Measure AI value against a defined business outcome and a comparable baseline. Include preparation, human review, corrections, support and recurring costs. Track quality alongside speed, distinguish released capacity from financial savings, and record uncertainty. Use the evidence and agreed criteria to decide whether to expand, revise, pause or stop an initiative.
A busy AI tool can still leave a business with the same delays, costs and unresolved decisions. Usage helps show whether people are engaging with a capability. Understanding its value requires a view of the finished work, the effort behind it and the outcome the organisation needs.
PwC's 2026 Digital Trends in Operations Survey examines gaps between technology investment and reported results. It surveyed 767 operations and supply-chain leaders at US organisations with annual revenue of at least $100 million. Those self-reported findings are context for asking better measurement questions; they are not an estimate of your organisation's likely return.
Start with a business outcome
Write down the result the initiative is intended to improve. “More employees use the assistant” describes participation. “Customer enquiries reach a complete, checked answer sooner” describes a change in performance that a business owner can assess.
Choose an outcome owner and define the relevant scope: which requests, users, locations or products are included? Agree what a successful result looks like and which conditions must remain acceptable. A faster process may still need to meet requirements for quality, access, review and customer service.
Establish a comparable baseline
Observe the current workflow before estimating improvement. Include preparation, waiting, checking, correction and follow-up. Record the mix of routine and difficult work, so a later comparison does not confuse easier cases with a better process.
Where possible, compare similar tasks over a suitable period and retain the definitions used. Record other changes, such as improved source records or additional team capacity. These may contribute to the result. Keep measured findings separate from assumptions, and explain what the evaluation cannot yet establish.
Count the effort around the output
The full operating picture includes licences or usage charges, integration support, source maintenance, administration and employee time. Human review deserves particular attention: checking facts, correcting a draft and handling exceptions all consume capacity.
Distinguish initial investment from recurring effort. Include the cost of operating a fallback when the system is unavailable, and the work required to evaluate changes. Use a consistent period when comparing costs and benefits. A short trial may reveal recurring tasks without providing enough evidence to estimate their annual volume confidently.
Keep quality beside speed
Agree how reviewers will assess completed work. Depending on the task, useful criteria might include factual completeness, appropriate escalation, missed information or the frequency of corrections. Review difficult cases as well as routine ones, and look for differences between user groups.
The NIST AI Risk Management Framework offers a voluntary reference for considering trustworthiness throughout AI design, use and evaluation. It supports a broader view than output speed alone. Your evaluation and review requirements should reflect the actual task and its consequences, with relevant specialists involved.
Consider a fictional procurement team using AI to prepare supplier summaries. Drafting becomes quicker, but reviewers spend longer resolving unsupported statements. The team would need to compare the total effort and final quality before recording a benefit. This example illustrates a measurement question, rather than a client result.
Separate capacity from financial savings
If a workflow saves employee time, establish what happens to that time. It may allow the team to clear a backlog, improve service or absorb additional demand. Those can be useful benefits when they are observed and connected to a business priority.
Turning minutes saved into a salary-based estimate does not establish a reduction in expenditure. Label that calculation as an estimated capacity value. Confirm financial benefits with the appropriate business or finance owner, and avoid counting the same improvement twice across several initiatives.
Use the evidence to allocate the next investment
Give each initiative a short review record: intended outcome, baseline, current findings, operating costs, unresolved concerns and recommendation. Include adoption evidence, since low use may reflect a poor workflow fit, unclear guidance or a capability that employees do not need.
- Fund or expand: the evidence supports a useful outcome, acceptable quality and a sustainable operating arrangement.
- Revise: a specific change to scope, information, workflow or support could address an identified weakness.
- Pause: a dependency or material uncertainty needs resolution before more investment is justified.
- Stop: the business case no longer holds, required controls cannot be met or there is no viable owner for the work.
Agree these criteria before the review, along with who decides and when. Across a portfolio, consider shared dependencies and competing demands on employees. An initiative with encouraging local results may still require a different sequence before wider adoption.
Authentia's AI strategy advisory helps clients connect priorities with a practical value case. Through ongoing advisory, we help leadership review evidence and recommend where to fund, adjust or stop. Your business owners and delivery teams remain responsible for implementation and reported results.