Key takeaways
- AI is useful in reconciliation for suggesting matches, classifying breaks, explaining likely causes and drafting investigation notes.
- Decisions that move money or affect customer funds should remain with accountable people, with AI providing information rather than approval.
- Regulators expect firms to understand and govern any AI they use, including how its outputs are checked.
- Evaluate AI features on whether they reduce the number and age of open breaks without weakening the audit trail.
Where AI helps
Reconciliation produces large volumes of structured data and a steady stream of exceptions that people must investigate. That makes it a reasonable place to use AI, provided it is used for the right tasks. AI can suggest matches between records that rules miss, for example where references are incomplete. It can classify breaks by likely cause, so they reach the right team faster. It can summarise the history of a recurring break, draft an investigation note for a person to review, and highlight patterns, such as a processor whose fees have changed, that would take a person hours to find.
Each of these reduces the time people spend gathering information, so they can spend more time on judgement.
| Task | Role for AI | Role for people |
|---|---|---|
| Matching routine items | Suggest matches beyond fixed rules | Approve new matching rules |
| Classifying breaks | Propose class and likely cause | Confirm for material breaks |
| Investigation | Gather history and draft notes | Decide cause and resolution |
| Moving money | Prepare the proposed transfer and supporting figures | Approve and execute |
| Regulatory reporting | Assemble figures from reconciled records | Review and sign off |
Where people must decide
Some decisions should not be delegated. Whether a difference is a timing issue or a real shortfall, whether to move money into or out of a safeguarding account, and whether a regulatory return is accurate are judgements with legal consequences. Under the UK's CASS 15, the firm must determine the cause of every difference and correct shortfalls the same day. Under the US GENIUS Act, the CEO and CFO personally certify reserve reports. In each case, an accountable person must be able to explain the decision.
AI should therefore prepare and inform these decisions, with its suggestions recorded alongside the person's final decision. That gives the firm the speed of automation and the accountability regulators expect.
Governance expectations
Regulators do not prohibit AI in financial operations, but they expect firms to understand what it does, test that it works, monitor it over time and keep people accountable for outcomes. In practice, that means documenting each AI use case, recording AI suggestions and the human decision for every material item, measuring accuracy against outcomes, and being able to switch an AI feature off without stopping the process.
Data protection and security also apply. AI features that send customer or transaction data to external services must meet the same requirements as any other outsourcing, and firms should know where data goes, how it is protected and whether it is used to train models.
How to evaluate AI features
Ask vendors to show AI features working on your own data. Measure whether they reduce the number and age of open breaks, as described in our guide to reconciliation KPIs, and whether the audit trail still shows who decided what. Be cautious of features that hide the reasoning behind a match or close breaks without a person's review.
Safeheld uses AI to suggest matches, classify breaks and prepare investigation summaries, with every material decision and approval retained by people and recorded. See reconciliation break management and Safeheld reconciliation.
Frequently asked questions
How is AI used in reconciliation?
To suggest matches, classify breaks, explain likely causes, summarise history and draft investigation notes.
Should AI approve safeguarding transfers?
No. Decisions that move money or affect customer funds should be approved by accountable people, with AI providing information.
Do regulators allow AI in financial operations?
Yes, but they expect firms to understand, test, monitor and govern it, with people accountable for outcomes.
How should firms evaluate AI reconciliation features?
On real data, measuring the effect on open and aged breaks and confirming the audit trail shows who decided what.
References
- FCA, PS25/12: Changes to the safeguarding regime for payments and e-money firms
- FCA Handbook, CASS 15.8 Records, accounts and reconciliations
About Safeheld
Safeheld reconciles client money every day, resolves every difference and prepares every regulatory return for sign-off, for regulated firms in the United States, the United Kingdom, the European Union, Canada, Australia, Singapore and Hong Kong. To discuss your operations, contact us at safeheld.com/contact.
This article is for general information and does not constitute legal or regulatory advice.