Every upload becomes structured, reviewable work.
A bank statement is not a PDF to skim — it is a set of facts to extract, checks to run and questions to answer. This page shows exactly what happens between a client pressing upload and a fee earner recording a decision.
From upload to human decision.
Every document moves through the same pipeline. The final step is always a person — analysis prepares the review, it never concludes it.
- Uploaded
- Queued
- Extracting
- Analysing
- Cross-checking
- Review complete
- Human decision
Each document type gets its own review.
Extraction and checks are specific to what was uploaded. Select a document type to see what the system looks for.
Bank statement
Extracted: account holder, sort code and account number, statement period, opening and closing balances, individual transactions with dates and counterparties.
Checked: name matches the matter, period covers the required window, salary credits match declared income, large or unusual credits identified and grouped, running balance consistency.
Passport
Extracted: full name, date of birth, nationality, document number, expiry date, MRZ data.
Checked: name and date of birth against the matter record, expiry against the expected completion date, internal consistency of the machine-readable zone.
Driving licence
Extracted: name, date of birth, address, licence number, issue and expiry dates.
Checked: identity fields against the matter, address against declared address, expiry validity — and whether a second identity document is still required.
Proof of address
Extracted: addressee name, address, issuer, document date.
Checked: address matches the matter record, document recency against firm policy, issuer type is acceptable (utility, bank, council).
Payslip
Extracted: employer, pay period, gross and net pay, year-to-date figures.
Checked: net pay against salary credits observed in bank statements, employer against declared employment, continuity across consecutive payslips.
P60
Extracted: employer, tax year, total pay, tax deducted.
Checked: annual income consistency with payslips and declared income — the anchor for whether declared savings are plausible from earnings.
Gift letter
Extracted: giftor name and relationship, gift amount, confirmation the gift is non-repayable, signature and date.
Checked: amount against the declared gift and the credit observed in bank statements, giftor identity requirements, wording covering repayment and interest in the property.
Mortgage offer
Extracted: lender, borrower names, loan amount, property address, offer expiry.
Checked: loan amount against the funds calculation, borrowers against matter parties, property address against the matter, offer validity at expected exchange.
Sale proceeds
Extracted: completion statement figures, property sold, solicitor details, net proceeds.
Checked: net proceeds against the declared contribution and against the corresponding credit in bank evidence, seller names against the matter parties.
Company document
Extracted: company name and number, officers, shareholdings, filing dates.
Checked: declared directorships and ownership against filed records, connection between company payments in bank evidence and the declared funding story.
Investment / crypto evidence
Extracted: platform, account holder, disposal transactions, proceeds and dates.
Checked: account holder against the client, disposal proceeds against the credit that reached the conveyancing account, the traceable path from platform to bank.
What the fee earner actually sees.
£24,500 received from ABC Holdings Ltd on 14 June. Not present in the declared source-of-funds story. Suggested question drafted for the client.
£45,000 credit on 3 June matches the declared gift and the signed gift letter.
The declared savings history requires the March–May statements to evidence accumulation.
Confidence and limits, stated plainly
Every review panel shows what the analysis could and could not read — page quality, unreadable rows, assumptions made. Nothing is presented as more certain than it is.
Fee-earner actions
- Accept a finding and record why.
- Send the drafted question to the client for response.
- Request replacement or additional evidence.
- Escalate the matter to the MLRO with context attached.
- Dismiss a finding — with a mandatory recorded reason.
Every finding ends in a recorded state.
Findings cannot be quietly ignored. Each one moves to a terminal state with an owner and, where required, a reason.
Raised by analysis, awaiting a human look.
Confirmed as understood and consistent with the file.
A question has gone to the client; the finding waits on the answer.
Referred to the MLRO with the full context attached.
Closed with a mandatory written reason that stays on the audit trail.
Documents are checked against each other, not just read alone.
| Data point | Matter record | Document evidence | Result |
|---|---|---|---|
| Name | Daniel Okafor | Passport | Match |
| Salary | £3,250 / month declared | Payslip and bank statement | Match |
| Gift | £45,000 declared | Gift letter and bank credit | Match |
| Corporate credit | Not declared | £24,500 on bank statement | Unresolved |
Illustrative fictional matter. Cross-document results feed matter readiness directly — an unresolved row is a blocker, not a footnote.
When analysis cannot finish, review does not silently pass.
Unreadable document
The document is flagged, the client is asked for a replacement in plain language, and the matter shows the gap as a blocker.
AI provider failure
The document falls back to a review-required state for manual handling. A provider outage never becomes a silent pass.
Incomplete extraction
Partial results are shown as partial — with the unread sections identified so a person knows exactly what to check by hand.
Manual review path
Any document can be reviewed entirely manually. Human findings are recorded in the same lifecycle as automated ones.
Next: the money side of the file.
Findings from documents feed the source-of-funds reconciliation — where the declared story meets the observed evidence.