Financial services & investment management
Agents now draft credit memos, settle claims and act for customers. Accountability stays with the institution.
The lines that matter Cost-to-income, net interest margin, credit losses, loss and expense ratios
Banks, insurers and wealth managers have moved from AI that drafts and summarises to agents that carry out multi-step work: assembling credit memos, handling claims and completing customer requests.
Regulators in Singapore, Hong Kong, Europe and the US are moving in step, and every one of them keeps accountability with the institution, including for models bought from vendors. The advantage will go to firms that redesign delegation, accountability and controls before the rules are set for them.
- Cost-to-income and expense ratios under pressure from competitors with lighter operating models
- Rising compliance cost, from KYC refresh to regulatory reporting
- Advisers and relationship managers spending too little time with clients
- New supervisory expectations for AI, including inventories, materiality and third-party AI
- Fragmented customer and product data across legacy systems
- Model risk frameworks that do not yet cover agents
- Vendor AI embedded in software without an inventory
- Unclear approval thresholds for agent actions
Where AI can move the numbers
Opportunities grouped by Agent Value Ledger category. They are possibilities to test, not promised results.
Revenue
Faster credit and underwriting decisions, adviser preparation, renewal and retention
Operating costs
Onboarding and KYC, credit memos, claims handling, operations exceptions
Capital
Collections, early-warning monitoring, liquidity and cash forecasting
Risk
Fraud and AML alert triage, covenant monitoring, conduct surveillance
The lines this sector is judged on
Banking
- Net interest income
- Faster credit decisions, pricing discipline, deposit retention
- Cost-to-income ratio
- Onboarding, KYC refresh, credit memos, operations exceptions
- Credit losses and provisions
- Early-warning monitoring, covenant checks, collections
- Compliance cost
- Triage of AML alerts and regulatory reporting
Insurance
- Gross written premium
- Quote turnaround, broker servicing, renewal retention
- Loss ratio
- Claims leakage, fraud detection, recovery and subrogation
- Expense ratio
- Claims handling, underwriting preparation, policy servicing
- Combined ratio
- The sum of the two, and the number boards are judged on
Wealth and family offices
- Revenue per adviser
- More client time per adviser as preparation and reporting shift to agents
- Cost to serve
- Reporting, rebalancing proposals, document handling
- Client retention
- Proactive service and next-best-action prompts
- Compliance cost
- Suitability checks and record-keeping
A credit memo agent in SME lending
Assumptions for a mid-sized bank: 4,000 SME credit applications a year; average facility US$400,000; net interest margin 2.5%; analyst cost US$90 an hour, fully loaded.
| Line | How the agent moves it | Measured by | Working | Per year |
|---|---|---|---|---|
| Revenue | Decisions in four days instead of ten win more approved deals | Conversion of approved applications | +2 points × 4,000 × US$400k × 2.5% | +US$0.80m |
| Operating cost | Analyst time per memo falls from 6 hours to 2.5 | Hours per memo; cost-to-income | 3.5 hours × 4,000 × US$90, counted only if redeployed or hiring is avoided | +US$1.26m |
| Credit losses | Consistent covenant and data checks on every file | Early-warning hits; expected loss | Not counted until 12 months of evidence | Excluded |
| Cost of the agent | Model usage, integration, oversight, model validation, assurance | Total cost of ownership | Annual run cost | −US$0.60m |
| Net annual value | ≈ +US$1.46m, before any credit-loss benefit | |||
Hypothetical illustration of the method. The figures are assumptions, not a client result or a forecast.
- Accountability stays with the institution, including for vendor models
- For Singapore institutions, MAS guidelines on AI risk management set expectations for inventories, materiality and life-cycle controls
- Human authorisation thresholds by product, value and customer type
AI Readiness & Value Diagnostics
Start with an Enterprise AI Value Creation Assessment focused on cost-to-income and credit or claims operations, then a governance review aligned to MAS expectations.
About this serviceFinancial Services in the Age of Agency
Banks, insurers and wealth managers are now letting AI agents carry out work, not just advise on it, so the advantage will go to firms that redesign delegation, accountability and controls before regulators set the rules for them.
Read the brief →Governing a bank's first credit agents
Situation. A mid-sized bank's pilot agent drafts credit memos well. Leaders want it rolled out within six months, but nobody can say exactly what it may do or who answers when it is wrong.
Our approach. A delegation map with approval thresholds by loan size, an agent register with named owners, a plan to suspend or replace the vendor model, and a workforce plan for the roles that change.
What good looks like. The bank scales the agent with board approval and a governance model it can show its regulator.
Scenarios and worked ledgers are hypothetical illustrations. They do not describe specific clients.
Start with an executive conversation
A confidential discussion of your priorities, and where AI could move the numbers in your business.