AI does the case. Your team decides.
One export goes in. What comes back is a completed case, every fact carrying its document, version and page, with the decision waiting for a named person in your house. Between those two points sit four steps — and none of them can be skipped, including by us.
A case file that has to prove what it says.
Ask who assembles a case file in your house today, and what that hour costs. Here it assembles itself — and unlike the version a person builds under time pressure, every line in it can be traced back.
Every fact carries its source
Document, version and page — not a footnote at the end but attached to the individual statement, so the first person who doubts it can check that one line rather than the whole file.
Anything unsourced is flagged, not kept
An answer that cannot be traced to a document is marked as such. This is the answer to "surely it just makes things up": it can, and when it does, the file says so instead of hiding it in fluent prose.
Every finding gets closed
Fixed, escalated, or accepted with a reasoned assumption. Nothing is left in a list for someone to notice later, and the report says what was checked and what was not.
An agent team, not a chatbot.
A chatbot gives you one confident answer. Cause, severity, contractual entitlement, precedent and compliance are different questions with different right answers, so they are worked by different specialists, in parallel, on the same case.
Disagreements are surfaced, never averaged
Where the specialists contradict each other you see the contradiction. It is never smoothed into a confident-sounding summary — because in a case that is difficult enough to be sitting in a backlog, the contradiction is usually the part that deserves a human.
Severity comes from your matrix
Not from a generic library. Your rules, your cases, your exceptions — which is also the honest boundary: your configuration is built with you, and nothing here pretends to know your rules before you have shown them to us.
Two gates. Nothing material runs without both.
Gate one asks whether the file is complete. Gate two asks whether the right person is deciding — routed by your own signing authority, not by ours.
Finding
A case needs a decision - context assembles from your systems.
Agent review
Specialist analyses challenge the recommendation - policy, compliance, evidence.
GATE 1Consensus
A structured verdict - rationale, disagreements, unresolved evidence.
Human approval
The named quality, clearing or compliance owner decides. Nothing runs without this.
✓ GATE 2Execution
Only approved actions execute in your systems - logged and signed.
logged · signedThree days, then your own clearance, then a month, then a standing arrangement.
The work sample
You export 20–50 real cases against a written checklist. We work them and send them back finished, with a recording of the path. Nothing is installed and nothing is connected.
Your internal clearance
Legal, compliance and information security have to sign off before real volume moves. In an organisation your size that is four to eight weeks, and it is the longest part of this. We bring the data-processing agreement, the impact assessment and the hosting sheet to the first conversation so the clock starts then rather than after it.
The first shift
Real volume at an agreed turnaround, against a quality threshold, with your named approver. Cancellable at the end of it — which is the only honest way to sell a first month.
Standing cover
The queue is staffed month after month. An adjacent queue is added only once the first one runs reliably — never because it was on a plan.