By James Proctor, Co-Founder and Managing Director, The Inteq Group
When AI initiatives deliver disappointing returns, the technology is rarely the reason. The disappointment was designed in months earlier, when the initiative was scoped, ranked, and justified as task automation: a business case denominated in hours saved. An hours-saved case caps returns at the labor content of the tasks it touches, which in most processes is a modest number, and the program then dutifully delivers to its cap. The program did not underdeliver. The business case underasked.
“The program did not underdeliver. The business case underasked.”
Why Do Hours-Saved Business Cases Cap AI Returns?
An hours-saved case can only see one kind of value: labor displaced from tasks. It is structurally blind to everything that happens when a process makes its decisions sooner and better, which is where agent-enabled redesign earns its money: cycle time compressing, rework declining because decisions are made with full context, quality produced in the flow rather than inspected at the end, and expert capacity redirected from routine confirmation to consequential judgment. These effects compound each other, and none of them appears in a template whose only line is minutes multiplied by volume. Scope to the template and you have pre-committed to marginal outcomes with perfect precision.
What Should the Business Case Measure Instead?
Decision-value cases are built on different questions. What does a day of cycle time cost us in this process, in working capital, penalties, attrition, or lost throughput? What does rework actually consume? What would our best people do with the hours they currently spend confirming the obvious? These numbers require more work to establish than a time-and-motion study, and they are the numbers in which transformation-scale returns are denominated. Organizations that cannot be bothered to establish them should stop being puzzled by automation-scale results.
There is a portfolio effect as well. Rank candidate initiatives by hours saved and the list fills with high-volume clerical work. Rank the same candidates by decision value and the list reorders around the processes where latency, rework, and expert capacity carry real money. The two rankings rarely agree, and the gap between them is a reasonable estimate of what the template has been costing you.
What Does the Difference Look Like in Practice?
Consumer packaged goods deduction management shows both programs side by side. Retail customers short-pay invoices continuously, claiming promotions, allowances, shortages, and compliance penalties. The automation program does what automation programs do: it captures claims, matches documentation, and codes deductions, and it genuinely saves analyst hours. The write-off line does not move, because the money was never in the clerical work. It was in the validity decision, which still waits weeks while analysts assemble promotion terms, shipment records, and deal sheets, and while validity teams triage by dollar size instead of by recoverability. Redesigned around that decision, with agents assembling the evidence case and resolving clear dispositions immediately, invalid deductions get challenged inside the customer's own dispute windows, recovery rates move, and the value shows up on a line the CFO actually watches. Same function, same data, different design target, different order of result.
Who Actually Owns This Problem?
Now the part that will not make me popular in finance: in many enterprises, the binding constraint on AI returns is the CFO's business case template. Hard FTE savings are auditable, so the template privileges them, so sponsors scope to them, so the portfolio fills with precisely the initiatives least capable of changing the operation. The measurement system is selecting for marginal projects and then presenting the marginal results as evidence that AI is overhyped. It is a self-sealing loop, and only finance can break it, by admitting decision-value lines into the template with the same seriousness it grants labor lines.
Constructing credible decision-value cases, with defensible baselines and measurable outcome lines, is a discipline we bring inside our agentic AI consulting engagements, because the case is where the ceiling gets set, long before any technology is chosen.
