How do you track AI benefits after go-live? Baselines, owners and reviews
Most companies approve AI projects on a business case and never check it again. A practical routine for baselines, named owners and regular reviews turns promised benefits into measured ones.
KEY TAKEAWAYS Summary by the editors
- Benefits tracking starts before go-live: without a measured baseline for the process being changed, no later number can be attributed to the AI tool.
- McKinsey's 2026 survey found that 37% of respondents attribute at least some EBIT impact to AI, essentially unchanged from the year before, so measured impact still lags investment.
- Each benefit needs one named business owner, one metric, one baseline and one review date; the AI or IT team should not own the business outcome.
- Time saved only becomes a financial benefit if the freed hours are redeployed, absorbed into growth or removed from cost, which must be decided explicitly.
- Gartner's analyst commentary notes that ROI is harder to calculate when many people use a tool and workflows are not clearly defined.
Tracking AI benefits after go-live means comparing the live process with a baseline you measured before launch, assigning each benefit to a named business owner, and reviewing the numbers on a fixed calendar until the benefit is either confirmed or written off. The tool vendor's dashboard is not enough, because it reports usage, not business results. This guide sets out a routine a fashion company can run with a finance partner and no new software.
Why is benefit tracking harder than the original business case?
Business cases are written once, by enthusiasts, with a clear target. Real use is messy: people adopt a tool unevenly, workflows change around it, and other initiatives run in parallel. McKinsey's 2026 survey of 1,719 respondents found that 37% attribute at least some EBIT impact to AI, essentially unchanged from the prior year, while 80% say AI has improved their individual productivity. The gap between personal productivity and enterprise results is where most unmeasured benefit disappears.
The same report observes that conviction in AI is growing faster than the financial returns organisations can attribute to it. In its survey, nearly three-quarters of high performers had fundamentally redesigned workflows around AI, against about one-quarter of other respondents. Redesign is therefore part of the benefit, and tracking has to follow the workflow, not only the tool.
What baseline do you need before go-live?
A baseline is a measurement of today's process, taken over a representative period, using the same definition you will use after launch. In fashion, seasonality makes this critical: comparing a post-launch autumn with a pre-launch summer proves nothing. Where possible, compare like-for-like periods or keep a control group that does not use the tool.
- Volume: how many items, orders, briefs, tickets or forecasts the process handles per week.
- Effort: hours per unit of work, taken from timesheets, ticket systems or a short work-sampling exercise.
- Quality: error, rework, return or override rates for the same output.
- Cycle time: elapsed days from request to approved result.
- Cost: external spend (agencies, photographers, temporary staff) that the tool is meant to reduce.
If a baseline cannot be measured, say so in the business case and downgrade the benefit to a hypothesis. A benefit with no baseline should not appear in a budget line.
Who should own each benefit?
Every benefit needs one owner with a budget or target that the benefit affects: the head of e-commerce for conversion, the planning director for forecast accuracy, the customer service lead for handling time. The AI team owns delivery of the capability; the business owner owns the result. Shared ownership usually means no ownership.
Owners should also sign the baseline. If the person who will be judged on the number has agreed how it is measured, disputes at the first review are far less likely.
How should you structure the review cycle?
A simple rhythm works for most mid-size companies. Reviews should be short, use the same one-page format each time, and end with a decision: continue, change, scale or stop.
| Review point | Typical timing | Question to answer | Possible decision |
|---|---|---|---|
| Adoption check | Weeks 4 to 6 | Are the intended users actually using it, and for the intended task? | Retrain, adjust the workflow or fix access |
| First value check | Months 3 to 4 | Has the metric moved against the baseline, allowing for seasonality? | Continue, or narrow the scope |
| Financial confirmation | Months 6 to 9 | Has finance agreed the saving or revenue effect, and where did freed time go? | Book the benefit or remove it from the plan |
| Annual re-test | Each budget round | Is the benefit still there, and what does it cost to run? | Scale, renegotiate or retire |
How do you turn time saved into a financial benefit?
This is the step most programmes skip. A tool that saves each planner two hours a week creates a benefit only if the organisation decides what happens to those hours. Options include absorbing growth without hiring, reducing temporary or external spend, or using the time for higher-value work that has its own metric. Each is legitimate, but each must be named, and finance should agree which one is being claimed.
Be cautious about self-reported savings. Surveys of users capture perception, and perception is useful for adoption but weak evidence for a budget. Where the stakes justify it, sample actual task times before and after.
What costs belong in the benefit calculation?
Net benefit means gross benefit minus the full running cost. That includes licences or usage fees, internal support, data preparation, monitoring and retraining. McKinsey's survey found that about 20% of respondents say operating costs, including tokens, have constrained their AI use, so usage-based pricing deserves its own line in the tracker. Gartner's analysis also notes that ROI is harder to calculate when many people use a tool and workflows are not clearly defined, which is an argument for defining the workflow before launch.
What should you do when a benefit does not appear?
Treat a missing benefit as information. The common causes are low adoption, a workflow that never changed, poor data quality, or a baseline that was wrong. Diagnose in that order before blaming the technology. If the cause cannot be fixed within a quarter, stop and release the budget. A tracker that never records a failure is not being used honestly.
- Write the benefit as one sentence with a metric, a baseline value and a target.
- Name the business owner and have them sign the baseline method.
- Measure the baseline over a comparable period before go-live.
- Fix review dates in the calendar at approval, not after launch.
- Ask finance to confirm how any saving will be booked.
- Record decisions and stopped projects in the same tracker.
The discipline is modest, but it is what separates companies that can say what AI delivered from those that can only say how much they spent.
Frequently asked questions
How soon should you measure AI benefits after go-live?
Check adoption within the first six weeks and the first business metric after three to four months. Financial confirmation with finance usually needs six to nine months, and longer where seasonality affects the metric.
What if we have no baseline for a process?
Measure it now, even if the tool is already live, using a group that has not yet adopted it if possible. If that is not feasible, label any claimed benefit as an estimate and keep it out of budget commitments.
Who should own AI benefits, IT or the business?
The business function whose target is affected should own the benefit. IT or the AI team owns delivery and operation of the capability, but not the commercial result.
Is time saved a real benefit?
It is only a real financial benefit if the freed time is redeployed, avoids a hire or removes external cost. Otherwise it is capacity, which has value but should be reported as such.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
