The traditional business case asks a project to predict cost, return and risk before work begins. This is sensible when technology and requirements are stable. AI initiatives often contain too much uncertainty for that precision to be honest.
Model capability changes, employees learn through use and the greatest value may emerge from process redesign rather than the first proposed feature. Months spent perfecting a forecast can make the original assumptions obsolete.
Do not replace discipline with enthusiasm
The answer is not to approve every experiment. It is to change what the investment decision purchases. An early decision should buy evidence, not promise a full transformation.
A small discovery can test whether useful data exists, whether users trust the output and whether the process has enough value to justify redesign.
Work with staged commitments
Divide investment into explicit gates:
- Strategic fit: does the opportunity reinforce the chosen position?
- Discovery: can value, feasibility and risk be tested cheaply?
- Controlled pilot: does the capability work with real users and limited permissions?
- Operationalization: can ownership, integration and governance support regular use?
- Scaling: does evidence justify a broader mandate?
Each gate has its own evidence and stop criteria. Ending a weak initiative early becomes a success of governance rather than a failed project.
Measure more than hours saved
AI may improve quality, consistency, speed of learning or access to expertise. It may also introduce review work and new risk. A useful case includes operational, human and governance indicators.
Examples include error rates, escalation, adoption, customer outcomes, human intervention and the cost of maintaining knowledge.
Treat options as value
A well designed pilot creates knowledge that informs several future choices. Shared infrastructure also reduces the marginal cost of later applications. These option values are real, even when they do not fit neatly into one project forecast.
Convene AI OS makes staged delivery practical by providing shared controls and reusable operational components.
A modern AI business case does not pretend uncertainty has disappeared. It turns uncertainty into a sequence of accountable learning decisions.