Trust
A word on AI
Finance teams are right to be wary of AI making up their numbers. So we built LeakIQ the other way round. The figures you see, what is leaking, how much is at risk, how confident we are, are produced by deterministic rules reading your real records. Not by a model. This page explains exactly where AI is used, where it is not, and how we keep it that way.
The short version
AI does not produce your numbers. It never has, and the way LeakIQ is built, it never will.
Any AI in LeakIQ is an optional assistant that writes in plain English about what the rules already found. It can draft a summary or suggest a next step. It cannot change a figure, and it is off by default.
Always deterministic, never AI
Finding the leak
Every issue is detected by fixed rules reading your source records: a failed payment, an overdue invoice, a duplicate charge. There is no model deciding what counts as a leak.
The money at risk
The pound figure on every issue is read or calculated directly from your data, not estimated by a model. Where a value is genuinely an estimate, it is labelled as one.
The confidence score
Confidence is a deterministic score from the detection rules and the materiality and age of the item. It is not a model's guess, and every score traces back to a real record you can open.
What counts as recovered
Whether a recovery worked is decided by comparing expected against actual value from your systems, and recorded in an immutable audit trail. A model never decides an outcome.
Where AI is, and is not, used
The optional assistant may
- Write a plain-English summary of an issue the rules already detected.
- Suggest, in words, the steps a person might take to recover it.
- Explain an outcome after you have recorded it.
Always a draft for a person to read. It is never the source of a number, and it never acts.
AI is never involved in
- Deciding what is a leak or how much is at risk.
- Setting a confidence score or deciding an outcome.
- Approving anything, or changing anything in your systems.
Recoveries are executed by your team, approved under segregation of duties, on read-only connections. LeakIQ never writes back to your systems.
Our commitments
Off by default. The assistant stays off until it is explicitly and deliberately switched on. Adding an AI key alone does nothing; a separate, documented switch is required, so it can never turn on by accident.
No training on your data. Your data is never used to train AI models. Any AI provider we use is contractually bound not to train on it or retain it beyond the request.
Named sub-processor, under the DPA. If we enable the assistant, the AI provider is added to our published sub-processor list with advance notice, and covered by our Data Processing Agreement before any data is sent.
Minimal data, only when asked. The assistant only ever receives the specific issue text needed to draft a summary, never your full dataset, and only for issues you open.
A human decides everything. Every recovery is reviewed, approved and executed by a person. The assistant informs; it does not act.
Where we are today
The optional assistant is currently off. LeakIQ runs entirely on deterministic detection, so nothing you see is generated by AI. We will only turn the assistant on deliberately, once the governance above is in place, and our intent is to make it a per-workspace choice so each team stays in control. When that happens, this page and our sub-processor list will say so first.
Reviewing us as a vendor?
We are happy to walk your security or data team through exactly how detection works and how the optional assistant is governed. Email security@leakiq.io.