Catch the paperwork problem before the custodian does.
Three checks before anything leaves the firm. Every result cites its page and is sealed in a signed receipt.
Three questions, before submission.
Each result gives its reasons, their pages — and exactly what to fix.
In good order?
- Required documents and fields
- Signatures and their dates
- Expired ID
- Beneficiaries that don't add up to 100%
Will it go through?
- Title, TIN or registration mismatch
- A stale statement
- Holdings that may not move in kind
- Margin debit, fractional shares
Corroborated?
- Independent documents for the declared wealth
- In proportion to the risk rating
- PEP, adverse media, screening flags
An amber or red result is a request about paperwork — never a decision about the client. Rules are set per custodian.
One application. Every document. One case.
The engines decide on what Gothink AIR publishes — never on raw text.
Read
Nine document types into governed records
Join
One case; conflicts kept with both citations
Decide
Sealed as a replayable, signed receipt
Fix once
Back to the client once, not three times
Tools for AI agents
A KYC platform hands over records; gets back a receipted decision.
One printable sheet
Result, what to ask for, amounts with pages. Personal values can be withheld.
Configuration, not code
Per custodian and per bank. Your compliance function sets every threshold.
Corroborated for the risk — or sent to a person.
Evidence weighed against the client's risk rating. Three possible outcomes.
Every result, whichever way it goes, is sealed with its page citations.
Proportionate to risk
Higher rating, more evidence.
Flags are never missed
A screening flag always reaches compliance.
Cited and sealed
Every result, with the pages it relied on.
Singapore and Hong Kong
A draft settings file for your compliance function to sign off — not in force until approved.
Not advice. Other booking centres are not built yet.
Change the case, watch it re-decide.
Five invented private-banking cases, each change sealed live.
The one number to ask for — on your files.
How often the engine agrees with your reviewers, and whether it ever clears a file they would not.
Agree criteria first
In writing. Default: no false clear
25–50 real packs
Anonymised by the bank
Reviewers decide blind
Without seeing the engine's result
Discuss every miss
Then a fresh set of packs
What a small pilot can prove
Zero false clears observed still allows a true rate of up to about:
The report prints that interval next to the count.
What is built, and what is not.
| Piece | Status |
|---|---|
| NIGO and ACAT engines, onboarding pack (nine document types) | Built |
| Source-of-wealth engine and pack (seventeen document types) | Built |
| Live demo in the Decisions console | Cloud · fictional cases |
| Pilot kit — decision agreement and field-level accuracy | Built |
| Accuracy figure | Not claimed |
| Console screen for onboarding operations | Not built yet |
| Custodian-specific rule sets | Not built yet |
No accuracy figure will be published until a pilot measures one.
Run the pilot on your own files.
Wealth managers and private banks: measure the engines against your own reviewers.