Yes. Reconciling safeguarded funds is detailed, repetitive work that an AI agent does well: it collects balances, compares what is owed to customers with what is held, and explains each difference. It does not move money into or out of the safeguarding accounts, sign off the reconciliation, or decide whether a shortfall must be reported. A named person does those.
What is safeguarding reconciliation for?
Payment and e-money firms hold money that belongs to their customers. Most regimes require them to keep it apart from the firm's own money, usually in designated safeguarding accounts or covered by an insurance or guarantee arrangement, so that customers get it back if the firm fails. Reconciliation is how the firm proves, usually every business day, that the amount protected matches the amount owed.
Two numbers are compared. One is the firm's own record of what it owes customers, from its ledger. The other is what actually sits in the safeguarding accounts, from bank statements. If protected money is less than owed, there is a shortfall the firm must fix quickly. If it is more, the firm's own money is mixed in. Either way, the reason must be found.
What decides how far the agent can go?
The deciding factor is whether the step changes money or a regulatory position. Gathering data, calculating, matching and explaining change neither. Moving money to correct a shortfall does, and so does concluding that a break is a reportable breach.
The agent also needs its inputs to be trustworthy. The ledger balance and the bank statement should both arrive from their systems of record, not from a spreadsheet someone edited. A reconciliation built on an edited figure is worse than none, because it looks like assurance.
What may the agent do, and what stays with a person?
| Step | The agent | Stays with a person |
|---|---|---|
| Collecting ledger and bank balances | Reads both at the cut-off time, depth 1 | Nothing |
| Calculating the requirement and the resource | Applies the firm's documented method | Changing the method |
| Matching and explaining breaks | Traces each difference to transactions | Accepting unexplained breaks |
| Proposing a top-up or withdrawal | Drafts it with the calculation, depth 2 | Approving and executing it |
| Recording the reconciliation | Files it with every input | Signing it off |
| Shortfalls past the firm's tolerance | Escalates at once | Deciding whether to notify the regulator |
| Recurring breaks | Groups them by cause | Fixing the underlying process |
Safeguarded funds should not be confused with a customer's source of funds, which is an anti-money laundering question about where money came from.
An example: a break from timing
Take an e-money firm that reconciles each morning against the previous day's close. The agent reads the ledger, which shows 4,200,000 owed to customers, and the safeguarding account statements, which show 4,150,000.
The agent traces the 50,000 gap. Card top-ups received late the previous day were credited to customers' balances in the ledger, but the card acquirer settles them into the safeguarding account a day later. The amount matches the expected settlement exactly. The agent records the break as a timing difference, attaches the settlement report, and drafts a top-up of 50,000 from the firm's own funds to cover the gap until settlement arrives, as the firm's policy requires.
The safeguarding officer reviews the explanation, approves the top-up, and signs off the reconciliation. The next day, when settlement lands, the agent drafts the matching withdrawal of the excess, which the officer also approves.
What controls make it safe?
- Both balances sealed. Ledger and bank figures come straight from their systems, with provenance recorded. Manually entered numbers are flagged.
- No money movement. The agent drafts top-ups and withdrawals; it has no tool that executes them.
- A documented method. The calculation follows the firm's written method; any change needs approval.
- Tolerance triggers. Shortfalls beyond tolerance go straight to the safeguarding officer, whatever the time.
- A complete record. Each daily reconciliation is filed with inputs, breaks, explanations and sign-off, ready for an auditor.
Reconciling correspondent accounts follows a similar pattern with different stakes; see reconciling nostro accounts with an AI agent. Merchant funds awaiting settlement are often among the money a payments firm must protect, which is one reason holding back a merchant's settlement funds needs care. The general limits are in AI agents in banking: what they may and may not do.
Questions people ask
Can the agent run the reconciliation more often than daily?
Yes, and intraday checks can catch problems earlier. The formal reconciliation and sign-off still follow the firm's documented schedule.
What if a safeguarding bank's statement is late?
The agent records that the reconciliation could not be completed and escalates. It does not estimate the balance to fill the gap.
Does reconciliation need customer-level data?
The requirement is often calculated from the ledger total, but investigating breaks may need individual transactions. The agent should read only what each break needs.