Case study — 02
Insurance agent co-pilot
Frontline agents were juggling dozens of open customer relationships at once, each with its own history, paperwork and next step. I built the co-pilot that keeps track of all of it — and drafts what to say next.
Suggestion acceptance rate
[placeholder — daily active agents]
[placeholder — reduction in prep time per conversation]
01 The problem
Relationship managers were expected to walk into every conversation already knowing a customer's history, goals and open items — but that context lived scattered across systems, emails and notes. Prep for a single follow-up could take longer than the conversation itself, and it was easy to miss a signal that mattered, like a life event that changed what a customer needed.
The job wasn't to replace the relationship manager's judgment. It was to put everything they needed in one place before they needed it, and to draft the first version of what they'd say next so they could spend their time deciding, not searching.
[A line from an agent about how the day planner changed their morning routine.]
[Frontline relationship manager]
02 What I built
A day planner that opens with a prioritized list of who to follow up with and why, a customer insight panel that surfaces what's changed and what might matter next, and a chat portal that drafts a reply in the agent's voice for them to review and send. I owned the product end to end for a team of more than ten engineers and designers, from the first prioritization model to the tone of the drafted replies.
Recreated illustration — not the shipped product; anonymized for confidentiality.
03 Key decisions
Decision 01
Prioritized queue over a flat task list
A ranked list of who to follow up with first, based on signals like recent activity and time-sensitive events.
[The alternative you weighed — a simple chronological task list, or something else?]
[What convinced you the ranking was worth the added complexity — and what you gave up in transparency or trust.]
Decision 02
Draft-and-approve over fully automated replies
The co-pilot drafts a reply for the agent to review and send; it never sends on its own.
[What a more automated version would have looked like, and why it was on the table.]
[The trust or compliance reasoning that settled it.]
Decision 03
[The trade-off you'd most want to be asked about in an interview]
[What you did]
[What you gave up]
[Why, and whether you'd make the same call again]
04 Outcome
Suggestion acceptance climbed to 30% — meaning close to a third of the drafted replies went out with little or no editing, on a team that had been skeptical an AI could sound enough like them to be useful. [Add how this was measured: baseline, evaluation period, sample size.]
[What shipped next, or what you'd change about the prioritization model in hindsight.]
Confidentiality note
This project was built within McKinsey & Company. Specific client names, proprietary architectures, and internal tooling details are confidential. What's shared here reflects the publicly describable scope of the work; product screens shown are generic, illustrative recreations, not the shipped product.