Article
AI ROI: Why Most Companies See No Payoff Yet
Most companies still can't show AI ROI. Here is what Apollo and MIT found, why pilots stall, and how to pick work that shows up in the numbers.
- ai
- roi
- strategy
- comparison
AI ROI is the question every leadership team is asking now. You bought the seats. People use the tools. So where is it in the numbers? For most companies, it isn't there yet. Two pieces of research from the last year say so plainly, and they also point to why.
This post covers what Apollo's chief economist and an MIT study found, the three patterns behind AI work that doesn't pay, and a comparison of the ways companies are trying to fix it.
What the research says about AI ROI
Apollo, October 2026. Torsten Slok, chief economist at Apollo, wrote that AI has delivered "zero financial payoff to corporate America outside of tech." He looked at the S&P 500 minus the seven biggest tech companies, which he calls the S&P 493. Revenue per worker is flat. Profit margins are flat. There are no signs yet of AI lifting productivity growth. TechCrunch covered the note and the line that stuck was this one: "AI may be a great calculator, but in business, intelligence is mostly judgment."
MIT, 2025. An MIT NANDA report called "The GenAI Divide" looked at 300 AI projects, interviewed 150 executives and surveyed 350 employees. As Fortune reported, 95% of AI pilots delivered no measurable savings or profit lift.
Those are big claims and you can argue with the methods. But they match what most operators see. Lots of activity. Not much that shows up on the P&L.
Why AI work doesn't show up in the numbers
From the research and from what we see in established companies, three patterns explain most of it.
1. The goal was "use AI," not a number. A team gets licenses and a mandate to experiment. Nobody writes down which metric should move. Six months later, people like the tools and nobody can say what changed.
2. It went to the visible work, not the expensive work. The MIT report found many companies put AI into sales and marketing when the bigger payoff may be in back-office processes. Writing more emails is visible. Cutting the hours spent on invoice matching or order entry is where the money often is.
3. Building when buying would work. In the same report, purchased AI tools succeeded about 67% of the time. Internal builds worked about one-third as often. Building your own makes sense in some cases. It's rarely the fastest path to a result.
There's a fourth one we'd add. The AI did its part and the process around it didn't change. An AI tool qualifies leads in seconds and then a rep calls them back in two days. The AI worked. The ROI didn't. We wrote about this in why AI isn't increasing sales.
How to measure AI ROI without fooling yourself
Pick one workflow. Measure it before. Measure it after. That's the whole method, but the details matter.
- Name the metric in dollars or hours. "Hours per week on quote prep" or "cost per booked meeting." Not "productivity."
- Take a baseline for at least two to four weeks before turning anything on.
- Count the full cost. Licenses, setup time, the hours people spend checking AI output.
- Compare the same period after rollout, and watch quality too. Faster and wrong isn't a win.
- Decide in advance what result means keep, change or stop.
If your marketing tracking is shaky, fix that first. Our guide to tracking marketing ROI covers the basics.
Ways companies are going after AI ROI
There's no single right path. Here's how the common options compare.
Buy a vendor tool for one job
Pick a tool built for one workflow, like support ticket triage, invoice processing or call notes. The MIT finding favors this path. The risk is buying five tools that don't talk to each other.
Roll out a general assistant to everyone
ChatGPT, Gemini or Copilot seats for the whole company. Good for adoption and comfort. Hard to tie to a number unless you pair it with specific workflows. We wrote about getting your team to use AI if adoption is the gap.
Build internally
Makes sense when your data or process is truly unique, or rules keep data in house. Expect it to take longer, and staff it with people who will own it after launch.
Hire a consulting firm for a roadmap
A firm studies your business and hands back a plan. Useful if the plan gets executed. Often it doesn't, because the people who wrote it leave when the project ends.
Work with a growth partner
AI Answered is one option here. We're a growth partner for established companies. We put AI to work on revenue: ads, lead follow-up and the sales process, and we measure the result in pipeline and sales rather than usage. We look at your numbers first, then price on the value we can deliver. If the real payoff for you is back-office automation, we'll say so, and that may be a better job for a specialist tool. More on our growth partners page.
Comparison table
| Approach | Speed to a result | Ties to a number | Main risk | |---|---|---|---| | Vendor tool for one job | Fast | Yes, if you baseline | Tool sprawl | | Assistant for everyone | Fast to launch | Weak | Usage without results | | Internal build | Slow | Yes | Low success rate in MIT study | | Consulting roadmap | Medium | Depends | Plan without execution | | Growth partner | Medium | Yes, revenue metrics | Wrong fit if the payoff is back office |
A simple first step
Pick the one workflow that costs you the most hours or the most lost revenue. Write down the number today. Then pick the smallest AI change that could move it, and check again in 30 days.
What the first two weeks should produce
A useful look at those first weeks ends with a short list, not a slide deck.
Days 1 to 3: watch the work. Sit with the people who answer the phone, send the quotes, and chase the invoices. Write down repeated steps, handoffs, waiting, and the spreadsheet someone built because the real system does not do the job.
Days 4 to 7: sort what you found. Four zones cover most of it:
- Conversion. Where visitors, leads, or quotes fail to become customers. Slow follow-up is the common leak. We cover that in why deals stall.
- Automation. Repeated tasks you can measure in hours per week: intake, reminders, data entry, first-draft emails.
- Custom tools. Only when nothing you can buy does the job.
- Lead sourcing. Where new customers come from, and which channels you are ignoring.
Days 8 to 14: put your own numbers on the top items, then start one. For each fix near the top, write what it costs you today, what you expect to change, and what it would take to build or run. Start the one with the best mix of value, speed, and low risk. Track it in your systems. If the projection was wrong, find out why before you go bigger.
At the end of two weeks you should have a map of how the work actually runs, a ranked list, a one-page sheet of the math, and the first fix already started. If you don't have those, you got a report.
Skip the exercise if the process still changes every week, you already know the exact fix and only need it built, or nobody has a few days to sit with the work.
FAQ
What is AI ROI? The return you get from AI compared with what it costs, measured in revenue, profit or hours saved. Usage alone isn't ROI.
Is it true 95% of AI projects fail? That's the MIT finding for pilots that showed no measurable financial lift. It doesn't mean the tools don't work. It points at how they were chosen and rolled out.
How long does it take to see AI ROI? For one well chosen workflow, a few weeks to a few months. For a company-wide change, longer.
Should we build or buy? Start by buying for common jobs. Build only where your process or data is truly unique.
Ready to put this into your business? We map how you run, build what you approve, train your team, and prove the hours and dollars saved.
Related articles
Nov 15, 2026 · 5 min read
Company-Wide AI Rollout: What Uber's Move Teaches
Uber just put all 30,000 employees on ChatGPT. Here is what a company-wide AI rollout needs, the tools companies pick, and the mistakes to skip.
Nov 22, 2026 · 5 min read
Pay for Performance PR: 4 Firms That Charge Per Placement
Pay for performance PR means you pay when coverage runs, not every month. Four firms compared, what they charge for, and what to check first.
Nov 21, 2026 · 5 min read
AI Consulting Firms in Ohio: 6 to Compare in 2026
Comparing AI consulting firms in Ohio? Here are six, from Cincinnati to Toledo, what each says it does, and the questions to ask before you hire.