An 8-week pilot on your own flow
We run next to your current review; every case is still reviewed during the pilot. At the end you have your own number: how much can go out without review under your error cap, or an honest “not yet” with what it would take.
Three phases, one flow
Weeks 0–2: set up
We connect in shadow mode and agree, in plain language, what counts as an error and which case types always stay with people, whatever the score. If you already have reviewed history, we start from it. By the end of week 2 you get a first read on whether your data and risk score can carry your cap, or what is missing.
Weeks 2–8: shadow mode
Your reviewers keep reviewing everything. We record the AI's original answer and their verdict, have a small sample reviewed twice, and run the watch alongside.
Week 8: readout
What can go out without review at your cap, or an honest “not yet”. What should stay reviewed, what limits more automation and what it would take. The time saved after audits and upkeep. What the watch saw, the receipts, and a plan for going live.
A good fit
- An AI flow where people review every answer today, such as customer emails, claim notes or case summaries, with an AI that is already right in most cases: the lower its error rate and the better its risk score, the more can go out under a tight cap.
- Steady volume, so the pilot sees enough reviewed cases: often a few thousand per case type (the calculator estimates it for your numbers).
- Reviewers who can mark each case: sent as is, edited, or rewritten.
What we need from you
- A risk score per case, and the AI's original answer saved before any edit (it stays on your side). If your model API returns token probabilities, the gateway computes the score; otherwise you send a score with each answer, one you already compute or one from an open judge model run next to your system.
- The reviewer's verdict per case, and a small sample reviewed twice.
- Your definition of an error, and the case types that must always go to people.
- One contact person.
Your data. The gateway runs on infrastructure you choose: your cloud or on-premises. The readout needs only scores and verdicts, never the content of your cases, and they stay in the EU. Receipts carry identifiers and keyed hashes, not your content; the key stays with you.
Pilot price
| What | Price (excl. VAT) |
|---|---|
| Pilot: 8 weeks, one flow | EUR 12–15k |
The price is fixed in the offer before we start. What comes after the pilot is priced after your readout, from your volume and the value it shows.
Ready to find your number?
Fifteen minutes to look at your flow and agree the scope.