← AI BUSINESS SYSTEMS
DIAGNOSTIC / 15 POINTS

Find the workflow worth automating first.

The best first AI project is not the flashiest one. It is a frequent, bounded and measurable workflow where people can retain decisions, exceptions and control.

READ THE WORKED EXAMPLE AND GUIDE ↓

Educational prioritization self-assessment. It is not a guarantee of ROI, security, compliance or technical feasibility.

01Frequency and valueDoes the workflow happen often enough to justify redesigning it?
02Structure and accessDoes the work have inputs, rules and systems that can be bounded?
03Control and oversightCan we test the workflow without handing sensitive decisions to a black box?
HOW TO USE THE RESULT

The score prioritizes workflow shape, not technology hype.

01A DIGITAL WORKER

Fits best when the work belongs to a clear queue, role or responsibility: intake, research, scheduling, follow-up or documentation.

02AN AI BUSINESS SYSTEM

Fits best when value depends on coordinating several stages, systems or handoffs while preserving context between them.

03DO NOT AUTOMATE YET

If the process constantly changes, has no owner, uses inaccessible data or cannot be measured, redesign the workflow before automating it.

READING AND APPLICATION GUIDE

Score a unit of work, not an entire department.

Choose a named workflow with a recognizable start and finish. “Improve operations” is too broad; “prepare a reviewed response to a complete service request” is testable. Answer for the process that exists today. If access, ownership or exception handling is only planned, leave the corresponding point unchecked.

Worked example: frequent work with weak oversight

A fictional coordinator evaluates recurring service-request preparation. The workflow earns 5/5 for frequency and value, 4/5 for structure and access, but only 1/5 for oversight: 10/15 in total. The missing oversight is not compensated by high volume. No one has yet agreed who accepts exceptions or how to stop an incorrect action.

A sensible first pilot produces drafts in a review queue. The coordinator compares them with the current process, records corrections and defines escalation ownership. Only then is it meaningful to discuss writing to live systems. A high total is a reason to investigate a pilot, not permission to make it autonomous.

Practical criteria for reviewing the example
DimensionWhat to observeCommon blind spot
Frequency and valueHow often work arrives, where it waits and what a useful result looks like.Counting clicks saved without counting new review work.
Structure and accessInputs, approved sources, repeatable rules and actual permissions.Assuming a tool can be integrated because someone can open it manually.
Control and oversightWho approves, who receives exceptions and how recovery is tested.Treating an unassigned notification as a completed handoff.

An exercise for your project

  1. Describe the start, finish and owner in one sentence; collect a small representative sample of past work.
  2. Include incomplete requests, contradictory records and a tool outage in the sample. Agree what a good response would be.
  3. Re-score after ownership and access are confirmed. Use your notes as the pilot brief rather than treating the total as a business case.

Concepts worth distinguishing

Baseline
Observed performance of the current process, including waiting and review time.
Exception
A case outside approved rules that needs an identified person or a defined alternate path.
Reversible pilot
A bounded test that can be stopped without losing the authoritative work record.