Start with a specific handoff
A customer sends an email with an attachment. A manager needs enough information to prepare the next step. The automation task is to turn that message into a structured request the manager can check.
This is the pattern behind part of my Grand Veil work: email intake, AI-assisted extraction, a manager review workspace and a connection to CRM records. The catalogue and configurator solve adjacent parts of the business workflow. Each part has its own job.
Define the draft before choosing the model
Write down which fields the next person actually needs. For a made-to-order request, that might include the requested product, measurements, quantities and unanswered questions. This is an example field list, not a universal schema.
A missing measurement should remain missing. A model should not turn a plausible guess into an agreed customer requirement. The review screen needs to let a person compare the draft with the source and correct it.
Keep review in the workflow
Extraction is one stage. Approval is another. A useful draft makes the manager’s next decision easier: check the details, ask for clarification or move the request forward.
Define which actions the automation may take on its own and which require a person. Creating an internal draft is different from confirming a customer specification or releasing work to production.
Make failures visible
Before building, agree what happens when an attachment cannot be read, a field is uncertain or a downstream system is unavailable. A request should have a visible state and an owner for the next action.
Scope a first version
Choose one incoming channel, one request type and one destination. Test with representative examples, including incomplete and ambiguous requests. Measure review effort and correction rates before claiming time savings.
Explore the Grand Veil example, or describe the workflow you want to automate.
