A revenue analyst for your Stripe account
Revenue answers you can check.
Ask about your business in natural language. Namuna writes the analysis, runs it against your Stripe data in an isolated sandbox, and shows the code behind every number. When an answer leads to a price change, it rehearses that change on Stripe before a customer sees it.
What is our current MRR in USD?
First attempt · 17 Oct 2026, 23:33 UTC
Doesn't match Stripe$285,985.11
Confident, cleanly formatted, and $50,795.60 short: it read USD base prices for customers who pay in 15 other currencies.
After the fix · 17 Oct 2026, 23:39 UTC
Matches Stripe to the cent$336,780.71
Graded against MRR recomputed independently from 420 customers' Stripe objects.
Real benchmark runs on a synthetic Stripe test-mode account. A number you can't check is a guess.
Revenue numbers are easy to get and hard to trust.
Multi-currency prices, coupons, trials and churn make a simple question like MRR surprisingly easy to get wrong.
Dashboards stop at the question they were built for.
Churn by market, the real cost of discounts, revenue by currency: every new question becomes a ticket.
AI answers sound right when they're wrong.
Our own agent's first MRR was off by $50,795, delivered with total confidence. Nothing on the screen said so.
Decisions get made on the guess.
Prices change, markets get cut, budgets move. Then the number turns out to be wrong.
A number you can't check is a guess.
Every answer shows its work.
The analyst writes Python, runs it in a sandbox that can only reach Stripe, and keeps the code next to the number. A benchmark grades it against values recomputed from Stripe, so you know how far to trust it.
What share of MRR comes from emerging markets?
25.3%
Emerging markets are $85,083 of $336,781 MRR.
Matches Stripe ground truth (expected 25.26%)
Code that produced this
- starter$9,214
- enterprise$61,816
- growth$77,817
- scale$187,934
5 of 5 benchmark answers match Stripe exactly.
Ask where your price is losing customers.
One price for the whole world quietly loses customers in part of it. Namuna reads your book market by market and shows where churn follows price, before you decide what to change.
Churn where customers outside the US pay in US dollars
31%
Kenya, Nigeria, Colombia, Indonesia
Churn everywhere else
13%
Local currencies, or the US at list price
Of those who left, citing price
8/12
$3,541/mo lost or at risk
Synthetic Stripe account; churn modelled to rise with price relative to local purchasing power. See every market →
When an answer means changing a price, rehearse it first.
Describe the change. The agent writes the migration, runs it against Stripe Test Clocks for 13 months, and checks every invoice Stripe produces against its own prediction. Stripe checks the agent, not the other way round.
- Natural-language plan
- Agent writes migration + prediction
- CreateOS sandboxegress: api.stripe.com only
- Stripe Test Clocks13 months
- Compare predicted vs Stripe invoices
- Verified migrationrequires human approval
↺ mismatch → fix and rerun
Monthly Pro · 20% off Example
| Month | Predicted | Stripe actual | Match |
|---|---|---|---|
| M5 | USD 39.20 | USD 49.00 | Differs |
| M6 | USD 47.20 | USD 59.00 | Differs |
| M7 | USD 47.20 | USD 59.00 | Differs |
- Runs in an isolated micro-VM
- Writes only to the rehearsal's own subscriptions
- Never touches live mode
One CLI set up the whole stack.
Stripe
Sandbox, Test Clocks, billing
CreateOS
Isolated sandboxes for agent code
OpenRouter
The models that write the code
stripe projects add stripe/sandbox stripe projects add createos/project stripe projects add openrouter/api