The suspicion phase, where most founders live
It starts as a feeling: the provider bill is bigger than the product’s usage seems to justify. The aggregate view can’t confirm or kill the feeling, because a bill total contains no users, per the structural gap in provider reporting. On client projects this phase lasted months: gross margin looked fine in aggregate while a handful of users quietly consumed the profit of everyone else. The diagnostic checklist for this phase is the three signs of a margin problem.
The sort, recreated in the demo
Open the demo, go to Users, sort by margin. The table is cost against plan price per user, and the shape appears immediately:
user plan cost (mtd) revenue margin
u_412 pro $29 $41.80 $29.00 -$12.80 <- the row
u_103 pro $29 $18.20 $29.00 +$10.80
u_309 starter $9 $6.40 $9.00 +$2.60
...most users: cents of cost against full plan price...That first row is the moment. Not an anomaly report, not a quarterly analysis, a user, on a plan, costing more than they pay, this month, visible in one sort. Click the row and the drill down shows *why*: which features drive the spend, which models, how many calls. In the demo dataset, as on the client projects that inspired it, the loss concentrates in one or two features, usually the conversational or agentic ones where context compounds, per the chatbot compounding math.
What the moment changes
- The debate ends.“Are some users unprofitable?” becomes “what do we do about u_412?”, a much better meeting.
- The options become concrete: cap the account (block or downgrade past a budget, the decision guide), reprice the tier (the heavy tail math), or knowingly keep subsidizing a strategic account. All three are fine; only invisibility is not.
- The averages lose their power to mislead. Once you have seen one red row hiding under a healthy mean, you stop trusting means, which is the correct posture for heavy tailed usage.
Having the moment on your own data
The demo gives you the moment on fictional data in thirty seconds. Having it on your real data takes the ten minute integration: wrap your client, pass user id and plan, and the same Users table starts filling with your actual accounts. Most products find their first red row within the first weeks of real traffic, and finding it in week two instead of month eight is the entire value proposition.
FAQ
How do I find my first unprofitable user?
Sort your users by AI cost against what each one pays. That single sort is the whole discovery: the moment cost per user sits next to plan price, any account past its price is visible instantly. The hard part is not analysis, it is that most stacks never join those two numbers, cost lives in the provider bill and revenue lives in Stripe, and nothing puts them side by side per user.
What does an unprofitable user actually look like in the data?
A red margin row: a user whose month to date model cost exceeds their plan price. In the Weckr demo dataset you can see the shape immediately, the top of the Users table sorted by margin shows accounts where a $29 plan carries $40 or more of monthly model spend, driven by one or two features. The pattern is always concentration: a couple of accounts, a couple of features, most of the loss.
Is this pattern real or an invented example?
The pattern is real and repeated: the founder watched it on multiple consulting projects shipping LLM features for SaaS clients, where a handful of users burned the margin of the rest and nobody could see it until the invoice. The specific numbers in the walkthrough come from the public demo dataset, which is seeded fake data built to reproduce exactly that repeatedly observed shape, no real customer data is shown.
What should I do once I find one?
Decide deliberately instead of discovering accidentally: cap them (block or downgrade past a budget), reprice (usage tier or overage), or knowingly keep them (a design partner or a strategic logo can be worth subsidizing). All three are legitimate; the failure mode is not choosing, because the account stayed invisible.
Can I experience this discovery without integrating anything?
Yes, that is what the demo is for: useweckr.com/demo is the full dashboard on the seeded dataset, no signup. Open the Users page, sort by margin, and you will have the exact moment of realization this article describes, on data shaped like the real thing, in about thirty seconds.
Keep reading
Thirty seconds to the moment
Open the demo, Users page, sort by margin. If the red row gives you the same jolt it gave the founder on client projects, the ten minute integration puts the same table on your real users, free for 50,000 requests a month.