Our team wants AI to route every incoming service request. How should a forward-deployed engineer decide which requests actually need AI?
Start with the routing rules your team can state reliably, and use AI only for requests that need interpretation. A forward-deployed engineer should observe how requests arrive, where employees make judgment calls, and what happens when a request goes to the wrong place.
For example, an explicit account number or a selected service category may be enough for a conventional rule. A free-form message describing several problems may benefit from AI classification. The engineer should test both approaches against a representative set of past requests, with sensitive details removed where practical.
Agree on the cost of a mistake before setting a target. A request sent to the wrong internal queue may be recoverable. A safety issue or urgent customer complaint sitting unnoticed is more serious and may need a separate escalation rule.
- Define routing categories and an owner for exceptions.
- Compare AI suggestions with staff decisions on a sample that includes ambiguous cases.
- Send uncertain or high-consequence requests to a person rather than forcing a classification.
- Track misroutes and time to resolution, not just the number of requests processed.
If the work supports a quarterly EOS Rock, make the intended business result explicit. The engineer can build the workflow around your existing operating system, but the leadership team should decide whether the result is useful enough to expand.
Category: Forward-Deployed Engineering