Sold my 7-figure agency.
Bootstrapping Lancelot to $1M, in public.
Helping apps that sell AI stop losing money on their best users.
Dad of 4. Follow along ๐ค
Sold my 7-figure agency. Now building Lancelot.
Companies that sell AI have a margin problem they can't see. Stripe shows what a customer pays. The model bill shows what they cost. Nothing joins the two.
Lancelot does. It matches AI cost to revenue per customer, finds who's underwater, and fixes it.
I'm building it in public, bootstrapping to $1M. Here you'll get:
- Real numbers, including the bad ones
- What I learn about AI cost and margins
- What I ship, as I ship it
If you sell AI inside a product, follow along. ๐ค
@BlockByBlockHQ I've ran studies with a few partnering SaaS (2) and and mobile apps that integrate with AI.
It was mind blowing how many power users were attributing to 50%+ of the monthly spend.
In tests, we were reducing spend by around 72.3%
Your AI feature probably makes money on average and loses money on your best users.
You won't see it in Stripe. You'll see it on the model invoice, a month late.
I wrote up the 5 ways this happens and how to fix each one. https://t.co/Pch3hvGl8R
@pradyum1916 A tool that stops at 2pm is a tool you plan your day around.
Your own users probably feel the same about your app's limits. Do you cap them anywhere?
@apoorvjain25 Caching fixes the bill, the next question is which users are still running up token costs.
Do you know your cost per user, or just the total AI cost?
@not_sacke@grok Better question for anyone building on Claude: what does your heaviest user cost you?
Stripe shows what they pay. The API bill shows the total. Nothing puts the two side by side.
That's what I'm building Lancelot for. Do you know yours?
@iancccho There's going to be copycat "SaaS founders". What matters is if you can build a better product, have a UX-focused website, great branding, and market better than they can.
Win on branding, features, and a marketing strategy and you'll be just fine!
@Zaydenf9 1. Ensure you win on quality of product / features.
2. Ensure your website and branding are uniform and steps above the competitors.
3. Out-market the "competitors".
4. Scale past the "competitors"
@soydiegoup Pre-launch, but our tiers are based on the AI spend we track for the customer, not seats.
It's an AI cost tool, so price scales with what it watches.
Which would you pick?
@_waela It depends - do those 10 customers bring another 20 customers? I've had partnerships where the other person was an expert at "xyz" - they brought their weight in value and success was equal part them.
Sometimes it works out, sometimes it does not.
@anah_sahh The vision, the thought, the corrections of Claude's thinking, the marketing, the website, the GTM plan - thats all you.
Did @elonmusk build the rocket bolt by bolt?
@CalumBrannan I ran a web design and dev agency and sold it in May. Path 2 is underrated: you already know where the hours go.
The catch is the team has to change how they work, not just add tools.
What was hardest to change after you bought?
@soydiegoup Pre-launch, but our tiers are based on the AI spend we track for the customer, not seats.
It's an AI cost tool, so price scales with what it watches.
Which would you pick?
@robbylit Simon's point is the real one. Once you can attribute token cost per project, you can price it.
Flat per-agent pricing breaks that.
Do you think agents end up usage-based, or flat with caps?
@Michaelchoong5@grok Usually nothing, unless you set it.
Common fixes: a usage cap per plan, a cheaper model past a threshold, or an upgrade offer when someone crosses it.
The hard part is spotting who's crossing it before the invoice. Which would you try first?