
November 17, 2026
How to Measure AI ROI: The One-Page Scorecard to Bring to Your Board
Most companies measure AI in hours saved. Boards want dollars. Here's the one-page scorecard that shows what each AI project cost, what it returned, and whether to scale it, fix it, or stop it.
Sometime in the next few months, someone on your board is going to ask a simple question. We've been spending on AI for two years. What did we get?
Most CEOs will answer with activity. The number of people trained. The number of tools rolled out. The hours employees say they're saving. All of that is true, and none of it answers the question. The board asked about money.
This guide shows you how to answer in dollars. It covers what the latest research says about who is getting a return from AI, why the most popular measure misleads, and a one-page scorecard you can fill in before your next board meeting.
How Do You Measure AI ROI?
To measure AI ROI, add up the financial return an AI project has produced, subtract everything it cost, and divide the result by the cost. A project that cost $80,000 and returned $200,000 in new gross profit and avoided spending has an ROI of 150%.
The formula is easy. The hard part is deciding what counts as return and making sure every cost is in the total. Most of this guide is about those two decisions.
What the Research Says About AI Returns
You've probably seen the headline that 95% of companies get zero return from generative AI. It comes from an MIT study published in 2025, and it deserves context. The researchers labeled the findings preliminary. The figure rests on 52 interviews, 153 survey responses, and a review of about 300 public projects, measured six months after each pilot. The 95% covers custom AI pilots that never reached production with a measurable effect on profit. It doesn't describe all AI use.
Larger and more recent surveys give a clearer picture, and they agree with each other.
- About half of companies see no financial benefit yet. PwC found that 56% of CEOs report no significant financial benefit, 33% report gains in either cost or revenue, and 12% report both.
- Individual gains aren't reaching the income statement. In McKinsey's 2026 State of AI survey, eight in ten respondents said AI improved their own productivity. Only 37% said it contributed anything to their company's earnings, a share that didn't move from the year before.
- The return is concentrated. Boston Consulting Group sorted 1,250 companies into three groups. Five percent are getting value at scale. Sixty percent report minimal revenue and cost gains.
- Few companies check. KPMG found that 61% of companies review AI costs at approval, and only 12% consistently compare value against cost afterward.
The pressure to produce a number is real. In BCG's survey of 640 CEOs, companies planned to double AI spending in 2026 to about 1.7% of revenue, and half of the chief executives said they believe their job is on the line if AI doesn't pay off. Boards aren't well equipped to judge it either. When Stanford researchers interviewed 20 corporate directors this fall, 80% said their boards had no formal AI policies or scorecards.
A CEO who walks in with a clear page of numbers will be ahead of nearly everyone.
Why "Hours Saved" Is the Wrong Number
Hours saved is the measure most companies reach for first. Wharton's 2026 enterprise AI study found it's the most commonly tracked AI metric. It's also the reason so many AI reports fall flat in a boardroom. There are three problems with it.
Saved time rarely turns into money on its own. If a salaried employee saves five hours a week and payroll stays the same, the company's costs haven't changed. The return only exists if those hours went to something that produces revenue, or if they let you skip a hire.
A lot of the savings get spent on cleanup. Workday surveyed 3,200 employees who use AI regularly. Nearly 40% of the time AI saved was lost to rework: correcting errors, rewriting output, and checking results. Only 14% of employees consistently came out ahead. Researchers at BetterUp and Stanford gave the problem a name, workslop: AI-generated work that looks finished and isn't. They estimate it costs $186 per employee per month.
People are poor judges of their own speed. In a controlled trial by the research group METR in 2025, experienced software developers took 19% longer to finish tasks with AI tools. Afterward, the same developers believed AI had made them 20% faster. The tools have improved since then, and METR now describes those results as out of date. The lesson about self-reported time savings still holds. Measure output. Don't survey feelings.
The same Wharton study shows what to measure instead. Among leaders who track at least one growth measure, such as new revenue, 35% report strongly positive returns from AI. Among those who track efficiency alone, it's 13%.
The Four Kinds of AI Return
Every legitimate AI return fits in one of four columns. If a claimed benefit doesn't fit, leave it off the page.
What counts as return
Revenue gained
New gross profit that exists because of the project. Faster lead response that raised the close rate. More quotes sent per week. A chatbot that books appointments after hours. Measure it against the period before the project started.
Cost removed
Money that stopped leaving the company. An outside vendor you no longer pay. Software you canceled. Overtime that went away. A role you planned to hire and didn't need to. If payroll and vendor spend didn't change, nothing was removed.
Capacity redeployed
Hours saved that you can trace to named work. Three estimators now bid 30% more jobs. Two support reps moved to account management. Count it only when you can say where the hours went and what they produced.
Risk and quality
Fewer errors, faster compliance checks, shorter customer wait times. Put a dollar value on these only when one exists, such as refunds avoided. Otherwise report the operating number itself.
The MIT study found something useful here. About half of generative AI budgets go to sales and marketing, where results are easy to see. The biggest measured savings often came from the back office: $2 million to $10 million a year from eliminating outsourced processing, and a 30% cut in outside agency spend. Those gains came from spending less with vendors, and the workforce stayed about the same size.
Count Every Cost
Return is half the equation. The cost side is where most AI math goes soft, because the software license is the only number anyone writes down.
A full count includes:
- Software and usage. Licenses, plus usage-based charges that grow with volume. In McKinsey's survey, about one in five respondents said AI operating costs were already limiting how much they use it.
- Setup and integration. Connecting the tool to your CRM, your ERP, your phone system, or your website, whether your team did it or a partner did.
- Your people's time. The hours your staff spent selecting, testing, and learning the tool, at their loaded cost.
- Ongoing care. Someone has to review output, update instructions, and fix what breaks. Budget for it every month.
The One-Page AI ROI Scorecard
Here is the page. One row per AI project. If a project can't fill in every column, that tells you something before you've done any math.
| Project | Owner | Business number it moves | Before | Now | Total cost to date | Return to date | Payback date | Decision |
|---|---|---|---|---|---|---|---|---|
| AI lead response and qualification | VP Sales | Lead-to-appointment rate | 18% | 27% | $64,000 | $210,000 gross profit | Month 5 (met) | Scale |
| Support chatbot on product manuals | Service Director | Tickets handled without a person | 0% | 41% | $48,000 | $72,000 (hire avoided) | Month 9 (met) | Scale |
| AI proposal drafting | Marketing Manager | Proposals sent per week | 11 | 12 | $31,000 | $0 | Month 12 (at risk) | Fix |
| Company-wide AI assistant licenses | None | None defined | n/a | n/a | $86,000 | Unknown | None | Stop or assign |
Illustrative numbers. Replace them with your own projects.
A few rules make the page work.
- One owner per row. Name a person. A department can't be held to a number. Google Cloud's 2026 survey of 2,403 executives found that among companies with accelerating AI returns, nearly half have clear ownership and decision authority for AI.
- One business number per row. Choose a number the business already tracks: close rate, cost per ticket, days to quote, revenue per employee. If the only measure is how many people use the tool, the project has no business case yet.
- A baseline taken before launch. Without a before, there's no after. If you didn't capture one, use the last full quarter before the tool went live.
- A payback date. Deloitte found most AI projects take two to four years to pay back. For a mid-size company, start with projects that can pay back inside 12 months, and say so on the page.
- A decision every quarter. Each row gets one of three calls: scale it, fix it by a set date, or stop it.
That last row in the example is the most common one in real companies. General-purpose AI licenses for everyone, with no owner and no target. It's worth keeping if people use it well. It still needs an owner and a number.
Take it to your next call
Five questions to ask about every AI project
Send these to your leadership team before the next review. A project that can answer all five belongs on the scorecard. A project that can't needs work before it gets more budget.
- 01Who is the one person accountable for this project's result?
- 02Which business number is it supposed to move, and what was that number before we started?
- 03What has it cost so far, including software, setup, and our own people's time?
- 04What has it returned so far in revenue gained, cost removed, or hours redeployed to named work?
- 05On what date does it pay for itself, and what happens if it doesn't?
What Companies With Real Returns Do Differently
The research on what separates the winners is consistent across firms.
They redesign the work. IBM's 2026 study of 2,000 CEOs found that organizations that redesigned five core business areas around AI were four times more likely to have delivered on their business objectives. Dropping a tool into an old process produces a slightly faster old process.
They track returns continuously and stop what isn't working. In a Gartner survey reported by TechInformed, high performers who track ROI continuously reported positive returns on 81% of their AI projects. Low performers didn't know the return on 29% of theirs. Gartner has also predicted that more than 40% of AI agent projects will be canceled by the end of 2027. Stopping a project early is a sign the scorecard is doing its job.
They aim AI at growth. The Wharton numbers above make the case. Cost savings are real and they have a ceiling. Revenue doesn't.
They buy more than they build. In the MIT study, AI tools bought from outside partners and customized reached full deployment about 67% of the time. Tools built internally got there about 33% of the time. The researchers caution that this is a correlation. For a mid-size company without a large engineering team, it's still a useful default.
We wrote about why projects stall before they reach this stage in The AI Pilot Graveyard. The pattern is the same: no owner, no number, no date.
What This Looks Like in Practice
The AI projects we see pay back fastest are narrow and tied to a number someone already cares about.
For Dupps, a manufacturer, we built a support chatbot trained on the company's own product documentation. It handles a large share of first-line questions without a person touching the ticket. The number it moves is support workload, and the return is capacity the service team gets back.
For TekStream Solutions, we built calculators and landing pages for each type of buyer. One campaign alone produced 308 high-value leads. Those tools aren't AI, and they make the same point: an asset earns its place on the scorecard when you can name the number it moved.
If you're deciding where the next project should go, start with the role you're about to hire. Our hire vs. automate calculator compares the two in about two minutes, and Grow Profit, Not Payroll explains the thinking behind it. Over time, the company-level number to watch is revenue per employee.
Interactive scorecard
Could you answer the board's AI question today?
Tick every statement that's true for your company right now.
Red flags checked
0 / 7
Nothing checked yet.
Tick each statement that's true for your company today. Nobody else sees this.
Book My Free Strategy SessionBring One Page
You don't need a consulting study to answer your board. You need one page that lists each AI project, who owns it, what it cost, what it returned, and what you've decided to do about it. Half the value is in the answers. The other half is in what the blank cells tell you.
If you'd like help choosing AI projects that can pay back inside a year and building them so the return is measurable from day one, that's the work our AI and automation team does.
Frequently Asked Questions
What is AI ROI?
AI ROI is the financial return a company gets from an AI investment compared with what it cost. It's calculated as return minus cost, divided by cost. Return includes revenue gained, costs removed, and staff hours redeployed to work that produces value.
How do you calculate ROI on an AI project?
Add up the gross profit from new revenue, the spending the project eliminated, and the value of hours redeployed to named work. Subtract the full cost: software, usage fees, setup, your team's time, and ongoing maintenance. Divide the result by the full cost. Compare against a baseline measured before the project started.
What is a good ROI for AI?
There's no universal benchmark, and most surveys rely on self-reported figures. A practical standard for a mid-size company is payback within 12 months on early projects. Deloitte found that most AI use cases take two to four years to reach satisfactory ROI, and only 6% pay back in under a year, so a project that clears 12 months is performing well.
How long does it take to see a return from AI?
Narrow projects tied to one process, such as lead response or support automation, can show a return in three to nine months. Company-wide programs take longer. In Deloitte's survey of 1,854 executives, most reported payback in two to four years.
Why do most AI projects fail to show ROI?
The common causes are no single owner, no baseline, no business metric, and tools added to a process nobody redesigned. Many companies also measure hours saved, and those hours never turn into revenue or lower spending. McKinsey found eight in ten employees say AI makes them more productive while only 37% of companies see any effect on earnings.
What should a CEO report to the board about AI?
Report one page: each AI project, its owner, the business number it's meant to move, the baseline and current result, total cost, return to date, the payback date, and a decision to scale, fix, or stop. Add total AI spending as a share of revenue and the trend in revenue per employee.

5K Team
Our team helps companies to increase revenue, decrease costs, increase efficiency, and scale employees using digital marketing and AI technology.
Tags
Ready to Grow Your Business?
SCHEDULE A GROWTH STRATEGY SESSIONRelated Articles

Revenue Per Employee by Industry: The Benchmarks Every CEO Should Know Before the Next Hire
November 17, 2026
Read More

Marketing Budget as a Percentage of Revenue: The Real Benchmarks, and a Better Way to Set Yours
November 17, 2026
Read More

AI Agents as Buyers: How to Be the Company the Machines Recommend
November 17, 2026
Read More
