On this page
- Key Takeaways
- Change Management Gets Cut First Because It Reports Activity
- The Productivity Dip Is a Cost You Can Measure
- Time to Value Is the Metric That Pays for the Program
- Hours Saved Mean Nothing Until You Say Where They Went
- Five Metrics That Belong in the CFO Report
- Adoption Numbers Belong Under the Line
- Measurement Is Part of the Change
- Report the Change in the CFO's Units
- FAQ
Key Takeaways
- Change management gets cut first because it reports activity. Sessions held and emails sent tell a CFO what you spent, not what came back.
- The productivity dip is a real cost. Research on AI adoption in manufacturing finds output falls before it rises. Change management is how you make that dip shallower and shorter.
- Time to value is the metric that funds the program. Every month a rollout sits below baseline is a month of cost with no return.
- Hours saved mean nothing until you say where they went. Capacity redeployed is the number that shows up in a budget.
- Adoption numbers are leading indicators. Keep them, report them, but put them under the financial line, not in place of it.
Here's the line item that gets cut first when an AI budget gets tight: change management. The model license stays. The integration work stays. The training budget, the champion program and the rollout plan go.
That's strange when you look at where AI projects actually fail. BCG's 2024 research on AI adoption found that 74% of companies had yet to show tangible value from AI, and that about 70% of the challenges came from people and process. Only 10% came from the algorithms. So the budget line aimed at 70% of the problem is the one finance trusts least.
I don't blame the CFO for that. I blame the reporting. Change teams tend to report what they did: 12 training sessions, 40 champions, three all-hands demos, a satisfaction score of 4.5. Those are honest numbers. They are also numbers a finance leader can't do anything with. None of them say whether the business got its money back, or when.
The way I've come to see it: change management is a cost you take on to shrink a different, bigger cost. That bigger cost is the time your people spend being slower, confused or working around the new tool before it pays off. If you report the change program in those terms, the CFO stops seeing a soft expense and starts seeing a lever.
The rest of this post is about which numbers make that case.
Change Management Gets Cut First Because It Reports Activity
Activity metrics are easy to collect, so they are what most change plans track. They also describe the input side of the ledger only. A finance leader reads "14 training sessions delivered" as "we paid for 14 training sessions."
BCG's 2025 survey of 280+ finance executives shows how wide that gap is. Only 45% said they could quantify ROI from their AI initiatives, and the median reported ROI was 10%. When half of finance can't measure the return on the AI itself, a change program that reports headcounts and session counts has no chance.
Keep counting activity. You need it to run the program. Just stop leading with it.
Lead with the outcome, then explain it with activity
Flip the order of your report. Start with the financial result, or the best estimate you have of it. Then use adoption and activity numbers to explain why the result is where it is. "Claims processing time is back to baseline in three of five teams; the two lagging teams have the lowest weekly usage and haven't had role-based training yet." That's a sentence a CFO can act on.
The Productivity Dip Is a Cost You Can Measure
Every serious AI rollout makes people slower before it makes them faster. They're learning a new tool, unlearning an old habit, and checking the output twice because they don't trust it yet.
There's good research on this. A study covered by MIT Sloan, using U.S. Census Bureau data on tens of thousands of manufacturers, found that adopting AI was associated with a 1.33 percentage point drop in productivity in the short run, and a much larger drop once the researchers corrected for which firms chose to adopt. Over the longer run, adopters pulled ahead. (Manufacturing isn't every business, and these were industrial AI tools, not chat assistants. I still think the shape holds.)
The detail that stuck with me: older firms saw their use of structured management practices decline after adopting AI, and that decline alone explained nearly a third of their productivity losses. That's a change management finding. A big share of the loss came from how the organization absorbed the tool.

Baseline the work before launch day
You can't measure a dip without a baseline. Before a team goes live, capture two or three numbers that describe its normal output: tickets closed per person per week, days to close the books, contracts reviewed per analyst. Then track the same numbers weekly through the rollout. The gap between the line and the baseline, multiplied by the loaded cost of the people in it, is the cost of the change. Now you have something to shrink.
Time to Value Is the Metric That Pays for the Program
If the dip is the cost, time to value is the clock on it. I define it plainly: the number of weeks from go-live until a team's output is back above its pre-launch baseline and stays there.
This is the number I'd put at the top of every change report. It's simple, it's in weeks, and it converts straight into money. Say you shave two months off a 200-person rollout. That's two months of payroll that stopped underperforming sooner.
It's also where change management earns its budget. Prosci's research, drawn from more than 2,600 change practitioners, found projects with excellent change management were about seven times more likely to meet objectives than projects with poor change management. That data covers change projects in general and comes from practitioners grading their own work, so treat the multiplier with care. The direction matches everything I've seen with AI rollouts.
Put a date on the business case
Most AI business cases include a benefit number and no date. Add the date. "We expect underwriting cycle time to drop 20% by week 12" gives you something to manage against. If week 12 arrives and you're at week-6 numbers, you know to look at adoption first, and you know which teams to look at.
Hours Saved Mean Nothing Until You Say Where They Went
This is the metric that breaks most AI ROI decks. "The tool saves each analyst five hours a week" sounds great. Multiply it by headcount and salary and you get a big number. Then the CFO asks the obvious question: so where's the money?
Usually there isn't any. The five hours turned into slightly longer lunches, more meetings, or a bit more polish on work that was already fine. Time freed up isn't value until someone decides what it's for.
That decision is a change management job. Someone has to redesign the role, move work into the freed time, and make sure the old process actually stops. Otherwise you pay for the AI and keep paying for the old way of working.
Ask "redeployed to what?"
For every hour-saved claim, write down where the hours go. More accounts per rep. A backlog cleared. A hire you no longer need to make next quarter. Contractors you can roll off. Track capacity redeployed as its own line, with an owner, and report it next to hours saved. The gap between the two is a pretty good measure of how much change work is left.
Five Metrics That Belong in the CFO Report
Here's the framework I'd use. Five numbers, each tied to money or time, each with a leading indicator underneath it that explains movement.
| Metric | What it answers | How to measure it | Leading indicator to watch |
|---|---|---|---|
| Time to value | When did this team start paying back? | Weeks from go-live to sustained output above baseline | Weekly active use by role |
| Cost of the dip | What did the transition cost us? | Output gap vs. baseline × loaded labor cost, summed over the dip | Support tickets and repeat questions |
| Capacity redeployed | Where did the saved time go? | Hours moved to named work, hires avoided, contractors rolled off | Workflows redesigned and old steps retired |
| Benefit realized vs. plan | Are we on track to the business case? | Actual vs. planned benefit at each dated milestone | Teams trained by role, champion coverage |
| Cost per unit of work | Is the work actually cheaper? | Total cost of a task (people + AI spend) ÷ tasks completed | Usage per task, rework rate |
Notice what isn't on the top line: training completion, logins, satisfaction scores. They're all still useful. They just live one level down.

Give every metric an owner and a review date
A metric with no owner is a dashboard tile. Assign each of the five to a named person, usually a business lead and not the change team, and review them monthly with finance in the room. The change team's job is to explain the movement and propose the next intervention.
Adoption Numbers Belong Under the Line
I want to be careful here, because adoption data matters a lot. On our own AI change management work we measure adoption with usage telemetry and team feedback, and adjust the rollout before momentum fades. When PressW launched an AI search assistant for Education Delta, 700 pilot users asked 1,600 questions in the first two weeks. That's a strong early signal, and it told us where to go next.
But a strong early signal is still a signal. It tells you the financial result is likely coming. It doesn't tell you the result arrived. The mistake I see is teams presenting adoption as the outcome, because it's the best number they have.
Use adoption to explain, not to prove
Treat usage as the "why" behind the financial line. If time to value is slipping in one department, adoption data tells you whether people aren't using the tool, are using it badly, or are using it for the wrong tasks. Each of those has a different fix. That's what adoption data is for.
Measurement Is Part of the Change
Most of what decides an AI rollout comes down to how people work. And here's the part I find interesting: the act of measuring well changes how people behave.
When a team lead sees their own team's time to value next to a peer team's, they ask different questions. When a business owner has to name where saved hours went, they actually go redesign the work. BCG's finance survey found that systematic tracking and change management both ranked among the ten most effective tactics for getting ROI from AI. I don't think that's a coincidence. They're the same muscle.
Start measuring before the first user logs in
Set the baseline, the dated business case and the five metrics before go-live. Retrofitting them three months later is close to impossible, because you no longer know what normal looked like.
Report the Change in the CFO's Units
I'm convinced the companies that get real money out of AI will be the ones that run change management as a financial discipline, with baselines, dates and owners.
I'll be honest about what's unsolved. Measuring the productivity dip is easy for work that's already counted, like tickets or invoices. It's much harder for knowledge work where output is fuzzy. I don't have a clean answer for a strategy team or a legal team yet. The best I've found is to pick a proxy everyone agrees on before launch and live with its flaws.
If you're about to roll out AI and your plan doesn't have a baseline and a date on it, start there. That's the core of how we run AI Change Management at PressW, paired with role-based AI Training and Token & Usage Analytics for the usage data underneath. If you want to talk through what your CFO is asking for, Book a Discovery Call.
FAQ
The ones that convert to money or time: time to value, cost of the productivity dip, capacity redeployed, benefit realized against the business case, and cost per unit of work. Adoption and training metrics still matter as leading indicators that explain why those numbers move.
Compare the cost of the change program with the cost it removes, mostly a shorter and shallower productivity dip and faster time to value. That requires a baseline captured before launch and output tracked weekly through the rollout.
Time to value is the number of weeks from go-live until a team's output is back above its pre-launch baseline and stays there. It's the clearest single number for showing whether AI change management is working.
Usually because the saved time was never redeployed. Hours saved only become financial value when someone moves that time into named work, avoids a hire, or retires an old process. Tracking capacity redeployed alongside hours saved closes the gap.
No. Weekly active use, usage by role and proficiency are good leading indicators. They work best when reported under a financial metric to explain its movement, rather than presented as the result.
Before go-live. You need a baseline of normal output, a dated business case, and owners for each metric in place before the first user logs in. Without a baseline, you can't measure the dip or prove time to value.


