The Value Lens: Choosing the Right Problems for AI-Era Transformation

Define the outcome before narrowing the solution

Start with what the enterprise needs to become better able to do. That might mean paying suppliers reliably, resolving customer problems with fewer repeat contacts, responding to disruption, or serving customers whose needs have previously been uneconomic to meet.

Describe the beneficiary, the current limitation, and the consequence of leaving it unresolved. “Reduce processing time” becomes meaningful when leaders can explain whether the delay affects service, capacity, revenue, or a critical obligation. Establish a baseline where possible; where it is missing, make discovery responsible for finding it.

The original Value Lens test remains useful: would this problem still deserve attention if AI were unavailable? Apply it to the desirability of the outcome. AI may change the economics enough to make a previously impractical service feasible. That possibility still needs evidence of demand and a credible way to deliver it.

Keep alternatives open while diagnosing the constraint. NIST’s AI Risk Management Framework Playbook recommends considering non-AI and non-technology approaches when establishing the context for a proposed system. NIST, MAP 1.1 guidance.

The comparison should include a simpler process, clearer authority, improved information, conventional automation, and AI where it adds something material.

Use the five lenses to understand the opportunity

The lenses reveal different things about the work. Strategic relevance establishes why an outcome matters. The others help explain the constraint, the appropriate response, and the dependencies that could determine its success.

Strategic relevance

Connect the outcome to a specific enterprise priority and an accountable leader. Volume alone is a poor guide: an infrequent failure can have serious consequences, while a frequent inconvenience may have little strategic significance. Ask how much the outcome matters relative to other demands on the same budget and people. A required obligation also needs an appropriate response, even when its financial return is difficult to express.

Operational friction

Trace the work far enough to locate what causes delay, rework, or uncertainty. A queue shows where work accumulates; it does not establish why. The cause may be missing information, conflicting policy, limited capacity, or an approval no one has authority to simplify. Involve the people doing the work: their compensating actions often reveal dependencies the process map omits.

Decision intensity

Examine which judgments influence the outcome and what makes them difficult. Stable rules may support conventional automation. Interpreting varied information may justify AI assistance. Consequential ambiguity may require human judgment supported by better evidence. Complexity increases the need to investigate; it does not establish a case for autonomous action. The proposed role for AI should follow the decision’s consequences and the ability to detect and correct mistakes.

Knowledge dependency

Identify the information people need, its authority, and who maintains it. Heavy dependence on contracts, policies, or prior exceptions can reveal an opportunity for better assistance. It can also expose a substantial prerequisite. A system drawing on conflicting guidance may distribute the inconsistency faster. High knowledge dependency is a reason to inspect the foundation carefully, rather than award the opportunity extra points automatically.

Scalability and repeatability

Ask what another operation could actually reuse: an intake pattern, an approved knowledge source, a decision control, or a measurement routine. Identify a plausible next user and the work required to adapt it. Reuse strengthens an opportunity when it reduces future effort or improves consistency. It should not become a speculative multiplier that makes a weak initial outcome look attractive.

Test whether the benefit reaches the enterprise

A proposal should explain how a local improvement becomes a better operating result. Time released is one input to that explanation. Its value depends on what people can do with it, whether demand exists, and whether the improvement survives checking, correction, and downstream handoffs.

A study published in February 2025 examined a 2020–21 rollout of AI assistance to customer-support workers. Across 5,172 agents, the researchers found an average 15% increase in issues resolved per hour, with substantially different effects across workers. Brynjolfsson, Li, and Raymond, Generative AI at Work.

My implication for selection is to examine the particular work and workforce behind a projected gain. An average from one setting cannot establish the return from another. Nor does faster task completion, by itself, demonstrate cash savings.

Ask who receives the benefit and who takes on new responsibilities. A drafting assistant may reduce preparation effort while adding verification work for scarce specialists. An intake change may help an internal team while making submission harder for customers or suppliers. Those effects belong in the same assessment.

Distinguish released capacity, improved service, risk reduction, and actual spending reductions. Include implementation, integration, training, knowledge maintenance, supervision, and running costs at a level proportionate to the decision. Also ask whether the proposed human role is sustainable: a smaller queue of harder exceptions may require more judgment and support per case.

Keep value and readiness separate

An additive score can conceal a decision that needs explicit judgment. Strong projected value and reuse should not compensate for an unresolved consequence the organization cannot govern. Equally, weak current readiness should not automatically eliminate an important problem with a realistic path to improvement.

I would make the next action explicit. Where the outcome matters and the conditions support a responsible test, fund bounded discovery. Where the outcome matters but a specific prerequisite is missing, fund that prerequisite or the investigation needed to understand it. If the prerequisite costs more than the prospective outcome justifies, reconsider the intervention. Name its owner, scope, cost limit, review date, and the evidence that would justify further commitment.

If a simpler intervention addresses the constraint, route the work there. If the outcome is weak, another opportunity is more consequential, or no acceptable path exists, defer or decline the proposal. Where knowledge, authority, or oversight cannot support the proposed action, narrow the action or pause it while the conditions change.

Learning can justify an exploratory investment when it resolves a material uncertainty. Specify which decision the experiment will inform and what finding would stop further work. “Learning about AI” is too broad to establish that discipline.

Finally, compare demands across the portfolio. Several attractive initiatives may depend on the same specialists, knowledge owners, or operational teams. A shared prerequisite may deserve investment before another pilot. Make that connection explicit, and count each resulting benefit once.

A hypothetical example: choosing where to act on invoice exceptions

Consider a hypothetical finance operation seeking more reliable supplier payments and less avoidable exception work. Its initial proposal is an AI assistant that prepares case summaries for analysts.

Assume a sample of delayed invoices reveals two recurring constraints: missing purchase-order references at intake, and disputed cases requiring contract interpretation by a small group of specialists. These are scenario assumptions, not findings from an actual organization.

The Value Lens changes the comparison:

Under these assumptions, I would prioritize the intake change and a limited investigation of the knowledge gap. A general summary assistant would need to show that preparation effort materially constrains the outcome before receiving a larger commitment.

Procurement and finance would need to agree who corrects incomplete submissions. Specialists would help define authoritative guidance and retain responsibility for disputed interpretations. This hypothetical design grants the assistant no authority to change payment details or release funds.

The proposed benefits are fewer preventable exceptions, more reliable resolution, and payments made according to agreed terms. None is an achieved result. Validation would compare equivalent invoice categories against a documented baseline, tracking exception recurrence, elapsed resolution time, total handling and review effort, payment errors, and relevant costs.

If intake improves but payment reliability does not, the team should investigate the next constraint before expanding AI use. The outcome should govern the next choice.

The Value Lens belongs in Frame the Opportunity, the first phase of the Transformation Operating Map. Once the problem and next commitment are justified, the operating case can develop the delivery economics, responsibilities, controls, and conditions for sustained use.