AI enablementFinancial services
A consumer lender handles 45% more tickets with AI triage and drafted replies
AI classification, account look-up and draft replies inside the helpdesk, with agent approval and a full audit trail, for a regulated consumer lender.
- tickets resolved per agent
- +45%
- tickets resolved per agent
- first-response time
- −58%
- first-response time
- of AI replies approved by an agent
- 100%
- of AI replies approved by an agent
- from audit to production
- 8 weeks
- from audit to production
The client
Quillfield offers personal loans and credit cards to UK consumers. Its support team handles around 40,000 requests a month under FCA rules on fair treatment and record keeping.
The challenge
- ✕Tickets arrived by email, web form and chat and were sorted by hand.
- ✕Agents switched between four systems to find account details for one reply.
- ✕Response times slipped during month-end and payment-date peaks.
- ✕Any automation had to be explainable and fully auditable.
The solution
AI triage
Incoming requests classified by intent, urgency and vulnerability signals, then routed to the right queue.
Context panel
Account, loan and payment details pulled into the ticket so agents stop switching systems.
Drafted replies
AI drafts responses from approved templates and policy, and an agent reviews and sends every one.
Governance
Prompt and model versions logged per ticket, with QA sampling and monthly accuracy reports.
How we delivered
- 01A one-week AI Readiness Audit sized the savings and risks before any build.
- 02A shadow-mode pilot compared AI suggestions with agent decisions for two weeks.
- 03Compliance reviewed prompts, guardrails and audit logs before launch.
- 04Monthly reporting on accuracy, cost per ticket and time saved.
The results
- ✓Each agent resolves 45% more tickets per shift.
- ✓First-response time fell 58%.
- ✓Vulnerable-customer cases are flagged and escalated faster.
- ✓Every AI suggestion is logged and traceable for audit.
“The AI does the sorting and the first draft. Our people make the decisions, and our compliance team can see exactly how each reply was produced.”