn't have an effect on failures but I think it's better still because it paves the way for building a more sophisticated upload manager which can implement two improvements.

First, it can progressively test the bandwidth and upload more parts in parallel if the bandwidth supports it. This means better networks can upload faster and not be penalized in a "least common denominator" way because we are trying to accomodate poor networks.

Second, (though less related to serial uploads) we can start uploads in the foreground initially, rather than waiting for the service worker to start he upload.

### Improve FB ads sign up conversion rate to 40%

Grade: D

This remained pretty flat with only a slight improvement around 25-35%. I'll credit the improvement to two things.

First, I implemented a sticky sign up form at the top of the mobile page as you scroll down. Second, I changed the headline to be a bit more punchy. I think a lot more work can be done on the hero section to increase conversions.

### Improve sign up -> upload conversion rate to 40%

Grade: C

Throughout June there was a slight uptick here, from an average of 40% to 50%. Potentially from our email drip campaign. I'll call this a C because apparently I wasn't measuring this correctly at first, and thought it was below 40% somehow.

### Improve paywall -> purchase conversion rate to 20%

Grade: F

This has been pretty consistently around 15-20% the whole month.

### Set up analytics for device type, acquisition channel, upload kind

Grade: F

Still need to set up acquisition channel.

### Set up landing page for new business

Grade: F

Didn't get to this. I'll probably remove this goal

### Hire an Intern

Grade: D

This didn't happen but there are updates in this direction, see the next section.

## Things I am Thinking About

### Hiring

Last month I wrote about hiring an intern, but after thinking about it (and interviewing one person) I realized that this is going to result in hiring someone who doesn't know much and requires a lot of mentorship. I don't mind coaching, but the cost to value tradeoff is pretty bad. In some cases, you can hire an unpaid intern but for software I don't believe it's common, and the laws about this in NYC are very strict around the idea that basically you have to lose money to train them. For me it likely doesn't make much sense.

Instead, I'm going a different direction. I'm currently interviewing Phillipines-based software engineers because you can find engineers who are much more senior for lest cost than a US-based engineer. So far I have interviewed a few dozen applicants, and there are some strong candidates.

I think my offer to these candidates can be a lot stronger, because I can set performance based compensation goals like "double revenue and you'll get a bonus" and still have that be close to break-even for the business. Versus an American engineer, I would need to make the goal "100x revenue" to maintain the same economics.

I'm optimistic that having more help on the engineering is going to accelerate the growth of the business dramatically.

### Product Quality

This goes back to the question "what is the bottleneck of the business?". I think it is product quality, and under this umbrella falls the tasks of improving conversion rates at every point in the funnel. Critically though, I want to start nurturing a new customer acquisition channel: word of mouth.

Word of mouth isn't going to happen unless the product is so good that people are willing to talk about it. This means that in addition to optimizing conversion rates, the work output needs to be completely stellar.

I want to start investigating how we can improve the quality of the generated minutes by looking at what we're generating already. I'm also going to build out a feature where you can regenerate minute with feedback to try and have a low-friction way to gather feedback from users.

Eventually for product-led growth, I think we'll need to add a referral system as well. Something like, "invite two people via email to the product and we'll give them 90% off their first month."

## Goals for July

- Hire an engineer
- Drive down failed transcripts to 0
- Improve FB ads sign up conversion rate to 40%
- Improve paywall -> purchase conversion rate to 20%
- Set up analytics for aquisition channel and upload kind


Monthly Retrospective - June 2024


## Stats

- Worked 31.75 hours (-48% M/M)
- Streamed 18.5 hours on Twitch (-66% M/M)
- Stripe payouts $1,003.22 (+40% M/M)
- 381 Twitter followers (+25% M/M)
- 168 Twitch followers (+6% M/M)

## May Goals

### Drive down failed transcripts to 0

Grade: F

Failure rate stayed flat around 20%.

### Improve FB ads sign up conversion rate to 40%

Grade: F

This was pretty flat between 20-30%.

### Improve sign up -> upload conversion rate to 40%

Grade: F

Uploads stayed flat.

### Improve paywall -> purchase conversion rate to 20%

Grade: C

Late in the month I made a change that may have boosted this conversion rate to over 20%, but it's hard to say with so few days of data.

I added a 14 day money back guarantee and tweaked some of the copy on the pricing table that shows on the paywall. I also ditched the no-code stripe paywall which had some usability and perf issues. Time will tell if the higher conversion rate sticks!

### Set up analytics for device type, acquisition channel, upload kind

Grade: D

Still need acquisition channel but I did add upload kind and device type. I need to revisit this.

### Set up landing page for new business

Grade: F

Didn't get to this.

## Things I am Thinking About

### Hiring an intern

This month I basically didn't do any work because I was travelling and on vacation. It was great but it made me realize that when I'm not working, nothing gets done.

My plan is to hire an intern and see if I can get more energy into the system even when I'm not working.

I think the goals will remain the same for this coming month.

The good thing is that MRR still went up!

## Goals for June

- Drive down failed transcripts to 0
- Improve FB ads sign up conversion rate to 40%
- Improve sign up -> upload conversion rate to 40%
- Improve paywall -> purchase conversion rate to 20%
- Set up analytics for device type, acquisition channel, upload kind
- Set up landing page for new business
- Hire an intern


Monthly Retrospective - May 2024


## Stats

- Worked 61.5 hours (-26% M/M)
- Streamed 53.75 hours on Twitch (-33% M/M)
- Stripe payouts $715.49 (+20% M/M)
- 305 Twitter followers (-2% M/M)
- 158 Twitch followers (+3% M/M)

## April Goals

### Drive down failed transcripts to 0

Grade: B-

I drove the failure rate from 46% to 9% over the course of the month, but they are ticking back up to ~15% in the first few days of the month. There is still more work to do.

This is due to a few improvements. The biggest cause of failures is failed uploads. Users with slow internet connections were leaving the site before the upload was complete.

Moving to a model where uploads are done in the background via the background sync api has helped more uploads succeed, but required a significant amount of engineering. The background sync stuff is more async than uploading in the foreground, so there is a time to first byte sent over the network tradeoff.

Even with background sync, some uploads would take too long, and so I also started storing the file locally in the browser via IndexedDB. Then on subsequent page loads, the service worker can resume an upload for unfinished chunks. The upload is "chunked" because I also switched the upload to a multipart upload.

Multipart uploads have a better chance of succeeding partially, because smaller chunks upload faster. Take too long and the user will leave, and if they don't leave, then there's a chance the service worker times out and kills your task.

The task can be killed due to system resource constraints, and there is no exit handler, so all retry logic has to have a timeout component. For example, "this chunk had an upload attempt longer than 2 hours ago, and isn't marked as completed locally, so we will retry".

I never imagined uploading a file would be this complicated.

Another factor here is that the Facebook and Instagram in-app browsers don't support background sync API, which means reliable uploads cannot be supported in this context. At first this seems ok - but actually all ad traffic starts in these browsers.

People would create an account and try to upload, only to have it fail, and essentially have no hope of being retried or run in the background. It was a huge problem.

I changed the entire sign up flow of my website for users acquired from paid ads because of this - more on that later.

Other failures were due to unsupported file types, and supporting Word document uploads fixed many of those - more on that later as well.

### Finish upload transcripts feature for Word docs and PDFs.

Grade: B+

I built this for Word documents but not PDFs. I'm happy we have a solution for folks here - Word docs are more common than PDFs.

The flip side is that only 15% of uploads are Word documents, and zero sales have come from paywalling this functionality. It's hard to justify investing in it with such a minority of users depending on it.

The paywall also comes in super slow and there are some other UI polish issues that could be better. But it's there.

### Make 15 sales via FB ads

Grade: A

We had 17 sales from FB ads this month!

This number needs to go higher because we spend about $1000/month, and average sale price is $20/month. Pessimistically assuming 100% churn, that puts us at $340, $660 short of break even for that acquisition channel.

### Ensure FB conversions API has 8/10 score or higher for Purchase events

Grade: A

We had a match score of 9/10 last time I checked.

We still don't get enough purchase events to reliably track this, but there was a huge improvement because it used to be 6/10 score.

The improvement is due to saving the FB click ids attatched to the URL when the visitor first hits the landing page and creates an account.

We can then always send the click id with the conversions API call when the customer makes a purchase.

### Build "email-only" invite sign up flow for FB ad traffic

Grade: A

I built it and it was great. We went from a sign up conversion rate of 5% to 25%, a 5x increase!

This is huge and I think there's still more work to do. The visitors to the landing page are engaged leads because they already clicked on an ad to get there. I think we should be able to get conversions to 50% at least.

Possible improvements are having the email form stick to the top as the user scrolls, and overall just making the landing page more clear.

### Improve landing page -> sign up conversion rate to 10% or higher.

Grade: A

See the above section.

## Things I am Thinking About

### Churn is down

April compared to March saw churn go from $199 -> $89.65, a 55% improvement!

I'm going to credit improvements to upload failures, and also lower prices causing higher customer retention.

### 4x plan

I made a to 4x my revenue. It has two parts:

Part 1 is to double my paywall -> purchase conversion rate which is currently 10%.

Part 2 is to double my sign up -> upload conversion rate which is currently 20%.

Both are low, and so have a lot of room for improvement. The plan is to programmatically set up an email campaign for both cohorts using the Instantly API.

We'll see where conversion rates end up, but I don't think it's crazy to imagine that we could double the current rates with good enough copy.

This is the biggest priority for May. If I 4x, then that will make the business profitable.

### OpenAI bills

My bills are a little higher than anticipated and may end up becoming a good target for optimization. 75% of costs are the whisper API, and this is an open source model. There may even be potential for running whisper in the user's browser.

I'm not going to optimize it until profitability, but it's on my radar.

### Analytics

I'm investing in improving my analytics so I can get a clear picture of what's working vs. not. The biggest dimensions I want to track are:

- Upload kind (Audio, Word doc)
- Customer acquisition channel (organic search, fb ads)
- Device type (desktop, mobile)

I can get some of this info now, but I have to do some analytics gymnastics to make it happen. I want it to be way easier.

Having this be easy is going to help identify where more investment is needed and where the big opportunities are.

### Disputed payment

We had our first disputed payment this month. The user let their subscription run too long and got billed for another month. When they realized, they had their card company dispute the payment rather than asking for a refund.

This scared me because I've heard horror stories of Stripe shutting down people's accounts. I pulled a list of users who hadn't use the product in the last 30 days and found about 15% of users fell into this category.

I'm considering implementing a system to email these "forgetters", but not sure exactly how long I should wait to email them, because it's not crazy to maintain a subscription for an extra month if you, let's say, take minutes every other month.

### New business

I have a new idea for a business that I think will print money. I'm going to carve out some time this month to at least set up a landing page for it. I don't want to say too much more yet.

## Goals for May

- Drive down failed transcripts to 0
- Improve FB ads sign up conversion rate to 40%
- Improve sign up -> upload conversion rate to 40%
- Improve paywall -> purchase conversion rate to 20%
- Set up analytics for device type, acquisition channel, upload kind
- Set up landing page for new business


Monthly Retrospective - April 2024


## Stats

- Worked 83.25 hours (+8% M/M)
- Streamed 80.75 hours on Twitch (+124% M/M)
- Revenue $596.44 (+89% M/M)
- 311 Twitter followers (-18% M/M)
- 154 Twitch followers (+18% M/M)

## February Goals

### Finish upload transcripts feature

Grade: C

Right now we support uploading text file transcripts (basically .vtt, or .txt). For most of the month this feature wasn't paywalled and did have some users discovering that they could upload all their transcripts for free. As of last night, it is paywalled, and remains to be seen what the conversion rate is.

I am happy that there will be more opportunities to show the paywall to users. Our number 2 cause of failures is from people uploading Word document transcripts.

I give myself a C grade because Word documents are still not supported. I investigated nice ways of showing Word documents in the UI and there aren't really great options. The best I've found is an iframe that you can host which sends the document to Microsoft, and renders it in the iframe. It works ok, but the styling options are limited so it looks a bit weird, especially on narrow screens. I'll probably try to go this direction anyway and ship something soon (TM).

### Make 5 sales via FB ads

Grade: D

Out of 11 new customers, only 3 definitively came from ads I ran. 3/5 = 60% which is a D. The rest of the customers came from google search.

There was a big win this month with FB ads because I discovered that CPMs for South Africa are very low, and was able to set up an ad campaign with very low cost of conversion, albiet for sign ups, not payments.

This has led to a flood of traffic and users, and I've been iterating on the product to try and fix all the failures that I've noticed.

I also localized the price of the product, including listing the price in Rand which is the currency in South Africa. I am already seeing some positive response here, so hopefully next month I'll be able to report that my ad spend buys more than it costs.

### Drive down failed transcripts to 0

Grade: D

We are not even close but instead of an F letter grade I get a D "for effort". I've spent a ton of time this month on improving and understanding reliability.

The site has gone from a single PUT request made from the web app to using service workers that can run in the background, and which do a multipart upload to S3.

The uploads are more reliable but there's still a lot more work to do. For one thing, the in app browser that opens when you click an FB ad doesn't support service workers. I've attempted to direct users to open the system browser before they sign in, but I think ultimately I'll just have users in this browser give me their email, then email them a login invitation. This will open in their system browser 100% of the time, and prevent people from using the product in the FB browser.

Safari and Firefox don't support service workers either, but fortunately these are not very popular browsers.

Ultimately this is still a work in progress, and it takes time to see how changes affect production.

Second biggest cause of failures is uploading Word documents, so that's getting fixed this month too.

### Measure finish transcribe time, and label time (create minutes start - transcript finish)

Grade: F

Didn't do it. I did relax the restriction on needing to label all speakers before generating minutes. GPT can usually discover the speaker names by reading the transcript, so it's a bit of a waste of time. I had a paying user who was very frustrated because they couldn't figure out the labeling system.

You still have the option to label speakers, but I think it's best to just leave it as optional.

### Use server side rendering and improve perf

Grade: B+

We use server side rendering now and it's a much better experience than it was, e.g. parts of the UI load faster. I think there's a lot of room to run in terms of overall performance of the app, but it's not the bottleneck.

What is the bottleneck? Probably upload failures, because even if a user becomes a customer by paying for a plan, they may still encounter failures that result in churn. And users considering becoming customers, often hit (mostly upload) failures, which in turn prevents them from becoming customers.

After that, I think the bottleneck is likely landing page conversion rate because 94% of vistors bounce. 90% of these visitors are people who literally clicked an ad, so given that context, this is especially terrible/can be improved.

## Things I am Thinking About

### Churn, Reliability

It's still the top priority. Close to or exactly 100% of uploads should succeed. It's the most tenuous part of the process because I am depending on the users' conditions being right, e.g. having internet that's fast enough to get the upload off reasonably quick.

I've already written about it above so won't rehash it here.

### Mobile

This month I wrote a mobile version of the website! Many users are on mobile when they come from the FB ad and while there are still upload failures, it feels good to have a better UI for such a common use case.

### New domain

I tried to purchase a new domain name that I think will give better SEO ranking and got totally screwed by name.com! They basically took my money (thousands of dollars) and then silently failed to purchase the domain. After complaining multiple times to customer support they are in the process of refunding me, but I'm expecting to lose a few hundred dollars because I was charged a currency conversion fee by my bank when making the transaction.

It's a really bad experience, and I'm considering migrating all my domains away from them, though not sure to who.

### Targeting purchase conversions

My FB ad campaign is targeting the Registration conversion event (sign ups). I did this because their docs recommend targeting an event that you will receive a few dozen of within a week.

I am guessing that there are a lot of people in this cohort that are likely to sign up, but not to purchase for whatever reason. I'd like to bootstrap an ad campaign targeting Purchase events, and I think it will be possible once I get enough customers from the current campaign to buy subscriptions.

## Goals for April

- Drive down failed transcripts to 0
- Finish upload transcripts feature for Word docs and PDFs.
- Make 15 sales via FB ads
- Ensure FB conversions API has 8/10 score or higher for Purchase events
- Build "email-only" invite sign up flow for FB ad traffic
- Improve landing page -> sign up conversion rate to 10% or higher.


Monthly Retrospective - March 2024


## Stats

- Worked 77 hours (-18% M/M)
- Of that work, 36 hours was done while streaming live on Twitch
- 47% of work was done live on Twitch
- Revenue $316.05 (+31% M/M)
- 379 Twitter followers (+28% M/M)
- 130 Twitch followers (+15% M/M)

## January Goals

### Build out feature to upload transcripts

Grade: D

I started this a few days ago and I want to finish this week. A lot of people upload .docx files believing that this will work, and it does not. It's the number one source of "failed transcripts", and I think there is a big opportunity here. Especially if I advertise this on the top of the landing page.

### Make 1 sale via FB ads

Grade: A

We made 2 sales with FB ads this month. The ad campaign is unprofitable right now, but I'm going to continue to optimize it and eventually make it work. The next targeting refinement is likely to add non US countries which supposedly have a lower CPM.

### Create metric for transcript quality

Grade: F

Didn't work on this yet.

### Create metric for minutes quality

Grade: F

Didn't work on this yet either.

### Run 1 survey

Grade: F

Didn't do this.

None of these goals felt like priorities over making the ads work well.

### Run 1 A/B test

Grade: A

I'm currently A/B testing a section on the landing page with a CEO letter. Right now it doesn't look like it moves the needle and I'll probably end up shipping it soon.

A more effective test would be something that gets more traffic rather than the bottom of the landing page.

## Things I am Thinking About

### Churn, Reliability

We had 4 customers churn this month from last, and 2 of 3 customers this month set their subscription to end the following month. I want to make sure people have a good experience. As we get more users from spending on FB ads, it becomes more apparent that many people have a suboptimal experience.

I integrated Sentry this month and I'm optimistic it will be helpful in identifying errors. I want to spend time this coming month driving down errors and transcript failures. Failures happen because of many reasons, so having good observability is critical.

A large cause of failures is people uploading .docx transcripts, so fixing this class of failures is going to involve finishing the goal from last month to allow uploading text transcripts.

Performance is another side of the story - a slow website makes people less likely to buy.

## Goals for March

- Finish upload transcripts feature
- Make 5 sales via FB ads
- Drive down failed transcripts to 0
- Measure finish transcribe time, and label time (create minutes start - transcript finish)
- Use server side rendering and improve perf


Monthly Retrospective - February 2024


## Stats

- Worked 94.25 hours (+38% M/M)
- Of that work, 55.5 hours was done while streaming live on Twitch
- 59% of work was done live on Twitch
- Revenue $241.37 (+834% M/M)
- 297 Twitter followers (+9% M/M)
- 113 Twitch followers (+11% M/M)

## December Goals

### Finish minutes template feature

Grade: D

I built out the UI and backend for this feature most of the way but didn't ship it yet. Why not? There are two or three reasons:

Reason 1 is that I got bogged down by actually trying to make the feature work well. The idea is to inject a template for the document into the GPT prompt, but creating the template is a can of worms. Which parts from the example doc get kept, and which get discarded?

I hit problems with GPT including information from the example doc in the generation for the current transcript, so if I'm not careful this can regress meeting minutes quality. Quality is something I already struggle to measure.

Reason 2 is that I made a huge improvement to meeting minutes quality that achieved much of the spirit of this goal. I had a user "thumbs down" their meeting minutes at the beginning of the month, and when I reached out, they told me they wanted more bullet points, and sections for key takeaways and next steps.

I tweaked the GPT prompt, and this user mentioned the minutes were much better. They seem a lot better to me as well, so this is a small victory.

Measuring minutes quality is an open question. I've thought about looking at number of button clicks per transcript for the "Copy to Clipboard" and "Export to Word" buttons. I think if I track this, along with listening for browser copy events (for people who manually copy) then I could get a sense of, at least, who is exporting the minutes out of the product.

Reason 3 is that, ultimately, once the GPT prompt fix went in, I didn't perceive this to be a bottleneck for the business. That's more based on my intuition, but it seems to me like getting more users is the bottleneck. What if the users are getting poor quality minutes, leading to churn? Well that will be hard to know without getting more user data and feedback. So the bottleneck is probably still users.

### Make 1 sale

Grade: A+

We acquired 9 new customers this month! This is really awesome, because I raised prices on Jan 1. In December, we acquired 4 customers, at a price point that is 6x less than the new price.

Users continue to be driven entirely by organic search, thank you Google and Microsoft.

### Send 3000 emails

Grade: F

I am cold emailing youtubers to offer them affiliate partnerships. They make youtube videos about MinutesGenerator, and they get 30% of sales.

I sent 705 emails across 186 unique leads (includes followups) with a 16% reply rate (30 replies). Of these 30 replies there has been some interest, 3 calls, and one video made! This video did not result in any sales, but is a nice milestone at the end of a long funnel.

I will say that I underestimated the work required to get this program off the ground. It's draining to respond to emails all the time, and I think about hiring this out a lot.

Hiring an SDR to manage the process sounds attractive, but I think it will cost $25/hour, and optimistically is 5 hours of work per week, so 20 hours per month = $500/month. If it's 10 hours per week, then $1000/month. I think I would need to make around 50 sales to really make the program profitable.

In part, I am losing steam on the effort because I think it may be much more effort than running FB ads. FB ads will also allow me to scale up or down more easily if I need to manage bottlenecks that arise (product quality, reliability issues, customer support burden, etc). I think I can also do it without needing an employee like an SDR.

### Build Zoom integration

Grade: F

Didn't even start on this one. Instead I set up analytics with Posthog. I'm glad I did, this is higher priority, and nobody is asking for Zoom integration.

### Build or buy a customer feedback corner widget

Grade: A

I setup Intercom! This has been pretty smooth sailing so far, and several users have used it to ask for support, which is great.

It's nice to have a clear pathway to being able to hire customer support staff when that becomes necessary.

## Things I am Thinking About

### FB Ads

I touched on this already, but I am getting increasingly excited about running ads on Facebook. I'm going to do it this month and see if we can generate any sales. I am hoping that it will provide a pathway to 100x'ing the business in a sustainable way.

I took a copywriting class this month and I think I can write some decent copy for an ad. I am not clear on the end to end process, but the first thing I did in February is setup the conversions API from Meta to optimize future ads, so hopefully we can get some momentum going here.

I think spending $1000/month of ads will generate more revenue than the same spend on the affiliate program for now. I want to do both, but I think ads may be more efficient, so makes sense to do first.

### Raised prices success

From last month:

"I raised prices Jan 1 after a 2 week long email campaign to existing users. That yielded 1 sale which is about a 1.5% conversion rate. I'm interested to see if new users will pay the higher price. Fingers crossed!"

More people buy at the higher price! Crazy, and definitely an unlock for many things, e.g. ads, hiring, acceleration.

### Analytics

I set up PostHog this month to track analytics and built a dashboard. It's been pretty good so far, definitely a learning curve around how to instrument things so that querying is easy. Ultimately I think this is going to be an unlock. Some things I'm excited about:

1. Dashboard has a failures chart so I can notice when failures happen and fix them, ideally this reduces churn.
2. A/B testing is possible, so I can experiment with things, including changing the copy on the landing page and measuring signup rate changes.
3. It offers surveys, so I can ask existing users "what is the worst thing about MinutesGenerator?" and have them tell me things to fix.
4. Paves the way for understanding traffic from affiliates and FB ads. For example, understanding what percentage of users who click an FB ad will purchase.
5. Gives a way to understand churn and lifetime customer value

### Uploading transcripts feature

5% of transcript uploads are of a .txt, .docx, or .pdf file type. This currently results in an error, but I think it is a revealed product feature request. How many users bounce from the landing page when they discover we don't do transcript -> meeting minutes?

I'd like to implement this feature, advertise it on the landing page, and see if we can improve the signup rate.

I'm not sure how to monetize it because the credit system works based on the length of the recording uploaded. No recording, no length. Some options might be:

1. Make it free at first.
2. Make it cost a fixed number of credits, maybe 30? That fits into the free 40 credit trial. Or make it cost more than the free trial to prevent free trial abusers.
3. Offer two types of credits, one for recordings, one for meeting minutes.
4. Pay fixed amount per use.

As I write these out, I like (2) the best. I could try out 30 credit cost and see how many free trial abusers there are (this happens currently, occasionally people make multiple accounts to use the free trial).

### Improving the model

I want to add some UI to "fix" a transcript that's been generated, either by combining multiple lines of dialogue into one, or by mis-labeling the speaker of a line of dialog. A third issue is dialogue that is not captured by the current model.

Creating a feedback loop is important for eventually fine-tuning the model to be better. It also lets me assess the current quality of the product.

UI will probably go in the speaker labeler, but I'm not sure exactly how it's gonna look.

Creating feedback loop for the transcript and minutes is important.

## Goals for February

- Build out feature to upload transcripts
- Make 1 sale via FB ads
- Create metric for transcript quality
- Create metric for minutes quality
- Run 1 survey
- Run 1 A/B test


Monthly Retrospective - January 2024


## Stats

- Worked 68.5 hours (-23% M/M)
- Of that work, 49.25 hours was done while streaming live on Twitch
- 72% of work was done live on Twitch
- Revenue $25.84 (+82% M/M)
- 272 Twitter followers (+36% M/M)
- 113 Twitch followers (+4% M/M)

## November Goals

### Measure detailed timing information

Grade: A

I record times of sections that happen when users go through the flow of the product. They are stored next to the data for transcripts so everything is in one place and I am very happy with it since it lets me see users' upload speed and gives me a sense of how long folks are waiting around, identifying bottlenecks, etc.

### Allow users to upload template meeting minutes

Grade: D

I've started this but it's not shipped yet. I'm hung up on how to inject the template into the GPT prompt, but I think I'm getting close to something workable.

The UI is built out most of the way, I'd estimate there's 6 more hours so I am hoping to be done by the end of the week.

In general, it's hard to assess how good the system does at creating minutes, especially if the goal is to do fine tuning. I think this ties into another issue which is that I'm not soliciting customer feedback very often and should have a way to do it consistently. Maybe an email campaign to paying customers.

### Send 10 cold emails to "admin professional"-relevant youtubers

Grade: A

I did this over the holidays and got a 0% reply rate. I'm in the process of setting up a larger campaign soon to hit 1000 prospects with an email sequence.

### Make more sales in December than I did in November (5 or more)

Grade: B

I made 4 sales, so same amount but not greater. I think this was because of a holiday slow down.

### Average 20 hours of work per week

Grade: C+

I needed 88.5 hours and did 68.5 hours, so fell short by a decent amount - it was a chiller month and also holidays.

## Things I am Thinking About

### Hiring and automation

I'm reading Buy Back your Time and I'm trying to think about how I would offload any of the tasks that I do when I sit down to work. Right now I'm feeling a tension between getting profitable by increasing revenue significantly, and optimizing for my own time.

I think what makes sense right now is to devote all efforts into acquiring more customers faster. More customers will give me more data for conversion rates and also qualitative data about how users use the site and what they want.

### Raised prices impact

I raised prices Jan 1 after a 2 week long email campaign to existing users. That yielded 1 sale which is about a 1.5% conversion rate. I'm interested to see if new users will pay the higher price. Fingers crossed!

## Goals for January

- Finish minutes template feature
- Make 1 sale
- Send 3000 emails
- Build Zoom integration
- Build or buy a customer feedback corner widget


Monthly Retrospective - December 2023


## Stats

- Worked 89.5 hours (-20% M/M)
- Of that work, 46.25 hours was done while streaming live on Twitch
- 52% of work was done live on Twitch
- Revenue $14.17 (+339% M/M)
- 200 Twitter followers (+3% M/M)
- 109 Twitch followers (+7% M/M)

## October Goals

### Record a long form podcast episode with a virtual assistant

Grade: F

I didn't end up doing this, but I did find more clarity on my ideal customer profile, namely: administrative professionals. This can include virtual assistants, but isn't limited to them. I'll talk about future plans for marketing later. Ultimately, I decided I can find a more efficient way to reach audiences.

### Time to Transcript for MinutesGenerator should be under 10 minutes per hour of recording

Grade: A-

The real value of doing this is to decrease the time to delivering value, resulting in a better customer experience, resulting in less churn and more revenue.

I removed the inital text prompt and the transcribe button - just upload your file and things will start automatically. You literally have to do one thing and it's really easy. I spent a lot of time optimizing the transcription flow after upload. I rented a GPU server so I didn't have to wait for my API provider's cold start boot times. I started streaming back results to the client from the server as soon as they were ready, instead of waiting for everything to be ready before showing. I created a progress bar that communicated where in the process the job was, and gives an ETA for being done. Other stuff too, but ultimately the take away is that people hate waiting, and the faster my service gets, the better experience people have. That reduces churn, and boosts the potential for sales. Waiting sucks and speed is a feature.

This goal is an A- not an A because I am almost certain every successful transcription in the current system is well under 10 minutes. However, I don't yet have analytics on how long each step in the process takes, so this is based on usage I have seen personally. I keep watch very diligently, so I am confident we're there, but speed is so critical that it has to be measured and tracked.

The north star is to create the best software in the world for generating meeting minutes, and that software should produce instant results.

All the optimization I did in November was such low hanging fruit that I didn't need to measure to see performance improvements. The improvements were obvious. Going forward, I have optimization ideas that are more experimental, and I want to track perf metrics more rigorously to make sure everything gets faster and stays fast.

### Make one sale

Grade: A

I made four sales this month! All with zero marketing - customers came through organic bing search. I was worried about this goal at the beginning of the month because it's more of a lag measure than a lead measure, for instance I can control sending X cold emails in a month, but it's harder to control the outcome of that activity.

I had a tiny bit of confidence that sales would come because I could see people signing up for MinutesGenerator, but then hitting issues with the product and then leaving the site. The biggest issue I noticed was that it took a long time to receive any value from the site. A user would sign up, and then be faced with three things to do: upload a recording of their meeting, write a text description of any names or proper nouns in the meeting, and then click a button to start transcribing.

People would do these steps, which is a testatment to how much they do not want to write their own meeting minutes. After clicking the button, they would see a loading spinner and transcription would take on average 45 minutes to finish. Super slow! There was no feedback things were working, most users assumed the product didn't work and would leave the site (many times this assumption was correct and no error was surfaced).

I realized that nobody wants to pay for a shitty product that doesn't work (duh). I also realized that I can make the experience more compelling by doing two things:

First, speed it up. This is what the last section is about so we already know about it.

The second strategy is to show results to the user as early as possible. If the customer sees the first few seconds of audio transcribed, they get more confident things will work. I think once they get invested, a better outcome is more likely.

I changed the product so that it produces a partial transcript, and if the audio is longer than the free trial allows, an upsell pricing table appears asking you to upgrade your plan to continue transcribing. The partial transcript shows the user this will work, which boosts confidence, which boosts likelihood of them paying. That's the theory anyway, and it seems to be working so far!

### Average 20 hours of work per week

Grade: A

I averaged 20.9 hours worked per week, slightly above my goal. This goal combined with live streaming my work has been the most transformative for my productivity. It's nice to have something measurable I can track.

## Things I am Thinking About

### Iterating on the Money Printer

I get a steady stream of real user feedback now and it's been helpful to see where the product breaks. My mental model is that MinutesGenerator is a machine that other people interact with. I input money and time into the machine, and users input files, clicks, scrolls, etc. into the machine. The machine outputs meeting minutes and also sometimes the machine outputs money. Most of the time something gets messed up, the machine breaks and outputs nothing and the user leaves unsatisified.

I've started looking at every visitor to the site as beginning a sequence of hundreds of small steps. Every step has to go perfectly in order to make money. Any of these steps can break, for example the user uploads a file and it takes too long, or they click a button and there's a UI bug, or they pay money and our stripe webhook handler isn't set up correctly, or they read something on the landing page and got confused, etc etc etc.

It doesn't matter if a failure happens on the first step or the 100th step - if something fails, it's game over and you make zero money and probably pissed someone off.

Because of this, I look very closely at how every user's experience is. If they encounter any failure at all, I immediately prioritize a fix so it can't happen again. I have done this for one month and there are still problems, but I can see the improvements reflected in sales numbers as well as successful meeting minute generations.

It's tempting to think that you can create a machine that works well without observing user behavior, but I think it's extremely difficult. For example, here are some differences between you and a new user - you may have a different OS, browser, internet speed, screen dimensions, mobile device, familiarity with online tools, file formats used with testing, expectations for what meeting minutes look like, expectations for meeting minutes file format, reading level (or else bias that your own copy writing is easy to understand), and on and on and on and on.

### Cold Email Influencer Outreach

One arm of my strategy is building a best in class product that users have an overwhelmingly positive reaction to. I want churn low and to retain users forever.

The other side of the story is the process of making known to strangers. To do this, I'm going to try and form some affiliate partnerships with creators who have audiences that have administrative professionals. I spent some time this month building a list (and a system to build lists) to do some email outreach. I'm currently warming up some email accounts to start doing cold outreach, and I'm hoping that I can land a few partners and drive hundreds or thousands of sign ups.

For example, if I can get a few creators to produce videos that get total 10,000 views, and this has a 1% conversion rate, then this would be 100 sign ups, and would 25x my current sales numbers. Part of this plan is going to include raising prices, and offering exclusive affiliate discounts. If I can generate $15/month subscriptions, then this would make MinutesGenerator profitable with an $18,000 annual recurring revenue.

In general, the plan is to build out, optimize, and automate customer acquisition channels. Right now organic search is my only channel. Becoming more well known will help this channel's efficiency improve via higher search result ranking.

## Goals for December

- Measure detailed timing information
- Allow users to upload template meeting minutes
- Send 10 cold emails to "admin professional"-relevant youtubers
- Make more sales in December than I did in November (5 or more)
- Average 20 hours of work per week


Monthly Retrospective - November 2023


## Stats

- Worked 111.75 hours (+35% M/M)
- Of that work, 47.25 hours was done while streaming live on Twitch
- 42% of work was done live on Twitch
- Revenue $3.23 (from Twitch)
- 195 Twitter followers (+8% M/M)
- 102 Twitch followers (+4% M/M)

## October Goals

### Publish 100 TikToks

Grade: F

I published about a dozen tiktoks with no viral successes.

### Get 1000+ views on at least one TikTok

Grade: F

All tiktok's published never made it out of 200 view jail.

### Create QOTD and share it every day on Twitter thereafter

Grade: F

I didn't do this because this wasn't my focus after discovering that MinutesGenerator was getting users via Google search.

### Get approved for Adsense on [AddictingTrivia.com](https://addictingtrivia.com)

Grade: F

This didn't happen.

### Average 20 hours of work per week

Grade: A

I worked 25.23 hours per week on average.

## Things I am Thinking About

### Pivoting back to MinutesGenerator

I discovered midway through the month that my first business [MinutesGenerator](https://minutesgenerator.com) has been steadily seeing signups via organic Google search traffic.

This flew in the face of an insecurity I had about the business, which was that people wouldn't want to use or pay for a service that I perceived to be less valuable then the well-resourced, older, more feature-rich competitors in the space like Otter AI and others.

I decided to try an invest more time into making MinutesGenerator work with a focus on generating leads. I decided my ICP was VA's and EA's, who as part of their job responsibilities have to take meeting minutes.

I found some EA and VA Facebook groups and was able to find that this demographic does seem interested in a tool like this, and got some feedback on the price being too high, and the time to transcript being too long.

I took another look at the product and decided to do a major quality push to try and tighten up everything and make the product super reliable and fast. For example, sometimes generating a transcript would take over an hour, my goal is to make this take under 10 minutes and ideally be fast as possible. Realtime is best.

I also changed the pricing model to be subscription based. Two tiers, a $5/mo and $10/mo option with 5 hours of transcription for the lower tier, and 20 hours per month for the higher tier.

I'm going to rent a GPU server to eliminate cold start time from Replicate instances, and try to sell enough subscriptions to cover the cost. For example, an A100 from vultr is $800/month, so that would require about 80 sales.

I want to start creating VA specific content to try and build an audience, I think I'm going to try and create a podcast to interview VA's and see if I can learn about their workflow.

## Goals for November

- Record a long form podcast episode with a virtual assistant
- Time to Transcript for MinutesGenerator should be under 10 minutes per hour of recording
- Make one sale
- Average 20 hours of work per week


Monthly Retrospective - October 2023


## Stats

- Worked 82.5 hours (-2% M/M)
- Of that work, 64.75 hours was done while streaming live on Twitch
- 78% of work was done live on Twitch
- Revenue $6.57 (mainly from Twitch prime subscriptions)
- 181 Twitter followers (+4% M/M)
- 98 Twitch followers (+34% M/M)

## September Goals

### Publish 1000 quizzes on AddictingTrivia.com

Grade: C+

I published 776 quizzes across 81 different topics selected due to their high traffic and low "keyword difficulty" according to Ahrefs.

I'm pretty happy with the approach that I took on this, which was end to end automation. This includes GPT4 generating quiz questions, descriptions, titles, and topics, as well as using the Bing image search API to automatically find thumbnails, and the Google indexing API to automatically request indexing of the page.

As a result, all of these quizzes are indexed by google and I already do see a small trickle of organic search traffic for my oldest quizzes.

Time will tell if the organic search traffic can be converted into a productive revenue stream, but I'm looking forward to running the expriment either way.

I've been thinking more about traffic acquisition, and have decided that I need to pursue additional channels, but I'll write about that below.

### Get one page on AT to rank in the top 10 results on google for any search query

Grade: B

The Google search console data is pretty broad, but it does show that I am ranking 9th position or above for few search queries like "football tactics quiz".

When I search for this manually, I don't see my website, so I'm giving myself a B because the data isn't lining up with reality. I also notice that my ranking is trending down over the month, but I think this may be because of introducing so many new pages, which should skew the averages.

I would love to see the ranking data per query but I don't see a way to do it in GSC. Regardless, AddictingTrivia has very low traffic, and I would expect SEO to be more of a long game. More and more my mindset is shifting to view SEO as a secondary reinforcing traffic source, as opposed to a primary channel.

Said differently: SEO is slow and I don't want to wait for traffic.

### Average 20 hours of work per week

Grade: A-

I averaged 19.25 hours per week - just short of my goal.

I love this goal because I am confident that if I work consistently that I will be successful after enough time.

I'll have the same goal this month.

### Investigate a solution for analytics that I can use for quality control

Grade: A

I have two channels for feedback, one implicit and one explicit. The explicit channel is a thumbs up/down button per question that lets a user indicate if a question is good or bad. The implicit channel is telemetry on the success rate per question, which tells me if a question is hard or bad, as well as if people are navigating away from quizzes without finishing.

I don't do anything with this data but if I can drive traffic to the site, then I can use it to automatically discard questions that are poor quality. This is especially important because sometimes the AI generated questions are ambiguous or have an incorrect answer marked as correct. That's a bad experience and this telemetry offers a low cost way to remove these low quality questions.

## Things I am Thinking About

### Adsense

I am currently awaiting a decision on my 3rd application for adsense ads to be run on my site. The previous two times had a turnaround time of 8 days for a review that is automated. To me this is insane, but maybe it's a good thing, because I was looking online and saw that some random redditor said that borderline cases go to a human contractor to review. Maybe my turnaround time is so long because I always get sent to human review due to being borderline?

I tried Ezoic as well, but they rejected me due to not being in compliance with the Adsense policy. I have no idea what specifically I am being rejected for because Google doesn't say, though they tell me it's "Low Value Content". Ouch.

I've made some efforts to improve the quality of the homepage, so we'll see if the 3rd time is the charm. Fingers crossed.

### Going Viral

I banked hard on SEO to be a driver of traffic to my site and so far the results have not been exemplary. They may improve over time but I have arrived at the conclusion that I want things to go faster.

Going viral is the best way to deliver organic traffic quickly, but how does something go viral? I don't really know, but I have a few ideas and those will be the focus of next month.

Idea 1: Generate TikToks

I have so many multiple choice questions in my DB that have been generated, why not package them up into a vertical video format and try and get some distribution on TikTok, IG, and YT shorts?

I'm already working on a system that can churn these out fast and cheap, so we'll see if we can make something happen in October.

Idea 2: Addicting Trivia QOTD

Wordle was able to get a lot of distribution from people sharing their scores, so I'm wondering if I can repeat that playbook and create a "Quiz of the Day" (QOTD). Every day there is a new one that is the same for everyone. Create a streak system and a little graphic with emojis that shares well on twitter, FB, etc.

I don't know the specifics yet, but I think there's a chance something along these lines can work.

### TikTok Creator Program

If an account gets enough followers it becomes eligible for monetization via the creator program. I wonder if I can grow a set of TikTok accounts with programmatically generated content. If a video can get 10K views on average with $0.50 RPM, then with 2 videos per day that works out to $300 per month.

Very optimistically, if I can create videos that do 100K avg views, and post 4 times per day with $2 RPM, then this would be $24,000 per month.

Expanding out to YT and IG, there is a chance this alone is a 6 or 7 figure ARR business on its own.

## Goals for October

- Publish 100 TikToks
- Get 1000+ views on at least one TikTok
- Create QOTD and share it every day on Twitter thereafter
- Get approved for Adsense on [AddictingTrivia.com](https://addictingtrivia.com)
- Average 20 hours of work per week


SEO is slow so I also want to get traffic from going viral

Monthly Retrospective - September 2023


I launched [AddictingTrivia.com](https://addictingtrivia.com), and became a Twitch Affiliate!

## Stats

- Worked 84.5 hours
- Of that work, 60 hours was done while streaming live on twitch
- 71% of work was done live on twitch
- Revenue $0.75 🤠
- 174 Twitter followers
- 73 Twitch followers

## August Goals

### Find a niche and target customer for MinutesGenerator

Grade: F

I deprioritized MinutesGenerator and as a result didn't invest time into pivoting to find a niche for it. More on this below on the last goal.

### Invest in learning about marketing

Grade: B

I watched several Ahrefs youtube videos about how to do content marketing and use their tool to identify keywords with low competition. I would say I learned just enough to be dangerous.

I won't give myself a higher grade here because it does seem like I could have done more, but frankly I think this goal isn't well defined because it's hardly measurable. I think I achieved what was in the intended spirit here though.

### Decide on and execute a marketing approach, e.g. SEO

Grade: A

I paid for an Ahrefs subscription and used my learnings from the Ahrefs tutorials to identify an SEO niche for my next business, AddictingTrivia. This involved spending 12 hours across two days typing words into their tool until I found what I think is the right balance of low competition keywords, and high traffic.

This process was slow, but educational and I feel much more familiar with the tool. I'll classify it as one of these one-time-cost-asset things because I intend to use this tool for every subsequent business idea I have.

This goal gets an A because I picked SEO, and if things pan out as promised, then I can build a website that has a high likely hood of aquiring traffic from the get go. This dovetails nicely with the maxim that "entrepreneurs should build something that people want". People search for things they want, so basing a business idea off search query data checks this box.

### Make a decision by EOM about continuing to invest in MinutesGenerator

Grade: A

As I said above, I'm putting MinutesGenerator on the backburner. There's a chance I'll come back and try to pivot this into a niche, but for now I think my efforts are better spent elsewhere, rather than trying to squeeze blood from a stone in a saturated market.

There are many competitors in the audio to text transcription space, and many of them are VC funded, and pay for ads, and teams of engineers and designers. I'm not optimistic about a head on competition here, and I'd prefer not to go on a suicide mission trying to win a David and Goliath style battle.

One of the luxuries that I can enjoy as a bootstrapped founder is that I don't need to make a billion dollar company, and I don't need to make commitments to investors about what my business is going to do. I can pivot as much as I want (and I plan to) in order to solve the optimization problem of finding the easiest, most sustainable way to build a business that pays me enough money to live.

I'll leave MinutesGenerator up and running for as long as I can, but for now I am on to bigger and better things!

## What got done:

- Created AddictingTrivia.com
  - Right now hosts multiple choice quizzes on Minecraft and The Office.
  - 11 pages indexed by google as of this writing, most of them quiz pages.
  - First commit went in August 15th
- Became a Twitch Affiliate
  - This makes me eligible for monetization, and I have made 75 cents so far 😎
  - My engagement stats on Twitch are rising as well, I've noticed I also have more and more regulars coming into my chat
- Helped build a collaborative connect 4 game using BashStack (powered by HTMX) courtesy of my fellow streamer BadCop. Super fun, good times.
  - Try it out [here](https://connect4.bashsta.cc/)
- SEO education and research
  - Becoming familiar with Ahrefs SEO data tool

## Things I am Thinking About

### AddictingTrivia

The basic version of the plan is to acquire traffic via google search. There is a volume of about 400K search queries per month for keywords that are supposedly easy to rank for, that are also some form of "quiz" or "trivia", for example "minecraft quiz".

I'm going to focus on multiple choice quizzes because they are the simplest to generate, but I like the idea of expanding to other quiz types eventually. I am building a system that uses GPT to generate quizzes given a topic, and I intent to have this system be fully automated end to end, starting with a root keyword like "minecraft" and ending by uploading quiz data to my DB, pushing the website, and requesting indexing from google.

My goal is to generate 1000 quizzes using this system, because I think this will lend legitimacy to my website. I believe that nobody will take seriously a quiz website with only a dozen quizzes. This will also cast a wider net in SEO-space, allowing my site to acquire more traffic.

Eventually I will put Adsense ads on the site to monetize it.

The two biggest things to figure out for this approach are backlink acquisition and quality control. Backlinks are hard to get, and I don't know how systematized I can really make this process. I'm punting on it for now, because I think it will be an easier sell to ask for a link to a website with thousands of quizzes rather than a dozen.

Ahrefs comes to the rescue here too, because I can look at my competitor websites and see who is linking to them as well. This gives a reasonable place to start searching for backlinks, but ultimately it seems like a manual process.

The second thing to figure out is quality control for quizzes. GPT hallucinates wrong answers and off-topic questions decently often, which leads to poor quiz quality. My rough idea here is to use explicit and implicit feedback from website visitors to dynamically make adjustments to quizzes. For example, if a quiz has a question that 100% of users get wrong, that is an indicator of the question being bad. For explicit feedback, probably a thumbs up/down button asking to rate the quiz would suffice.

### Twitch Collabs

I had a lot of fun doing a costream with BadCop and it led to a few followers on twitch and twitter which was nice. I spend a lot of time streaming and it's rewarding to make more and deeper connections in that space. I definitely would like to do more costreams.

## Goals for September

- Publish 1000 quizzes on AddictingTrivia.com
- Get one page on AT to rank in the top 10 results on google for any search query
- Average 20 hours of work per week
- Investigate a solution for analytics that I can use for quality control


Monthly Retrospective - August 2023


## Stats

- Worked 85.75 hours
- Of that work, 52.75 hours was done while streaming live on twitch
- 62% of work was done live on twitch
- Revenue $0

## What got done:

- Created a landing page
  - Optimized for mobile
  - Attempted to make this SEO friendly by adding customizability to the <head> tag for things like title, description, open graph meta tags.
- Finished MVP for [MinutesGenerator.com](https://minutesgenerator.com) product
- Integrated stripe
  - Checkout page
  - Insufficient credits checking
  - Flow to create user post-checkout
    - Not sure about this, on the one hand could be lower friction for conversion, on the other hand it's weird and unclear if people will even convert via buying right away on the landing page.
- Created a zero downtime solution for updating the server
  - Patches Nginx config and reloads
  - Able to "drain traffic"
    - Logic to wait for async jobs to finish on the server before killing server process and restarting
- Read "The one thing"
  - Try hard to always be working on the right thing AKA the most important thing AKA the one thing such that by doing it, everything else becomes easier or unnecessary.
  - If you don't know what the one thing is, then your one thing is to find out what is your one thing.
  - It's hard to know what this thing is because sometimes you don't know what you don't know AKA "unknown unknowns"
    - When it comes to marketing, for me, there are many unknown unknowns.
- Wrote a statically generated blog, with AI generated content. Takes 10-15 mins to publish a post end to end.
  - I may not end up pursuing the AI generated angle depending on the marketing strategy I choose.
  - Regardless I think having a simple, customizable blog platform that I control can pay dividends in the future.

## Things I am Thinking About

Go to market strategy (what is my "one thing")

I had a "now what" moment after launching officially on twitter. No one is going to use this product unless there is a clear value being provided and potential customers are aware of this solution. "Who is my ideal customer?" is a question I need to answer. I think answering this question is the most important next thing to figure out.Strategy-wise I came at this project backwards in the sense that I did not first identify a niche or who my customers are. I created a product that I know has broad value, but there are other competitors in the space who are well resourced with large marketing budgets, designers, engineers, etc.

I will have a hard time competing with these companies head-on in the broad space of speech to text transcript and minutes generation. It's possible that I can gain traction here but I want to find the easiest path to success, and that's likely going to involve "going smaller". If I can find a niche community on Reddit, Facebook, or LinkedIn and target MinutesGenerator to that community, talk with users to get feedback, and specialize it to a use-case, then I think I will have a better chance of finding paying customers.

When I create my next business, I need to find the niche first. Ideally talk to customers first. For MinutesGenerator my "one thing" is to drive likely-to-convert traffic to my site. How to do this will depend on who my ideal customer is, which is something I don't know yet. So figuring this out is, I think, the most important next step, and will involve finding a niche. Doing this will make everything else easier - for example, I could invest into targeted SEO "minutes generation for ESL social workers" or something like that.

Shared work making subsequent businesses easier

One thing I'm really happy about is the investment I've put into what I think will be reusable assets. My landing page, stripe and clerk webhook handlers on my server, my blog, learning about NextJS, PlanetScale, and zero downtime server update system. The cost of setting up a new blog for a new website is dramatically lower now that I have the code ready to go. Writing the code for MinutesGenerator took me about 3 months, and I estimate ~160 hours. Without being too scientific about it, I would estimate that at least half of this time was spent doing expensive "one time cost" activities such as building/designing/optimizing a landing page.

There will be more things to learn and more work to do on these things, but I am optimistic that I'm building a cache of assets that will help accelerate my productivity. If it took 160 hours to launch MinutesGenerator, then perhaps business number two will take 80 hours. Doable within one month, which is a ~67% reduction in wall time. How fast can I go?

Tracking my hours and streaming has been great for my productivity

I didn't start tracking my hours until the last week of June, but forcing myself to fill out a timesheet and be intentional about my work has been transformative. I set a goal to work 20 hours/week and didn't even hit it (I barely missed it would have been (20 \* 31 / 7 = 89 hours) , however I am confident that I worked more hours this month than any previous month.To be clear, when I talk about working X hours, I really mean working X hours. Almost no time checking my phone, getting coffee, chatting with people, no meetings, no emails. Just sitting at my desk writing code, or reading docs, or asking chatGPT questions.

Working more hours isn't a perfect measure of productivity, but I can get a lot done in an hour. If I can consistently put in hours, and consistently ensure I am working on the most important thing, then I am confident that I will be successful eventually. Working hard is hard work.

Streaming has been a great motivator to work long stretches without interruptions. Not everything can be done on stream, and there's a balance to strike because streaming can lead to situations where you don't work on the most important thing, because you are biasing towards more interesting content vs. what needs to be done. But despite the nuance, it's a great tactic, and I'm gonna keep at it.

## Goals for August:

- Find a niche and target customer for MinutesGenerator
- Invest in learning about marketing
- Decide on and execute a marketing approach, e.g. SEO
- Make a decision by EOM about continuing to invest in MinutesGenerator


Max Sherman

Recent Blog Posts