Complaints handling
Case summarisation, root-cause classification and first-draft response against your own precedent - with the decision, redress and sign-off staying with the case handler.
Industries
AI in a regulated firm, evidenced from day one
In financial services the constraint is rarely capability - it is being able to show, afterwards, how an outcome was reached, who was accountable and why it was fair. We build AI that produces that evidence as a by-product of doing the work.
The pressure right now
What we hear from financial services
- Consumer Duty obligations to evidence good outcomes across every customer journey
- Complaints volumes and root-cause analysis expectations rising
- KYC, AML and periodic review backlogs absorbing operational capacity
- Model risk expectations extending to AI tools nobody classified as models
- Supervisory and audit questions arriving faster than the evidence can be assembled
Where it pays off
Six places AI earns its keep here
Not everything on this list will apply to you. Most organisations start with one and extend once it has been measured.
KYC, AML and periodic review
Document gathering, adverse-media screening triage and risk-assessment drafting, with the MLRO decision and record-keeping unchanged.
Suitability and file quality
Checking advice files for the evidence a reviewer would look for, and drafting the client-facing explanation in plain language against your templates.
Policy and product retrieval
Grounded answers for front-line staff drawn from your actual product terms and procedures, with citations, replacing the wiki nobody trusts.
Quality assurance sampling
Extending QA coverage from a sample to something closer to the whole population, with the exceptions routed to human reviewers.
Regulatory horizon and change
Tracking consultations and policy statements against your own product and process inventory, so the impact assessment starts with a real list.
Where we would start
The first three moves
1
Classify the AI you already run
Including the tools embedded in supplier products. Model risk governance now has to reach them, and most firms' registers do not.
2
Take one customer journey
Complaints is the usual candidate: high volume, well-defined outcomes and a direct Consumer Duty read-across.
3
Make the evidence automatic
Design so the audit trail is a by-product of the workflow rather than a report someone compiles later.
Risk and regulation
The part most suppliers skip
Where the risk sits
- Consumer Duty: outcomes must be evidenced, including for customers with characteristics of vulnerability
- SM&CR accountability - a named individual is answerable for an AI-influenced decision
- Model risk governance extending to AI tools that were never registered as models
- Operational resilience and third-party dependency on AI providers
- Fairness and explainability where an outcome affects a customer's access or price
How we handle it
Accountability is designed in: every AI-influenced decision has a named human owner, a recorded rationale and an override that is logged rather than silent.
Consumer-affecting decisions are built to be explainable in the language the customer will receive, not only in the language the model uses internally.
Vulnerability signals are treated as escalation triggers to a person, never as a routing efficiency - and we test that behaviour before go-live.
Where this fits
Where to go next
Questions
Questions from financial services
The regulator's consistent position is technology-neutral: existing obligations apply regardless of the tool. What matters is governance, accountability, fairness and evidence - which is exactly what we build for. Firms get into difficulty over absent evidence, not over the technology.
In most firms, yes, and it usually does not yet. Extending the existing framework is far cheaper than building a parallel AI governance structure, and supervisors respond better to it.
We do not build it that way. It assembles evidence, drafts and recommends; the decision stays with an accountable individual. That is both a regulatory position and, in our experience, the design that actually performs better.
Start with an audit of what you already run
Two to four weeks to an evidenced picture of your AI use, spend and risk - and a ranked list of what to do first.