01

Start with operating pain, not a tool

List the work that repeatedly delays revenue, consumes skilled time, creates avoidable errors, or leaves customers waiting. Describe the current process before deciding whether it needs AI.

A narrow operational question produces a better project than a broad ambition. ‘Reduce the time between an accepted job and a correct invoice’ is actionable. ‘Use AI in finance’ is not.

02

Score the opportunity

Compare candidate processes using four lenses: business value, technical feasibility, operating risk, and adoption effort.

  • Value: time, cash flow, capacity, response, or control
  • Feasibility: accessible data, stable process, and supported integration
  • Risk: cost of an incorrect or delayed action
  • Adoption: clear ownership and a team willing to change the workflow
03

Choose a process with a clear boundary

Strong first projects have a recognizable trigger, a finite outcome, visible exceptions, and an owner. Lead routing, appointment reminders, document intake, or completion-to-invoice preparation often fit this shape.

Avoid starting with a process nobody agrees on. Automation scales the operating design you give it—including confusion.

04

Define success before building

Use measures tied to the work: response time, queue age, incomplete records, rework, unresolved exceptions, days between milestones, or staff time spent coordinating.

A useful first project proves that the process can run differently and creates the operating discipline for what comes next.