Short answer
Measure the ROI of AI by choosing one business number the AI should change, such as pipeline, cost per unit of work or hours on a process, and recording its baseline before anything ships. Report every release against that baseline and count the full cost, including people's time. ROI is the change in the number times its value, minus total cost, divided by total cost.
Key takeaways
- Only 18% of professionals surveyed by Thomson Reuters in 2026 knew their organization tracked AI ROI.
- Adoption metrics like logins and seats measure activity, not return.
- Without a baseline taken before launch, every later result is an argument.
- Model spend is often the smallest cost; the people directing and reviewing the work cost more.
Most companies can't say what their AI spending returns. In Thomson Reuters' 2026 survey of more than 1,500 professionals, only 18% said they knew their organization was tracking the ROI of AI tools, and most of those tracked internal, operational metrics rather than business results. MIT's Project NANDA reported in 2025 that 95% of the organizations it studied saw no business return from generative AI. The fix is to decide what the AI is supposed to change before anyone builds it.
Step 1: Pick one number
Choose the number that matters most this year and that the AI can plausibly move within a few months. Good candidates are specific and already reported somewhere:
- Revenue: qualified pipeline, conversion rate, revenue from a new product.
- Margin: cost per unit of work, such as cost per order, per claim or per report.
- Hours: time spent on a named process, such as proposals, research or onboarding.
Avoid adoption metrics (logins, prompts, seats used). They measure activity, not return.
Step 2: Measure the baseline first
Record where the number stands before anything ships, over a period long enough to smooth out noise. Without a baseline, every later result is an argument.
Step 3: Report every release against it
Ship in small releases and report each one against the baseline in business terms. If a release doesn't move the number, change the work. This is the core of revenue-focused AI consulting.
Step 4: Count the full cost
The cost side includes build time, model spend, tools and the hours your team puts in. Model spend is often the smallest line. Our autonomous agent control plane merged 23 pull requests in one 8-hour run for $1.42 in model spend; the people directing and reviewing the work cost far more than the model did.
A simple formula
AI ROI = (change in the number × its value − total cost) ÷ total cost.
If a proposal tool saves 200 hours a quarter at a loaded cost of $100 an hour, that's $20,000 of value a quarter. If it cost $40,000 to build and $2,000 a quarter to run, it pays back in a little over two quarters. The inputs are yours; the discipline is agreeing on them before the build.
What we do at Cosmic
Every engagement starts with a free 30-minute call, then a fixed-scope discovery session of two or four hours with your executive sponsor, where we agree on the number and measure its baseline. Each sprint has a fixed price plus an outcome-linked fee tied to that number, so part of what we earn depends on the ROI being real. More on that in how AI consulting is priced, and on the full approach in AI implementation.
FAQ
How do you calculate the ROI of AI?
ROI equals the change in the chosen business number times its value, minus the total cost of building and running the AI, divided by that total cost. The change is measured against a baseline recorded before the AI shipped.
What metrics should you use for AI ROI?
One business number: revenue (pipeline, conversion), margin (cost per unit of work) or hours on a named process. Avoid adoption metrics such as logins or prompts, which don't show a return.
Why do most AI projects fail to show ROI?
They start with a tool instead of a number, skip the baseline and measure usage. MIT's Project NANDA reported in 2025 that 95% of the organizations it studied saw no business return from generative AI.
Sources
Founder of Cosmic. Nearly 15 years in strategy, management and sales as a digital strategist, director of sales and product manager. He closed over $25M in new revenue at VMware Pivotal Labs, launched digital products at Capital One, and helped Viget grow from 30 to 75 people. He started Cosmic and built the Revenue Design® process after advising friends whose companies struggled to close deals and keep revenue steady.