Boards increasingly ask what an artificial-intelligence investment produced, not how impressive the demonstration looked. Yet many smaller companies still begin with a tool or a broad instruction to “try AI,” leaving success undefined until after money and employee time have already been spent.
Pete Smyth, chief executive of Leading Resolutions, argues that AI should be treated as a structured business programme rather than an exploratory technology exercise. The project begins with the outcome and a way to measure it, then works backward toward the model, data, process, and controls required.
Return on investment is therefore not a calculation performed at the end. It is a design decision made before a pilot receives approval.
Start With a Business Constraint, Not a Model
Useful candidates include automating routine analysis, predicting inventory demand, or streamlining customer service. Each connects to an operational constraint: excessive manual effort, working capital tied up in stock, or slow responses that weaken retention.
Stakeholders across departments can identify opportunities, but the list should be filtered by potential value and readiness. A promising idea may still be unsuitable if the necessary information is unavailable, the process has no clear owner, or the organisation cannot tolerate the risk of an incorrect result.
This assessment creates a shortlist for pilots rather than a collection of experiments competing indefinitely for attention. Technology choice follows the business case instead of defining it.
It also reveals dependencies before they become expensive surprises. An inventory forecast, for example, may look like a modelling project but depend on inconsistent product codes, delayed supplier updates, and local spreadsheets. Fixing those inputs may create value even before AI is introduced, while ignoring them makes an impressive prototype difficult to operate.
Agree on the Evidence Before Building
A structured value assessment combines benefits, implementation cost, feasibility, and risk tolerance. Leaders should approve the key performance indicators before the system is deployed: cost reduction, decision speed, productivity, margin, customer retention, service quality, or another observable measure.
The baseline matters. A company cannot credibly claim that AI shortened a workflow if it never measured the previous duration. Costs must also include integration, data preparation, oversight, training, and ongoing operation rather than only licence or API fees.
Some value appears as risk reduction or resilience rather than immediate revenue. Those outcomes can still be measured through fewer incidents, improved forecast accuracy, shorter recovery time, or reduced exposure to a known operational failure.
Measurement should compare like with like. A faster customer-service response means little if resolution quality falls or employees spend the saved time correcting generated answers. A useful scorecard pairs the headline benefit with guardrails, so efficiency cannot be purchased through a hidden decline elsewhere in the process.
Scale Only After the Outcome Survives the Pilot
A pilot should test both value and the conditions needed to maintain it. Governance, explainability, and risk controls belong in the early design because a result produced without acceptable oversight cannot be scaled responsibly.
Once the agreed metrics show improvement, deployment can expand through selected business units. The organisation should continue checking whether the gain persists as user numbers, data variety, and workflow complexity increase.
Smyth emphasises a culture grounded in information quality, collaboration, and evidence-based decisions. AI ROI is rarely delivered by the model alone; employees must adapt the surrounding process and trust the measurement enough to act on it.
The discipline also creates a stopping rule. If a pilot does not meet its predefined threshold, leaders can end or redesign it without inventing a softer definition of success.
Credible AI investment is not determined by budget size. It is determined by whether the organisation can explain the target, measure the baseline, account for full cost, and demonstrate a result worth scaling. The strongest ROI story begins before anyone writes the first prompt.
