What Stays Human

Merrow’s board has approved an AI-first operations plan. The specific proposal on the table: an agentic workflow that takes an inbound service request end to end — triage it, schedule a technician, select and order parts, and issue a quote — with no human touching it when the quote comes in under $2,500. 

Nineteen operations staff touch those four steps today. The CFO has committed the board to $2.1M in annual operating cost reduction from this program, in this fiscal year. 

Here is what you know. 

Per-step accuracy, measured over six weeks of shadow running against human decisions: 

Step	Agent accuracy	What “wrong” costs
Triage (priority + trade code)	96%	A missed P1 on a chain account: $5,000 contractual penalty, plus product loss the customer may claim
Scheduling (right tech, right window)	93%	A wasted visit: $340 and a burned SLA clock
Parts selection	88%	A repeat truck roll: $340, plus a second customer touch
Quote generation	91%	Under-quote margin loss averaging $610; over-quote drives disputes and rework

Volume and scope 

• 3,100 requests per month fall under the $2,500 threshold — roughly 62% of all requests by count, about 21% by revenue. 
• Median end-to-end handling time today: 5.5 hours. Shadow-run agent median: 11 minutes; p95: 4.2 hours (the tail is parts availability lookups). 
• Fully loaded cost of the 19 operations staff: $2.4M/year. 
• Current human-run accuracy on the same four steps, measured the same way: triage 98%, scheduling 95%, parts 94%, quote 93%. 

Governance 

• Chain customers can request, under contract, the reasoning behind any quote for 24 months after it is issued. 
• Merrow’s insurer requires that any automated decision affecting a safety-related dispatch be reversible and logged. 

Write the automation boundary and make the call. 

Constraints to consider 

• The $2.1M target is fixed for the fiscal year. The CFO has already told the board. You can change how you get there; you cannot change the number without going back to the board, and you must say so explicitly if that is your recommendation. 
• Any headcount change requires 60 days’ notice to the affected staff before it takes effect. Today is August 10; the fiscal year ends June 30. 
• You have one engineer after month three. Whatever you design has to be operable by that one person plus the operations team that remains. 
• You cannot retrain or replace the models. You can change thresholds, routing, scaffolding, guardrails, the human review path, and what an operator sees. 
• You will present this to a leadership group that includes the CFO, who wants the full $2.1M, and the VP of Operations, who has told you privately that she thinks the whole thing will cost more in errors than it saves. 

AI Usage Guidance 

We expect you to use AI tools. We evaluate how you use them — not whether you use them. Evidence of iteration, redirection, and critical evaluation scores higher than a polished output with no process documentation. 

The single highest-signal indicator: your video answer to the mandatory AI question. If you cannot name a specific moment where you redirected AI output, evaluators will assume you did not. 

Mandatory AI question for your video: Walk me through one moment where you disagreed with, pushed back on, or redirected what the AI gave you — and what you did instead. Name the specific moment. Explain what the AI produced that didn’t meet the bar, what you did differently, and why. 

Communication clarity matters for this role. We assess structure, stakeholder awareness, and ability to simplify complex ideas — not presentation style or accent.