By Max Sherman ยท February 2, 2024
I (almost) 10x'ed the business
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.
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.
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.
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.
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.
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.
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.
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:
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:
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).
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.