Watch a run

Demo

A sped-up screen recording of Critias answering a start-up founder's questions about MRR, cash and ad spend from Stripe, ad-platform and card exports, with checked figures and charts.

Demo. Captions are on by default; the full transcript is below.

The recording is sped up while Critias works and while questions are typed. Every question, step and figure is shown as it happened, and each answer is held at normal speed. The data is synthetic: Kiezly, a fictional Berlin SaaS company, with eight exports from Stripe, Meta Ads, Google Ads and an Amex card. The model is Claude Sonnet 5.5 with thinking set to high, on Critias 0.5.1.

What you are watching

  1. A founder asks for monthly MRR and paying customers, and gets a table and a chart.
  2. They ask why the cash Stripe paid out differs from MRR, and get a line-by-line reconciliation.
  3. They ask whether Meta or Google wins cheaper customers, and Critias says what the data can't show.
  4. They ask which platform actually works, judged on the oldest customers, and payback decides.

Transcript

Caption: Stripe, Meta, Google and card exports, straight from the dashboards.

Question: What was our MRR at the end of each month this year, and how many paying customers did we have?

Caption: MRR from invoices, annual plans spread over their months.

Answer: MRR grew from €33,268 at the end of January 2026 to €55,737 at the end of September, and paying customers from 466 to 800. Annual plans count at a twelfth a month. A month-by-month table and a line chart of MRR follow, with the method and caveats.

Question: Why is the money Stripe paid us in 2025 so different from our MRR? Explain it to me like a layman.

Caption: Cash and MRR reconciled line by line.

Answer: In 2025 Stripe paid €312,143 to the bank, while the twelve month-end MRR figures add up to €262,704. In plain English, MRR counts recurring business each month, and payouts count the cash that reached the bank. The gap comes from VAT collected for the tax office (+€45,141), annual plans paid up front (+€21,734), Stripe's fees (−€10,825), refunds and disputes (€989 and €1,034), and money still held in the Stripe balance (−€4,581). A waterfall chart shows the bridge from MRR to payouts.

Question: Is Meta or Google getting us cheaper customers? Use the last 12 months.

Caption: It says what the data can't show: most customers carry no channel tag.

Answer: Meta costs €1,491 per new customer (124 customers) against €2,404 for Google (112). But 544 of the 780 new customers in the window have no channel tag, so both figures cover tagged customers only, and the ranking holds only if both platforms tag customers at a similar rate. A bar chart compares the two.

Question: Which ad platform is actually working for us? Judge it on the customers we won in our first year (April 2022 to March 2023), since we have watched them the longest.

Caption: Cheaper isn't better. Payback decides.

Answer: Google is the platform that is working. Its first-year customers cost more to win (€1,353 against Meta's €634) but paid back their cost in month 34. Meta's customers stop paying sooner and have not paid back within the data. The cohorts are small (20 Meta and 14 Google customers), so the figures show a direction, not precise values. A chart compares the cost of each customer with what they have paid to date.

End card: Critias, your own personal pricing agent. npm install -g critias. critias.ai

Last updated