Strategy & value

Make AI an operating capability

Five decisions that turn an AI initiative into something a business can own and improve.

The decision this supports

The unit of progress is a workflow the business can accept, operate, and improve. Define those conditions before scaling the technology.

For executives and transformation sponsors

01 / Perspective

Start with a workflow

Define the work to be completed, the person accountable for it, and its current operating baseline. An attractive demonstration is useful evidence of possibility; the investment decision also needs a credible route to everyday use.

For client onboarding, define completion as a reviewed case that meets the service requirements. Measure the time spent gathering information and resolving gaps as well as the time spent producing the initial summary.

Ask the workflow owner to define a case that should be accepted, one that should be returned for correction, and one that cannot proceed. These examples establish the boundary of the capability more clearly than a broad target such as “improve productivity.” Keep queue time, handling effort, and correction effort separate in the baseline.

02 / Perspective

Design the human role

Decide who supplies context, reviews exceptions, approves consequential actions, and accepts the result. Make those responsibilities part of the workflow design.

Separate preparation from commitment. An assistant may extract information and prepare a recommendation; an accountable person can retain approval of an external communication or a change to a business record.

Make the review interface useful. Show the evidence behind a recommendation, the unresolved issues, and the action that approval will authorize. An approval button without that context creates a new responsibility for the reviewer without necessarily giving them the means to exercise it.

03 / Perspective

Test the difficult cases

Evaluate incomplete information, conflicting sources, interruptions, and unavailable systems. Record where the system must ask for help or stop.

Build an evaluation set before rollout. Include a complete case, a missing document, contradictory records, an expired policy, and an interrupted handoff. Agree which results are acceptable and when the system must ask for review.

Assign a domain reviewer to expected outcomes and an engineering owner to failure behavior. Do not let a good average result conceal a serious failure on restricted information or approval boundaries. Decide which conditions block release before evaluating the candidate implementation.

04 / Perspective

Plan ownership before rollout

Agree access, code and configuration ownership, documentation, operating procedures, and support. Identify the internal owner who can make ongoing decisions.

The handover should include the source inventory, configuration, evaluation cases, operating handbook, and named owners. Agree how the team will evaluate a model change and how it will pause or revert a workflow change.

Confirm the receiving team can demonstrate the operating procedures. Have them locate a failed case, pause new work, inspect a source update, and evaluate a proposed change. Agree what remains dependent on a supplier and what can be operated or changed internally.

05 / Perspective

Measure the work after release

Review completed work, quality, rework, staff effort, adoption, and operating cost together. Use this evidence to decide whether to expand, adjust, or retire the capability.

A faster draft can still create more work downstream. Include reviewer time and correction effort in the measure. Compare representative cases over an agreed observation period before expanding the scope.

Use an operating review to choose an action. A recurring source gap may call for better source ownership; excessive review effort may call for a narrower task. Expansion is one possible decision. Redesigning or retiring a workflow that does not deliver value is also part of owning the capability.