The open-source engine shows you, privately, whether your AI works. Cloud proves it to the people who act on it, from your CFO to the vendor, and turns what it finds into money back and a better deal at renewal, every month.
For the analyst who wants to know, privately, whether the AI works.
For IT, finance and procurement teams who have to act on the answer. Early access: I set it up with each design partner.
Design partnerships are free: you get the check and a readout; I ask for feedback, an anonymized case study and a reference call if it helped.
See pricingThe same method as the free engine, turned into results your team can act on.
Per resolution, per conversation, per session, in assists or credits, or per seat: send each bill or usage export to the program's upload address and it is checked against your own records as it arrives. Billing errors become a dispute pack the vendor can recheck; usage shows units spent in test instances and on conversations that failed. Every model ends in what each resolution that held cost, and I help you take the findings to the vendor.
Bill reconciliation · usage and seat checks · help with the vendor
I run the analysis, independently of the vendor, and the question is locked with a server timestamp before any result exists, so nobody can move the goalposts. The vendor can rerun every number with the open engine, but can't pay to change it. Once five organizations contribute, you also see your result beside your peers'.
Evidence pack · resolution audit · peer benchmarks
Read-only connectors pull fresh tickets, and the checks run by themselves every month. An alert reaches Slack, Google Chat or any webhook when claimed resolutions stop holding, a bill has items to dispute, or the data goes stale, so a vendor update nobody announced shows up at the next check, not at the renewal.
Monthly checks · regression watch · alerts
A portfolio view across every AI program and a cost ledger, with the unit values that turn hours into money signed off by finance, not by the vendor. Every registration, sign-off, result and download is kept in an append-only audit log.
Portfolio view · cost ledger · finance sign-off · audit log
I set up each program with you and present the findings to IT, finance and procurement. Read-only connectors (in beta) pull tickets, and any system can send the AI's conversations and bills to the program's upload address, so after setup nobody on your team has to run anything.
Setup and readouts with me · read-only connectors · upload address
The method and the numbers are the same. What changes is what you can do with them.
| Topic | Open-source engine | Probative Labs Cloud |
|---|---|---|
| Every analysis and report | Every one, run by you | The same, run every month for you; I run the action audit and bake-offs with you |
| Money back on your AI bill | Check any bill or usage export yourself | Every bill or usage export checked as it arrives, and help taking disputes to the vendor |
| Proof for the renewal | Your own numbers | An independent finding, plus peer benchmarks once five organizations contribute |
| Early warning when the AI gets worse | When someone reruns the analysis | An alert from the next monthly check |
| Answers for the CFO and board | A report per program | Portfolio view, finance sign-off and an audit trail |
| Getting data in | Exports you download | Read-only connectors (beta) and an upload address |
| Effort each month | Your analyst runs every check | Nothing to run after setup |
| Where your data lives | Your computer | Pseudonymized; in your own cloud, or hosted for you (planned) |
| Price | Free, Apache 2.0 | Fixed-fee project or annual plan |
Every analysis is in both: evidence pack, resolution audit, bill reconciliation and dispute pack, regression watch, action audit, vendor bake-off, measurement plan, benchmark contributions and re-performance.
You can, and your numbers will match mine. What you can't produce on your own is a finding the vendor has to take seriously: one from a party with no stake in the result, with the question locked before any result existed. Nor can you compare yourselves with peers, because that data only exists when many organizations contribute. And someone has to pull the data, run the checks and take disputes to the vendor every month. With Cloud, nobody on your team has to.
They don't have to take my word for it. Anyone with the data, the vendor included, can rerun every number with the open engine, and Cloud records the engine version, data fingerprints and settings behind each result. Vendors can't pay for, preview or suppress results: see our independence rules.
No. Every analysis is in the free engine. Cloud never runs a private version of the method, new checks land in the open engine first, and what is released as open source stays open source.
Yes. Cloud runs the same engine and method, so what you find on your own is what I would find. Try it on one export, then book a pilot when the result has to stand up with your CFO or the vendor.
Design partnerships are free: you get the check and a readout; I ask for feedback, an anonymized case study and a reference call if it helped.
Probative Labs