The brief
You lead the deployment of an AI support agent at Kestrel Broadband (fictional), an Indian home-broadband provider with 2.3 million subscribers. The 12-week pilot ends on Friday.
The agreed success criteria, signed in the pilot charter:
- Containment (chats resolved without a human): 60%
- CSAT on AI-handled chats: 4.0 or higher out of 5
Actual results over the last 4 weeks:
- Containment: 41%. CSAT on AI-handled chats: 4.2.
- For chats that only needed information (outage status, plan details, router reset steps), containment was 72%.
- About 30% of chats needed account actions (plan change, refund, technician visit). Kestrel's billing API access was promised for week 3 but arrived in week 9; containment on those chat types since week 9 is 48% and rising.
- 38% of chats were in Hinglish. Your team built the evaluation set mostly in English and underestimated this; containment on Hinglish chats is 29%. That was your team's miss.
- Cost per AI-handled chat is ₹9 versus ₹46 for a human agent.
The CTO, Ritu Malhotra, emails you on Wednesday night:
> "I'm hearing the numbers are well off. Steering committee is Friday. What happened, and give me one reason not to shut this down."
Write your reply and the plan you will put in front of the steering committee.
How it's graded
- Leads with the real number, no spin3 pts
- Owns the team's share without blaming the client2 pts
- Uses segmented evidence to support a recommendation2 pts
- Concrete next steps with owners, dates and stop criteria2 pts
- Brief, calm, respectful of the CTO's time1 pts