Catch the paperwork problem before the custodian does.
Three decision engines check a new client’s file before anything leaves the firm: is the paperwork in good order, is the account transfer likely to go through, and is the source of wealth corroborated for the client’s risk rating. Every result cites the page behind it and is sealed in a signed receipt.
Three questions, asked before submission.
Each engine returns an outcome, the reasons for it and the page each reason rests on. A result that needs work says exactly what to fix or what to ask the client for.
| Engine | Outcomes | What it looks at |
|---|---|---|
| Not in good order paperwork | IN_GOOD_ORDER · NIGO | Documents required for the account type, required fields that are missing or held, signatures, signature dates in the future or too old, expired ID, beneficiary percentages that do not add up to 100%, and owner details that differ between documents |
| Account transfer ACAT | LIKELY_CLEAN · AT_RISK | Account title, TIN or registration type that does not match the delivering account, a stale statement, holdings that may not transfer in kind, a margin debit, fractional shares on a full transfer, a missing account number |
| Source of wealth private banking | CLEAR · REVIEW · ESCALATE | Whether independent documents corroborate the wealth the client declares, in proportion to the bank’s customer risk rating — with PEP, adverse-media and screening flags escalated wherever they appear |
One application, every document, one case.
Gothink AIR reads the documents with the onboarding pack; the engines decide on what it publishes, never on raw text.
Read
Application, government ID, W-9, certification of trust, beneficiary designation, entity resolution, transfer form, held-away statement and advisory agreement are read into governed records.
Join
Records are joined on the application into one case. Every value keeps its page; owner details that differ between documents are kept as conflicts, with both citations.
Decide
Each engine decides the case and the decision is sealed as a replayable, signed receipt — the same receipts Gothink Witness puts in front of an examiner.
Fix once
A NIGO or REVIEW lists exactly what is missing, so operations or the relationship manager goes back to the client once, not three times.
On the MCP tool plane
The engines are exposed as tools alongside the rest of the decision layer. A KYC platform can hand over its verified records and get back a receipted decision, its reasons, the documents to request and the page citations.
One printable review sheet
Each source-of-wealth decision has a one-page sheet: the result, what to ask the client for, evidence amounts with their pages, and the receipt. Personal values can be withheld when the sheet leaves the bank.
Configuration, not code
Required documents, fields and age limits per custodian; risk tiers, currency rates and booking-centre settings per bank. Every threshold is a starting point for your compliance function to set.
Corroborated in proportion to the risk, or sent to a person.
The source-of-wealth engine was rebuilt on review of its first version. What it will and will not count as evidence:
Only independent documents corroborate
The client’s own assessment, their passport and the screening report never count as evidence of wealth. A sale agreement, grant of probate, tax filing, company accounts or trust deed can.
Proportionate to the rating
How many documents, whether they must state an amount, and how much of the claim they must cover all follow the bank’s customer risk rating. An unknown rating is treated as high risk.
Currency aware, no rate assumed
Each document’s amount is tested against the claim. Where currencies differ, the rate comes from the bank — the engine never assumes one.
A flag anywhere escalates
A PEP, adverse-media or flagged-screening value in any document escalates the case, even when another document says otherwise.
Trusts, companies and gifts
A trust needs a named settlor, a business source needs the company, a gift needs the letter and the transfer. The category “other” always goes to a reviewer.
Statements go stale; probate does not
Statement age limits are set per document type, historical evidence stays valid, and future-dated documents are caught.
| Booking-centre settings | What we ship |
|---|---|
| Singapore and Hong Kong | A draft settings file for the bank’s compliance function to sign off, with its basis stated (MAS Circular AMLD 05/2026, MAS Notice 626, the HKMA AML/CFT Guideline) and the open decisions listed. It is not advice and is not in force until your compliance function approves every value. |
| Other centres | Built with a design-partner bank from its own policy. |
Change the case, watch the decision change.
Five invented private-banking cases — a company sale, a self-declared claim, an inheritance part-covered in two currencies, a trust with no named settlor, and a politically exposed person. Change the risk rating, the amount, the PEP flag, the documents on file or the booking-centre settings, and each change is decided live, sealed in a signed receipt, with its review sheet and a one-click check of the receipt chain. The cases are fictional: the demo shows how the engine behaves, not how accurate it is.
The one number a bank should ask for, measured on its own files.
How often the engine agrees with your reviewers on real files, and whether it ever clears a file they would not. A gold set someone writes by hand tests software; it measures nothing, and the scorer refuses to call it a measurement.
Agree the criteria first
GO criteria are agreed in writing before any pack is scored. The proposed default is strict: no false CLEAR and no missed escalation.
25 to 50 real packs
Anonymised by the bank before they leave it, mixing the risk ratings and categories the booking centre actually sees — trusts and company sales included.
Reviewers decide blind
Your KYC reviewers decide each pack without seeing the engine’s result. Nothing leaves your machine.
Every disagreement discussed
False CLEAR, missed escalations, straight-through, over-review, review minutes and reason codes — then a fresh set of packs, never the same one again.
What is built, and what is not.
| Piece | Status |
|---|---|
| NIGO and ACAT engines, onboarding pack | Built — pack wealth_onboarding v1.0.0, nine document types |
| Source-of-wealth engine and pack | Built — engine and pack source_of_wealth v1.1.0, seventeen document types |
| Live demo in the Decisions console | On the cloud service — fictional cases only |
| Pilot kit | Built — decision agreement and field-level accuracy |
| Accuracy figure | Not claimed — none will be published until a pilot measures one |
| Console screen for onboarding operations | Not built yet |
| Custodian-specific rule sets | Not built yet — they come from each custodian’s requirements and reject codes |
Run the pilot on your own files.
We are looking for wealth managers and private banks to measure the engines against their own reviewers, and to shape the custodian rules and the operations screen with us.