Analyse a statement file

Sometimes you need to check activity that is not in FlowSentry's transaction book — a statement from another bank, a fund or broker ledger, a mobile-money statement supplied during due diligence or an investigation. File analysis runs such files through your institution's live monitoring rules, statement analytics and screening lists in an isolated workspace. The rows never enter the transaction book, cash reports or dashboards; only the findings you choose to promote become alerts.

Before you start

• You need the Analyst, Senior analyst, MLRO or Deputy MLRO role.
• Have up to five files about one subject (one person or one company): CSV or XLSX, at most 10 MB each. No template is needed.
• Password-protected workbooks are refused; save an unprotected copy first.

Step 1: Start an analysis

1. Open Monitoring › File analysis (/analysis/files).
2. Under Start an analysis, give it a name you will recognise (for example, the subject and the source of the statements).
3. The wizard opens with seven steps: Upload, Understand, Map, Validate, Subject & options, Run and Results.

Step 2: Upload

Add the statements. Each file is stored in FlowSentry's encrypted document vault. A file uploaded twice to the same analysis is refused.

Step 3: Understand — check what FlowSentry found

FlowSentry reads each file's layout for you and reports:

• the header row, even below a letterhead, and the statement details such as client name, currency and period;
• the rows set aside: blank lines, opening and closing balances, totals, repeated headers, footers and disclaimers;
• non-cash rows such as profit/loss or revaluation lines, excluded by default;
• any columns it realigned;
• how amounts are written: one signed column, separate debit and credit columns, or an amount with a direction indicator.

The step is labelled AI-assisted when the AI advisor read a redacted view of the file's structure, or Heuristics only when it did not. Only structure and value shapes are ever sent to a model — never names, account numbers or raw cells.

Check that this matches what you see in the file. Anything wrong is fixed in the next step.

Step 4: Map — confirm what the columns mean

1. Confirm each column's meaning: date, amount, type, narration, counterparty, channel, balance and so on.
2. Confirm the amount convention.
3. Answer How did money move? for each type value: what event it is, its direction and channel — or exclude it. This matters: cash deposits feed cash and structuring rules, bank transfers feed wire, funnel and turnover rules, and securities trades feed market rules.
4. Use Preview to see the first rows exactly as they will be analysed.
5. Select Save mapping. Your edits always win over the heuristics and the AI.

Step 5: Validate — check before running

FlowSentry reads every row with your mapping and shows:

• rows that cannot be read;
• control totals against the file's own total row, and a running-balance check;
• the period and currencies, and any gaps in exchange rates;
• the mix of events;
• a coverage forecast: which of your active rules can judge this file.

Blocking problems stop the run. Warnings need an explicit acknowledgement, which is recorded under your name; changing a file or the mapping later voids it.

Step 6: Subject and options

1. Name the subject, or link a customer already on file so that any promoted alerts land on their profile.
2. Choose the options you need:
   • use the latest exchange rate when no rate was in force on a date;
   • judge sub-day rules day by day when the file has dates but no times (findings are marked as degraded);
   • screen counterparty names from the files as well as the subject;
   • screen instrument and issuer names;
   • ask the AI for a draft findings memo, when AI is available.

Step 7: Run

Start the run. Progress moves through mapping, storing rows, rules, analytics, screening, the memo and saving findings. If a run stays Queued, background processing is not available — tell your administrator.

Step 8: Read the results

1. Coverage comes first. If fewer than a third of your rules could evaluate the file, coverage is called out as limited — "no findings" is then not a clean result. Each rule is classed as evaluated, out of scope (its transaction types or channels are absent), not evaluable, or excluded.
2. Read the activity summary and, if you asked for it, the AI memo. The memo is a labelled draft; it is never attached to an alert or case.
3. Review the findings. Rule findings come from your live rules; statement analytics flag patterns such as near-round amounts, the same amount repeated on one day, an escalation in monthly inflows, rapid drawdowns and small "test" deposits followed by large ones. Data-quality findings flag balance breaks, total mismatches and unreadable rows.
4. Open a finding's Details to see each condition against its threshold and the contributing rows exactly as the file showed them.

Step 9: Promote or dismiss

• Promote a finding that deserves investigation. It becomes an alert in Alerts & cases (/alerts) with the origin Ad-hoc file. A case opened from promoted alerts carries the files and contributing rows.
• Dismiss the rest, giving a reason.

Screening findings are screened for real into the screening workbench.

Note: Rule thresholds are your institution's. If your rules are calibrated for bank-scale amounts, small-value wallet patterns may not trigger on a mobile-money statement. That is a calibration question, not a fault in the file.

Limits worth knowing

• One subject per analysis; up to 5 files, 10 MB each.
• A file that never states its currency is read in your base currency unless you set it on the Map step.
• Rules that need network features cannot be evaluated on a single file.
• Excel formulas are read as their stored values.

Related

• Investigate an alert end to end
• Analyst guide
• AI and data protection FAQ