AI implementation
Build the AI business case before you build the pilot
Select an AI use case around a real operating problem, compare alternatives and test the assumptions behind value, quality and total cost.
At a glance
- Choose a specific task with a clear beginning and end.
- Compare AI with simpler process and automation changes.
- Test the assumptions that could change the investment decision.
The first AI decision is usually a business decision: which piece of work is worth changing, and why? Starting with a model or a demonstration can make a solution look inevitable before anyone has agreed the problem. A useful business case makes the alternatives and uncertainties visible enough to challenge.
The working method
- 01
Observe
Establish the task, effort and failure patterns.
- 02
Compare
Assess value and complete operating cost.
- 03
Test
Resolve the uncertainty before expanding.
Part 01
Describe the job in ordinary language
Choose a task that has a beginning, an end and an owner. 'Improve productivity' is too broad. 'Prepare a service handover from approved records for a human to review' can be examined. Observe how the work is completed today, including the informal checks that experienced people perform.
Record demand, elapsed time, human effort, error patterns and the consequences of a wrong result. Separate a lack of information from a lack of automation. If staff cannot find an agreed policy, generating a faster answer may simply distribute the confusion more efficiently.
Part 02
Compare AI with simpler changes
Consider process simplification, better search, a deterministic workflow or a reporting change alongside AI. The right comparison is the practical alternative available to the business, not an exaggerated version of today's process. Keep a do-nothing option so the cost of delay can be discussed explicitly.
Score each option against the same task and constraints. Consider variability, integration needs, data readiness, review effort and the cost of failure. AI may be helpful where interpretation is needed; a fixed rule can be preferable where the decision is already precise and stable.
| Question | Evidence | Decision it informs |
|---|---|---|
| Is the work repeatable? | Observed cases, including exceptions | Whether a narrow pilot is viable |
| Are the sources dependable? | Approved documents and record ownership | What preparation is needed |
| What happens when it is wrong? | Operational consequence and recovery route | Review and action boundaries |
| Can the business use the capacity? | A credible operating plan | Whether modeled value can be realized |
Swipe the table to view all columns →
Part 03
Keep the value calculation honest
Model eligible volume and the change in total effort per completed task. Include checking, approvals, correction and escalations. Time removed from one screen may reappear in another team's queue. Released capacity has value only when there is a credible way to use it, and it is not automatically a reduction in payroll cost.
Include implementation, integration, testing, training, tool consumption, monitoring and ongoing maintenance. Show a range of operating scenarios rather than one confident number. Identify which assumptions could change the investment decision and prioritize those in the pilot.
Part 04
Test the uncertain part first
A pilot should resolve an investment question. If the uncertainty is answer accuracy, test reviewed cases and source traceability. If it is integration reliability, exercise failed calls and duplicate requests. If it is adoption, observe real users completing the task rather than asking whether they liked the demonstration.
Use a representative sample and agreed acceptance criteria. Keep excluded cases visible. Set limits on information access and actions from the start, and provide a route to a person when the system cannot complete the task. Record the limitations of the evidence when presenting the result.
Part 05
Make the next investment a decision
Conclude with a choice: stop, improve the underlying process, run another bounded test or expand a specific part of the workflow. Name the owner of the next decision and the evidence still missing. Do not let a successful demonstration quietly become permission for unrestricted production use.
A sound business case can justify choosing less technology. Its purpose is to direct investment toward a useful, supportable change. The strongest output is a decision the business can explain, with assumptions it knows how to revisit.
- A defined task, baseline and operating owner.
- Comparable alternatives, including a simpler intervention.
- A complete cost model and a plan to realize value.
- A pilot that tests the uncertainty behind the investment.
