Your AI usage is up. Your revenue isn’t. Here’s how to measure enterprise AI ROI.
Let me guess.
You rolled out Copilot. Usage dashboards look great. Every month, more people are “using AI.”
And then your CFO walks into the board meeting and asks the one question nobody prepared for:
“Okay. So where’s the money?”
Silence.
This is the enterprise AI ROI problem, and almost every company has it right now. Our CEO, Derek Harrar, just wrote about it in Technology Record’s Autumn 2026 issue, and I want to break down the big ideas here, because this is the conversation every exec team needs to have before budget season.
Usage going up is not a success story
Here’s the uncomfortable truth.
If adoption climbs every month and revenue stays flat, you don’t have an AI win. You have a warning sign.
Even Microsoft gets it. This year it changed how it scores partners on Microsoft 365 Copilot: no more counting seats sold, now it’s paid monthly active usage. And instead of customer references, partners get an independent third-party audit.
Read that again. Seats became measured use. Claims became evidence.
If Microsoft is holding its own partners to that standard, your board is going to hold you to it too.
Why measuring AI value is so hard
We’re running old software playbooks on something that doesn’t behave like software.
Old software: you ship it, it stays shipped. The code doesn’t change on its own.
AI: models update, get swapped and get retired. A workflow that worked great six months ago can quietly get worse, and nobody touched it.
Most companies can’t even see that happening.
So the answer isn’t another dashboard. It’s a discipline. At Algoworks we call ours AI:CI, continuous improvement applied to AI itself.
Step 1: Measure leverage, not imaginary headcount
What is AI leverage? It’s how much faster a task gets done now compared to how long it took before. If a task took 10 hours and now takes 2, that’s 5x leverage.
Simple. Countable. Defensible.
Compare that to the AI ROI math most companies show the board: “Without AI, we’d have needed twice the headcount.”
Nobody can prove that number. Your board will tear it apart in about four seconds.
Leverage compares two things you can actually count: before and after. That’s a number that survives a CFO.
Step 2: Find your friction (this is where the real ROI hides)
Here’s the part most people miss.
Leverage doesn’t keep climbing just because the model gets smarter. It hits a ceiling, and that ceiling is friction: the prep work someone does before the AI runs, plus the cleanup after.
The math is brutal:
- A 5% prep-and-rework tax caps your return at 20x.
- A 10% tax caps it at 10x.
No matter how good the next model is.
So stop chasing model releases. The ROI is in cutting the manual cleanup.
Step 3: Agents raise the stakes
Everyone is excited about AI agents. Me too.
But agents make measurement harder, not easier. With one prompt, friction is cleaning up one output. With a chain of agent steps, friction is verifying a whole sequence.
And the key question becomes: which numbers in that chain were actually measured, and which were guessed?
That whole agent governance debate happening right now? It’s the measurement conversation, just bigger.
Step 4: Put token spend next to the results
Token spend is what you actually get billed to run AI. It needs its own line of scrutiny.
Because a workflow can look like a win (people are getting more done) while quietly costing more in AI fees than it’s worth.
That’s AI that works but doesn’t pay.
Remember the cloud wave? Everyone moved to cloud to save money. About 18 months later, finance wanted to know why compute was growing faster than the business.
AI is at that exact moment. Except it compounds faster. Get a handle on token spend now, or have the “what happened to our margins?” talk next year with a much bigger number.
The 5 questions to bring to your next board meeting
- What’s stuck?
- Why is it stuck?
- What’s it worth to fix?
- What proof do we have, from delivered work and not usage stats, that we’ll get that value back?
- Which of our numbers are proven, and which are still estimates?
Two more mindset shifts:
- Unused AI capacity you’ve already paid for is the real waste. The fix is pushing more work through it, not shrinking the commitment.
- Proof unlocks budget. It’s what gets you the next round of investment, not what kills it.
And be clear about the work that should stay human: architecture, judgment calls, sign-off. Skip that and your own team will quietly resist the whole measurement program.
The bottom line
AI value doesn’t show up because you deployed a tool.
It shows up when people get better at using it, when that skill turns into work that ships and holds up, and when you can measure it against what it cost in time and tokens.
Until you connect those dots, your AI story and your financial story will keep telling two different versions of the same spend.
Read Derek’s full piece in Technology Record, Issue 42.
And if your board is asking where the AI value is and you don’t have a confident answer yet, talk to us. We run this discipline on our own delivery first, so we’ll show you the numbers, not just the pitch.
FAQs
How do you measure enterprise AI ROI?
Compare how long a task took before AI with how long it takes now (leverage), subtract the prep and cleanup time (friction), and weigh the result against token spend.
What limits AI ROI?
Friction. A 10% prep-and-rework tax caps your realized return at 10x, regardless of model quality.
Why track token spend separately?
A workflow can boost output while costing more in AI fees than the value it creates. You need both numbers side by side.
What is AI:CI?
Algoworks’ measurement discipline: continuous improvement applied to AI workflows, so you can track performance as models change.
P.S. If you only remember one thing: usage is not ROI. Leverage minus friction, measured against cost, is.
