AI use-case brief · Fictional example
Customer enquiry reply drafts
Customer Operations at a fictional electricity retailer
- Stage
- Discovery
- Example
- Complete
Problem and current approach
Problem: Agents spend substantial time assembling routine replies, and customers receive answers with inconsistent detail. Today: An agent searches approved guidance, reviews account information, writes a reply and asks a specialist for help when needed. Searching and drafting create most of the delay.
Evidence and outcome sought
Evidence: Team observations show repeated policy searches and rework. Response-time and correction-rate baselines still need to be confirmed before any trial. Outcome: Give customers faster, more consistent routine answers while keeping agents responsible for what is sent.
How success would be measured
Establish the current median handling time, then compare it during a trial. Track agent correction rate and substantiated complaints linked to drafted replies.
People and responsibility
People: Service agents would use the drafts. Customers, including people experiencing financial hardship, could be affected by the final communication. Owner and decision: The Customer Operations Director owns the outcome. The Customer Service Risk Owner decides whether the idea may move to a controlled trial.
AI-supported task
Prepare an editable response draft using approved guidance and selected account facts, with links to the guidance used.
Information, output and action
Information: Approved service guidance and selected customer account fields. Account data may contain personal and financial-hardship information. Output and action: The system produces a draft. A service agent reviews, edits or rejects it and decides whether to send a response.
Human responsibility and initial scope
Human responsibility: The agent checks account facts and cited guidance. The agent can reject the draft and must escalate hardship, safety or disputed-account enquiries. Initial scope: Begin with routine written enquiries. Exclude hardship, safety, disputed accounts and automatic sending.
Other approaches and consequences
Other approaches: Improved templates and policy search were considered. Drafting may add value where a reply must combine several approved sources with customer circumstances. If it goes wrong: A wrong draft could expose incorrect account information, misstate customer rights or give vulnerable customers inconsistent service.
Dependencies and unknowns
Dependencies: Approved guidance must remain current. Access to account fields must be limited. Agents need time and authority to review every draft. Supplier and model limitations are not yet confirmed. Unknowns: Current performance baseline, privacy review, acceptable failure level, accessibility review and supplier data-handling terms.
Proposed next step
Customer Operations should confirm the baseline and sample workflow with Privacy, Security and the Service Risk Owner before deciding whether to design a restricted trial.