Source-led executive learning · reviewed August 2026

What must change in the work for AI activity to become value?

A faster task is not yet a business result. Value appears only when released capacity, improved quality or new capability changes a workflow outcome that someone owns.

1. Begin with friction in a priority workflow

Tool-first proposals start with a product and search for places to install it. Decision-quality proposals begin with a workflow that matters: resolving a customer enquiry, preparing a proposal, replenishing stock or reviewing a control exception. Name the outcome, the people who receive it, the current steps and the friction that constrains performance. Only then ask whether an AI-enabled change is a proportionate intervention.

A task is one unit of work: draft a response, classify a message, extract a contract date. A workflow is the connected sequence that turns several tasks and hand-offs into an outcome. Improving a task may leave the workflow unchanged if the bottleneck sits elsewhere. A drafting assistant that produces copy in five minutes creates no customer value if legal review still waits ten days, the draft needs extensive correction, or salespeople use the released time for low-priority administration.

The first discipline is therefore causal: describe how a change in the task is expected to change the end-to-end workflow. Avoid the phrase “AI use case” until you can complete this sentence: For this group, changing this task and these surrounding practices should improve this outcome because…

2. Choose the value mechanism

Most business hypotheses use one or more of four mechanisms. Growth or revenue means the intervention helps win, retain or expand valuable demand. Cost or capacity means it reduces resources required for an outcome or releases capacity that can be deliberately redeployed. Risk or quality means it reduces material errors, losses, rework or variation. Experience or service means it improves timeliness, accessibility, relevance or effort for customers, employees or other stakeholders.

These mechanisms are not interchangeable. If the hypothesis is better service, minutes saved is at most an intermediate signal. If it is risk reduction, the right support may concern severe exceptions rather than average speed. If it is capacity, management must say what happens to the released hours. If it is revenue, adoption metrics do not prove conversion, retention or contribution margin.

Write one primary mechanism and, if necessary, one secondary mechanism. Listing every possible benefit weakens the hypothesis because no support could disconfirm it. A useful value statement exposes the causal bet: “By reducing the time advisers spend locating approved policy passages, we expect more same-day responses without reducing source accuracy.”

3. Released time is capacity inventory, not cash

Suppose 20 employees each complete ten reports per month. Drafting falls from 60 minutes to 40 minutes. The arithmetic suggests 20 people × 10 reports × 20 minutes = 4,000 minutes, or about 66.7 hours released per month. That number is not automatically a saving.

To become cash, paid hours or an external expense must actually fall without shifting cost or risk elsewhere. To become capacity value, the organisation must redirect those hours to named work, and that work must improve a measurable outcome. To become service value, the end-to-end cycle time or quality experienced by the recipient must improve. Small fragments of time may be difficult to redeploy; review work may increase; demand may vary; and adoption may be uneven. Record these leakages rather than treating every minute as fully fungible.

A credible capacity hypothesis therefore includes volume, time difference, adoption, quality and a redeployment decision. For example: “At 70 per cent sustained adoption, we estimate 47 gross hours released monthly. The proposal manager will redirect up to 30 usable hours to qualified customer discovery. We will track accepted draft quality and discovery meetings; we will not book a financial saving during the experiment.”

4. Complementary change is part of the intervention

An AI system rarely creates value alone. The intervention may require an approved information source, a revised intake form, new quality criteria, role clarity, exception handling, training, incentives or a changed approval threshold. These are not implementation footnotes. They are causal components of the value pathway.

A peer-reviewed field study of a generative AI assistant in customer support found that effects varied across workers in that setting [C005-S07]. Its population, task, treatment and operating context do not establish the same result elsewhere. Use published support to shape a local hypothesis, not to pre-fill its benefit number.

Ask three questions. First, what behaviour must the user change? Second, what must the manager or process owner change? Third, what dependency outside the team must change? If none of these changes is visible, the proposal probably describes product activity rather than operating value.

5. Measure the chain, not only the endpoint

A useful measurement set reveals where the pathway succeeds or leaks. Activity measures whether the system ran: drafts created or messages classified. Adoption measures whether intended users use it appropriately and persistently. Quality measures whether outputs meet the criteria needed for the workflow. Operational outcome measures the end-to-end result: cycle time, first-contact resolution, conversion, rework or severe exceptions. Value connects that outcome to an organisational consequence, with an owner who can validate the method.

Do not collect dozens of metrics. Select one or two at each material link and define them before the test. A baseline and comparator matter because a change over time can reflect seasonality, demand mix, staffing or another initiative. Segmentation matters because an average can conceal a team that receives no benefit or a category where quality fails.

6. Worked case: proposal drafting assistant

What changed?

The technology did not change. The workflow, quality standard, review service level, redeployment decision and benefit ownership changed. These additions turn a task-speed observation into a testable value pathway.

7. Contrast case: every saved minute becomes cash

8. Practice: build the pathway

9. Decision summary

A task improvement becomes organisational value only through a chain of complementary changes and decisions. Make the chain visible. Distinguish gross capacity from usable capacity and cash. Measure adoption and quality so you can diagnose failure. Give the benefit to an owner with authority to redirect work, change the workflow or stop the intervention.

Review the annotated exemplar Open the manager discussion guide