Context
Council Tax calls combine policy rules, local exceptions, vulnerability signals, and strict time pressure. Agents are expected to make accurate decisions while keeping the call moving and staying compliant. Most organisations respond with more training, but the operational reality is different: even well-trained people still need support at the exact point of decision.
This project explored a different pattern. Instead of treating support as a separate learning activity, the product would sit inside the live workflow and surface the right guidance when the work demanded it.
Problem framing
Three constraints shaped the problem:
- Guidance had to be fast enough to help without interrupting the call.
- Every suggestion had to be auditable against policy and system evidence.
- Human judgement moments had to be protected, not automated away.
The core failure mode is not knowledge absence. It is timing failure: the right knowledge is inaccessible when the decision has to be made.
The practical design question became: how do we reduce decision friction for frontline staff while preserving accountability and escalation boundaries?
Approach
The product was framed as a workflow coach with explicit guardrails:
- Listen: live transcript ingestion and intent detection from call context.
- Ground: retrieval from policy sources, case patterns, and relevant records.
- Guide: compact suggestions with confidence, rationale, and source links.
The interaction model focused on progressive disclosure. Agents see concise prompts first, with deeper policy explanation one click away. Escalation triggers are visible and configurable so the system can support judgement rather than replacing it.
What was built
The delivery produced a full concept stack:
- Product PRD and service framing.
- Domain model and data design for guidance generation.
- UX specification across representative case journeys.
- AI prompt and grounding architecture with source traceability.
- Analytics and QA framework for confidence, uptake, and error monitoring.
- MVP backlog and implementation plan for phased rollout.
The prototype demonstrated how policy-grounded AI could fit the operational rhythm of contact-centre work without introducing a second workflow.
Demonstrated
The project showed that learning innovation and AI product design are strongest when combined in one system. Instead of asking teams to remember everything from training, the tool shifts support to the moment where mistakes are costly and time is limited.
The outcome was a credible blueprint for council environments: operationally realistic, technically grounded, and clear about where human discretion remains essential.