How Rekava's AI matching stays trustworthy

When people hear 'AI reconciliation', they often picture a system that quietly decides where every payment goes and hopes it got it right. That is precisely the design we rejected. Here is how matching in Rekava actually works — and why it is built to be trusted, not just clever.

Rules first

Most payments are not ambiguous. A payment with an exact amount and a clear invoice reference does not need a model to interpret it — it needs a rule. Rekava runs deterministic rules first: exact reference match, exact amount against an open invoice, a known customer phone number. These matches are fast, explainable, and correct by construction.

Doing the easy 80% with rules means the AI is only ever asked the genuinely hard questions — and a system asked only hard questions is far easier to keep honest.

AI second, with its reasoning shown

For the remainder — a slightly different name, a part-payment, a transfer with no reference — Rekava scores candidate matches. Crucially, it never just returns an answer. It returns an answer with its evidence: the name similarity, the amount fit, the timing, and an overall confidence score.

A match you can't question is a match you can't trust. Every suggestion comes with the reasons behind it.

A human always in control

Above a confidence threshold the business sets, a match can clear automatically. Below it, the payment goes to a review queue where a person sees the suggestion, the evidence, and can accept, reject, or reassign it in one click. The threshold is yours to move — cautious businesses keep it high and review more; confident ones lower it over time.

This is the core principle: AI proposes, a human disposes. The model is a very fast assistant, never the final authority on your money.

History that can't be quietly rewritten

Every match, every reminder, every status change is written to an audit log. Nothing is silently edited. If a payment was matched, you can see when, by whom, and on what evidence. Trust in an automated system comes less from it being right every time and more from it being inspectable every time.

Used well, AI in reconciliation should feel less like magic and more like a careful colleague — one who does the boring work instantly, shows their working on the hard calls, and never pretends to be certain when they are not.