Skip to content

AI that explains the evidence and never makes the call

Revo Intelligence is the assistant layer across Map, Verify and Control. It explains evidence and drafts suggestions for people to review. It has no permission to change anything.

Connect for freeBook a demo

Failed test QUOTE-004

For your review

Suggestion, for your review

The recheck appears to add a city tax that the quote left out, and doesn't flag the change.

Evidence cited

  1. run_014 · QUOTE-004Quoted 184.00 EUR, recheck 197.50 EUR.
  2. tr_9f27c1, recheck responseA city_tax line of 13.50 EUR, with price_changed false.
  3. tr_9f27c1, quote responseNo city_tax line in the original quote.

Couldn't check: the seller's contract terms on when tax is included.

Waiting for a person

Nothing has been rerun or changed. Following it up is the engineer's decision.

Illustrative. Companies and data are fictional.

Suggestions cite the mapping evidence and test results they read.

0
mappings, bookings or routes changed by AI
Every
suggestion shows the evidence it read
Every
decision stays with a named person

The limits, written down

Each job has a boundary, listed below.

What AI helps with, and what it must never do, for each job
Mapping triageAI helps withRank candidate matches with reasons and unknowns, explain the evidence, and point out where it contradicts itself.AI must neverApprove a mapping, assign a hotel's final identity, invent hotel attributes or redistribute unlicensed content.
API onboardingAI helps withExplain required parameters and known adapter quirks.AI must neverRead or export credentials, or generate production requests.
Test designAI helps withSuggest edge cases from schemas and scrubbed failures, for a person to approve.AI must neverRun traffic outside the agreed test policy.
Incident diagnosisAI helps withRank plausible causes, with a citation for each.AI must neverRetry an unknown booking, cancel a stay or change a route.

A suggestion looks like a suggestion

It's labelled, it cites its evidence, and it says what it couldn't check. Following it up is the engineer's decision.

Suggestion, for your review

On failed test QUOTE-004, Alpine Distribution → Corvane Travel

The recheck appears to add a city tax that the quote left out, and doesn't flag the change1,2,3. Check whether the adapter reads the tax line from the quote response.

Evidence cited

  1. run_014 · QUOTE-004

    Quoted 184.00 EUR, recheck 197.50 EUR. Difference 13.50 EUR.

  2. trace tr_9f27c1, recheck response

    A city_tax line of 13.50 EUR, with price_changed false.

  3. trace tr_9f27c1, quote response

    No city_tax line in the original quote.

Couldn't check: the seller's contract terms on when tax is included. Nothing has been rerun or changed.

Illustrative. Companies and data are fictional.

It stays out of the booking path

A search or a booking never waits on a model. Assistance happens at the desk where people review the work.

The booking path

  1. search
  2. quote
  3. book
  4. read
  5. cancel

Deterministic rules and seller responses only, with no model involved.

The review desk

  • mapping cases
  • failed tests
  • incidents

Models can draft and explain here, and a person acts on it or doesn't.

What it does

Cites what it read

A suggestion points to the assertion, the trace or the record it came from, so you can check it in a click.

Carries no confidence percentage

We don't present an AI's own certainty as a score. You get the evidence and a plain statement of what it couldn't see.

Passes through the same checks as a person

Its output is treated as untrusted. Schema validation, business rules, permissions and approvals all still apply.

What it covers, and what it leaves out

The list on the right is deliberate, and we'll say when any of it changes.

Included

  • Ranked candidate hotel matches, with reasons and unknowns
  • Explanations of matching evidence and contradictions
  • Suggested causes for failed tests, with citations
  • Suggested edge cases for a person to approve
  • Every suggestion labelled as one

Deliberately left out

  • Confidence percentages on suggestions
  • Any part in the search or booking path
  • Alias and description suggestions, subject to data rights
Does AI approve mappings?

No. It can explain the evidence and a person decides. Approval records the person who made it.

Could a bad suggestion cause a wrong booking?

It has no route to one, because it can't approve, book, cancel or change traffic. A person could still accept poor advice, which is why each suggestion cites its sources and we sample them for review.

Is our data used to train a model?

We redact sensitive content and check model terms before passing seller data to any model. Ask us and we'll tell you exactly how that applies to your data.

How accurate is it?

We don't publish a figure. We track how often suggestions are accepted, corrected or found to be unsafe.

Questions we'd ask too

Straight answers, including where the answer is no.

Where would you trust an assistant, and where wouldn't you?

Tell us where your limits are.

Connect for free