By Max Sherman ยท April 3, 2024
FB ads work
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).
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.
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.
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.
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.
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.
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.
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.
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.