S18 · lifecycle simulation

A case-summary assistant from intake to model retirement

People, data and outputs in this scenario are simulated. Each stage is recorded when you commit; the reveal comes after.

5stages
45 minabout
0stages committed

The brief

A customer-care team wants an assistant that summarises long conversation threads into a five-line case summary so that agents can pick up a case without rereading it. All numbers are simulated; prices are hypothetical.

Volume: 38,000 threads a month. Today agents read the thread; the average handling time is 6.0 minutes.
The assistant calls a hosted model from a vendor (details below).
A fallback exists: the first three messages of the thread, truncated.

Stage 1 of 5

The product owner asks for "summaries that agents like". You have two weeks before the build starts.

What do you define at intake?

"Agents like it": run a satisfaction survey after two weeks of use and keep the feature if the average rating is above 4 out of 5, because the agents are the users and they know the work best.Match the style and length of the agent-written notes already in the system, measured by an overlap score against them, and ship when the score passes a level chosen after the first run.Two measured things: summary faithfulness and completeness judged on about 300 labelled threads, and agent handling time in a controlled rollout, with the truncation fallback as baseline.Cost per thread under 0.02 USD and latency under 3 seconds, since a quality bar is hard to define for free text and the budget is known.