Sell process improvement
Clients understand time lost copying data or chasing approvals. They understand a proposal based only on models, prompts and agents much less.
Describe current volume, time per case, errors and target outcome. Technology becomes credible when tied to an operating metric.
Do discovery before the demo
Observe the process with the people doing it and collect normal examples, edge cases and unwritten rules. Identify systems, permissions and actions needing confirmation.
Then offer a short paid phase delivering a workflow map, risks, prototype and return estimate. Do not promise full automation immediately.
Build the exception path
The agent must know when to stop, ask or hand a case to a person. A review queue is part of the product, not a failure.
Use minimum permissions, logs, idempotent identifiers and tests on real examples. These choices separate a professional service from a fragile demonstration.
Make pricing and maintenance explicit
Separate discovery, implementation and ongoing operation. Recurring cost covers monitoring, providers, updates and evaluation improvement.
Show variable costs and thresholds. Useful automation needs observation because processes, data and external services change.