Start with the work, not the technology
Organizations often begin by asking where they can “use AI.” A better question is: which recurring workflow consumes skilled time because people repeatedly gather, organize, compare and draft before they can make a decision?
That framing produces a testable business opportunity rather than a speculative AI project.
Five signs of a strong candidate
Repeated frequently
The workflow happens often enough for improvements to accumulate.
Several information sources
People repeatedly assemble context from documents, systems, messages or meetings.
Preparation is recognizable
The work has repeatable stages even when individual cases differ.
Outputs can be reviewed
A knowledgeable person can evaluate completeness, accuracy and usefulness.
Ownership is clear
A named leader owns the process and remains accountable for its decisions.
Red flags that suggest waiting
- No one can describe how the workflow currently operates
- The desired output changes every time
- There is no accountable owner
- Source information is unavailable or unreliable
- Errors cannot be detected before harm occurs
- The project depends on unrestricted system access
- The proposed value is based only on reducing headcount
Establish the baseline before building
Record how long the work takes, who participates, where information comes from, which errors or delays occur and what a useful output looks like. Without a baseline, it is difficult to distinguish genuine improvement from an impressive demonstration.
Choose the smallest useful starting point
A focused agent may be enough to prepare one brief, compare evidence or draft one approved output. If that capability works, it can connect to additional stages. Starting small reduces risk and creates evidence for the next decision.