We read a replica of your database and your affiliate-platform export → build cohorts and payback → at 09:00 a post with verdicts and buttons to dig in lands in your Slack.
Rates, cohorts, payback — any question in plain words. The model parses it, Python computes on your cohorts, and the answer returns to the same thread in whatever form fits: a plain-text reply, a BI chart, a table or a CSV.
| NGR · 7d | −38% vs baseline |
| ROAS · 28d | 0.54 |
| NGR / player · 90d | €142 |
| payback @ 40% now | 3.1 mo |
| payback @ 45% | 4.2 mo |
How it works
Four steps — switch between them to see the mechanics.
The collector runs on your server and reads a replica of your database and your affiliate-platform export — read-only, no SDKs, no event rewrites.
Only cohort aggregates leave your perimeter, per a declared manifest, and every crossing is logged on your machine. The collector's source — all the code that sees your data — is open to you.
| user_id | geo | deposit | ts |
|---|---|---|---|
| 84120 | DE | 120.00 | 09:41:22 |
| 84121 | TR | 45.00 | 09:41:25 |
| 84122 | DE | 80.00 | 09:41:31 |
Baselines and rules live in a knowledge base inside the engine — built from your own cohorts: geo, device, source, seasonality, payback windows.
Every threshold is a number that can be traced back to the data it was computed from — nothing is pulled from thin air.
Every flag names the baseline it broke — you can trace any threshold back to the cohorts it was computed from.
Your question goes to the model — the arithmetic never does. Python computes every figure; the model only narrates the result.
It cannot change a digit or a verdict: if its text disagrees with the computed numbers, the answer is dropped, not published.
| geo | avg deposit | vs prev week |
|---|---|---|
| DE | €96 | +4% |
| TR | €41 | −3% |
| BR | €38 | flat |
Daily, weekly or monthly — on your schedule. Any data request in between: aggregations, complex queries, advice on rates and next actions.
The morning post ships even if our side is unavailable — it then carries only your own measurements, with a plain note that there are no recommendations today.
Advice included: rates, actions, what to pause and what to scale — answered from your own data, in your messenger.
Capabilities
Live captures of the product on synthetic data — the mechanics are real. Click a small tile to bring it forward; the large one opens full-screen.
The 09:00 post arrives: scroll the verdicts, press View — the reasoning unfolds in the thread.
A question in plain words — the answer comes back as a table, CSV attached.
Asking what to do — the numbers run first, the advice shows its working.
Cohorts charted right in the thread — payback curves without leaving Slack.
One metric, split by geo, source and device — asked in a single message.
The same numbers on the web — drilling from summary down to source.
Price
The rate follows the complexity of your project. Three parts — each billed on its own.
Contact
30 minutes, your metrics, no slides. We connect a sample source live and you watch the first verdict arrive.