Risk managers in Nigerian retail banks need training to set and test AI agents' bounds on instant transfers, failed-transaction disputes and agency banking oversight. The core skills are reading verdicts, checking provenance and approving held actions. CBN expectations and the Nigeria Data Protection Act 2023 frame the work. Fraud conclusions and risk appetite stay with people.
What fills a Nigerian retail bank risk manager's day?
Nigerian retail banking moves fast and at volume. Instant transfers are the default, USSD channels serve customers without smartphones, and a large network of banking agents with point-of-sale terminals brings cash services to places without branches. Each channel has its own fraud pattern: social engineering on mobile, compromised terminals and fake reversals in agency banking, and insider collusion in account opening.
The risk function covers operational and fraud risk, credit risk in consumer and SME lending, IT and cyber risk, AML and sanctions, and customer complaints, with reporting to the board and the Central Bank of Nigeria. Failed-transaction disputes are a constant: a customer is debited but the beneficiary is not credited, and the bank must resolve it within the regulator's timelines.
A note on terms. Nigerian bankers already use "agent" for the people who run agency banking outlets. Training should make the difference clear early, or every conversation about AI agents will be confused.
Which uses come first, and at what depth?
- Failed-transaction disputes, depth 2 to 3. The agent reads the switch and core banking records, confirms whether the debit and credit legs completed, and within bounds reverses a clear failed debit. Anything ambiguous, or above the bank's own limit, is held. The customer's screenshot is external and cannot settle the dispute either way.
- Agency banking monitoring, depth 1. The agent reads terminal transaction data and flags outlets with unusual reversal rates, split transactions or out-of-hours activity for the fraud team.
- Account opening quality checks, depth 2. The agent compares onboarding documents with the national identity and bank verification records and drafts exceptions.
- Loss event capture, depth 2. Fraud and operational losses drafted into the register with cause and channel.
Freezing accounts, releasing large held transfers or writing off losses are depth 4 and held for a named approver.
What must risk managers learn, and what do they keep?
Every risk manager should be able to read an agent's verdict and the record behind it. When a dispute is held, the verdict names the bound that decided: perhaps the amount exceeded the bank's own limit, perhaps the switch record was missing. Our guide to how risk managers review an agent's decisions sets out the routine.
Provenance comes next. Core banking and settlement records are sealed. A confirmation from the beneficiary bank is signed. A customer's account of events is declared. A forwarded alert or screenshot is external, and an external claim cannot authorise a payment. Fake reversal messages are common enough that this distinction needs drilling.
Training also has to reach beyond the risk team. Operations, customer service and fraud staff will approve held actions too; see which roles in retail banking need AI agent training.
What stays with people:
- conclusions that a customer, staff member or banking agent committed fraud;
- suspicious transaction reports;
- credit policy exceptions and risk appetite;
- reporting to the board and the CBN.
Which Nigerian rules shape oversight and competence?
The Central Bank of Nigeria regulates and supervises banks, payment service providers and agency banking. Its expectations on risk management, IT and cybersecurity, consumer protection, dispute resolution timelines and AML reach any agent that acts in a bank's name. A bank should be able to show the CBN which agents resolve disputes or monitor channels, at what depth, and who supervises them.
The Nigeria Data Protection Act 2023 governs customer data, including identity numbers, biometric data from onboarding and transaction histories. It requires a lawful basis, purpose limits, security, and respect for data subjects' rights, including where decisions are automated. The Nigeria Data Protection Commission enforces it.
For AML, the FATF Recommendations set the international standard, with a risk-based approach, customer due diligence and suspicious transaction reporting. Basel Committee principles on operational resilience apply to banks generally, and model risk is supervised in most banking markets.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
A 90-day plan
| Days | Focus | Output |
|---|---|---|
| 1 to 30 | Foundations for risk, fraud, disputes and agency banking teams; clear terminology on "agent" | Inventory of AI agents with depth and lapse dates |
| 31 to 60 | Replay a month of failed-transaction disputes at depth 1; set the bank's own reversal limits; approvers sit the operator exam | Limits documented; exam results |
| 61 to 90 | Two-week practicum on live disputes and agency alerts; calibration across regions | Practicum sign-off; board report |
The training plan template for risk managers turns this into a working plan, and the full role map is in AI agent skills by role.
Where the Academy fits
Pryme Intelligence Academy runs every lab in a free Build workspace, so risk staff can test how an agent handles a fake reversal before it touches live work. Approvers become Certified Pryme Operator after a practicum of two weeks of real review decisions with no missed red line. A bank can tie the right to release held disputes to a current certificate.
Questions people ask
Can the agent reverse a failed debit on its own?
Within bounds, once certified for that capability, if both legs of the transaction are clear in sealed records. Ambiguous cases and amounts above the bank's own limit are held for a person.
How should banking agent outlets be told about AI agents?
Briefly and plainly: what the bank's AI agents monitor, what they will never do, and how outlet operators can query a flag. They do not need certification unless they approve held actions.
Does the NDPA allow agents to use biometric data?
Biometric data needs strong justification and safeguards. Keep it out of an agent's sources unless the capability requires it, and confirm the basis with your data protection officer.