Most people who ask me about AI consulting ROI can’t tell me how many hours a week their team spends on the work they want automated. We’re already talking about getting those hours back, so I’d like to know how many we’re shopping for.
MIT looked at over 300 AI rollouts in 2025 and found 95% of companies had nothing they could point to on the P&L. The models themselves worked. The tools built around them didn’t learn from use or fit how people worked. A lawyer in that study kept using the $20 ChatGPT plan after her firm paid $50,000 for a contract tool because ChatGPT gave her better answers. I’d bet the $50,000 got recorded somewhere, even if nobody thought to check what it bought them.
I run a $3M business on about three hours a week at my laptop, which probably gives away where I stand on whether this stuff pays off.
I keep watching people cancel at day 45, though, when the systems have only been running for part of the month and the spreadsheet still looks a bit thin. Another two weeks could have shown the savings in the same sheet they were using to talk themselves out of it.
What AI consulting ROI measures
Your accountant won’t need a new formula for AI consulting ROI. It’s the one you’d use for a new hire or a piece of equipment. Subtract what you paid from what came back. Divide that by what you paid. A $15,000 build that gives you $68,000 of value in year one comes to about 350%, which they could probably confirm in thirty seconds.
They’ll need your numbers first, though. Most owners haven’t counted those three inputs, so the accountant’s thirty seconds tends to come after a few awkward questions about where everyone’s week went.
Hours given back to your team
Start with the hours your team spends on the work the system will take over. Include salary and benefits in the cost, along with software seats, then divide that total across the hours your people really work. Twenty hours a week at $60 an hour comes to $62,400 a year.
Most owners guess low here because four or five people each do a bit of the work. Everyone’s bit sounds small until you put them all on the same spreadsheet.
Outside spend that stops
This is the agency retainer, the contractor, the firm that processes your documents, any invoice that stops arriving once the system is running.
MIT’s Project NANDA study, The GenAI Divide, looked at over 300 public AI deployments in 2025 and found the clearest savings right here. Its best-in-class organizations were replacing $2 million to $10 million a year of outsourced processing with back-office automation. Agency spend in the same group dropped about 30% on external creative and content.
Revenue from moving faster
Among those same best-in-class organizations, lead qualification ran 40% faster once it was automated. Customer retention went up about 10% where the follow-up had been automated too. I keep this column separate from the other two. It takes months to attribute with any confidence, so it shouldn’t be propping up your first ROI number.
On the cost side, you’ve got the fee, your team’s time on the calls, plus whatever it costs to keep the system running afterwards. For me that’s a $100-a-month Claude plan on top of tools I already pay for.
One more thing before the studies. Whatever ROI figure you end up with is only as good as the baseline you took before the build started. Hardly anyone takes one. That single habit explains most of the numbers in the next section.
Why most AI budgets never turn into an ROI number
Three studies say the same thing, and it isn’t the technology. The tools mostly worked. What was missing was a number somebody wrote down before anyone started.
IBM asked 2,000 CEOs across 33 countries in early 2025 how their AI spending was going. 25% of their AI initiatives had delivered the return they expected. 16% had scaled enterprise-wide. 64% of the CEOs admitted they’d invested before they understood what the technology was worth, mostly because they were scared of falling behind.
Project NANDA’s number that July was worse. Enterprises had put $30 to $40 billion into generative AI. 95% of the organizations studied had nothing measurable to show for it on the P&L. The researchers said the models weren’t the issue. The tools built on them didn’t learn from anyone, held no context from one session to the next, and didn’t fit how work actually happened inside those companies. A lot of pilots looked great in the demo and then never got used.
One caveat, since I’m leaning on it three times. That report is a preliminary working paper, not peer-reviewed, from a group that builds agentic AI infrastructure and concluded the fix was agentic AI infrastructure. It has taken criticism for that, and for how the headline got repeated as 95% of pilots failing, which is not what it says. The part I’d stand on is narrower. 95% had no measurable P&L impact, and mostly that’s because nobody wrote down a baseline to measure against. Which is the argument I’m making anyway.
Gartner’s survey of 782 infrastructure and operations leaders, run in late 2025 and published in April 2026, has the detail I keep quoting. 28% of their AI use cases fully met ROI expectations and 20% failed outright. 57% of those leaders had had at least one failure, and most of them gave the same reason. They’d expected too much, too fast.
Line those three up and the shape repeats. Nobody had agreed what working would look like, and no date had gone in the calendar to check. So month two came around, the dashboard was empty, and the CFO asked what the money had done.
The AI consulting ROI timeline, one month at a time
The ROI formula has no clock in it. An AI build needs one, because the value ramps up over weeks. This is how the AI consulting ROI timeline runs inside our two-month engagements, with the studies above dropped in where they apply.
Days 1 to 14, baseline and build plan
Implementation starts within 15 days of the deposit. Week one is what we call Map and connect. We audit your tools, your goals and your content, then agree the build plan. It’s also the last clean moment to take a baseline, before anything changes, so we take it then.
Hours per workflow, spend per vendor, lead response time in minutes, all written down with a date on it. Everything you measure at day 90 gets compared against this sheet, so be fussy about it.
ROI at this stage is negative. You’ve paid a deposit and nothing has shipped.
Days 15 to 30, first systems live
From here it’s working calls every week. Systems go live as they’re built, one at a time. The early ones are usually the loops that handle your inbox, drafts in your voice, a summary waiting when you wake up. They’re small enough that you notice them straight away. What you can’t do yet is prove the hours on paper, because the baseline sheet doesn’t have a full month to compare against.
Deloitte’s October 2025 survey of 1,854 executives across Europe and the Middle East found only 6% of companies saw AI payback inside a year. Most reported satisfactory returns in two to four years, against the seven to twelve months people expect from ordinary tech spending. The gap between that and a 60-day build is scope.
A company-wide AI program has a hundred moving parts and a steering committee for each one. A build aimed at the workflows that eat your week has a handful. You’re the only person deciding anything about them.
Days 31 to 60, the day-45 trap
This is where people cancel. Your first full month of comparison data lands around day 45. It looks thin, because the systems only ran for part of that month while the sheet compares it against a full one.
Without a warning, it looks like the build isn’t working. The Gartner leaders who said they’d expected too much, too fast were standing about here when they pulled the budget. I wrote a whole post on why most AI implementations stall in month two. The short version is that day 45 is a data problem. The system underneath is usually fine.
Deloitte and the University of Hong Kong put numbers on that from the other side. In their January 2026 index of more than 100 executives across mainland China and Hong Kong, the second most cited reason AI underperforms was a lack of early visible results, at 32%, which they describe as stakeholders pulling support before a project can deliver. Technical limits ranked below organizational and execution problems, 39% against 50% and 47%.
Inside our builds, day 60 is handoff. You own the Claude setup, the skills, the routines, the playbooks. There’s no ongoing dependency on us. From here, your running cost is a Claude plan plus whatever tools you already had.
Days 61 to 90, the first number you can trust
Now you’ve got a full month of the system running at proper speed against a dated baseline. Hours per workflow then and now. Vendor invoices then and now. This is the first AI consulting ROI figure worth showing a partner or a board. At this point it’s usually the direct savings on their own. Revenue effects are still fuzzy at 90 days. Leave them out of the first report.
Days 91 to 180, where it compounds
The systems you took ownership of at day 60 have now had four months of your corrections. This is where it begins to compound. Voice DNA gets tighter with every draft you edit. Routines get extended to the next workflow once the first one runs without you. A year in, most clients are running something a lot closer to an AI employee than a set of tools.
It’s also the window where IDC’s number starts to make sense. Their Microsoft-sponsored survey of more than 4,000 business leaders in late 2024 put the average return at $3.70 per dollar, top performers at $10.30, with value typically landing inside 13 months. It’s vendor-funded, so I wouldn’t plan around it. The pattern underneath it matches what I see. A system that’s been learning from you for five months does a lot more for you than the same system did at month two.
| Window | What’s happening | What to measure | Realistic ROI read |
|---|---|---|---|
| Days 1 to 14 | Audit, baseline, build plan agreed | Hours per workflow, spend per vendor, response times | Negative, deposit paid, nothing live yet |
| Days 15 to 30 | First systems live, weekly builds | Which workflows are running, first hours back | You’ll notice hours coming back before the sheet shows it |
| Days 31 to 60 | Remaining systems live, handoff at day 60 | A partial month of comparison data | Looks thin, a partial month against a full baseline |
| Days 61 to 90 | Full speed against the baseline | Direct savings only, hours and invoices | First figure worth reporting |
| Days 91 to 180 | Corrections compound, routines extended | Direct savings plus attributed revenue | Cumulative net usually positive on Tier 1 inputs |
What ROI looks like on my own calendar
My week went from 90 hours at a laptop to about 3. Since I ask clients for their numbers, here are mine.
Before AI, I ran a content agency with close to 100 people. It needed 7 years and 60,000+ projects to reach $100k a month. First Movers started with 2 people and passed $100k a month in month 11. As of September 2026 it’s 15 people doing $250k a month. The rest of the week happens with a notetaker wired into Claude while I’m out walking, plus a set of loops that summarize my email, pre-write my drafts, and hand me action items every two hours.
Put that into the formula and the laptop hours alone are 87 a week returned, on a business at a $3M run rate. Count it the way I’m telling you to count yours and it’s less than that, because I still review on my phone through the day. I’ve never once needed the revenue column to justify what we built. Most clients find the same once they’ve actually added up the hours.
Inside a build, what gets tracked and where
Every engagement follows the same six stages, which we call the MOVERS methodology. Each stage leaves something behind that you can measure. This is what each one looks like from the inside and which number it gives you.
Mission, week one
Who you are, who you’re reaching, and the numbers that say the opportunity is real. This is also where the baseline table gets built, because you can’t set a target without knowing what you’re starting from.
Opportunity
What you sell and the opening it fills. The output is the build plan document: the workflows we’re taking on and the order we’ll ship them in. Every number you look at later in the engagement points back to that page.
Voice
Voice DNA is how we capture how you think, talk, decide and sell, then install it as your Claude’s brain. Mine is trained on over 100,000 words of my own writing. The number to watch here is editing time. Drafts that used to need an hour of fixing come back needing a quick read.
Efficiency
These are the workflows, automations and loops that run without you touching them. Each loop logs what it produced and when. That’s how you know a routine ran at six in the morning and the summary was sitting there four minutes later. If marketing is where you’re the bottleneck, this is where AI marketing automation gets wired in.
Results
Hours back, pipeline moving, content shipping every week. Dr. Miranda, a medical practice owner who did a Done-for-You build with us, has her patient list ready every morning before she’s awake, labs pulled, medication and peptide recommendations drafted to her own prescribing parameters. She’s since referred ten other doctors who want the same setup.
Sustainability
This one’s about whether the system keeps running next quarter without us in the room. The handoff folder holds the playbooks, the skills and the routines. All of it is yours.
Four things that move the AI consulting ROI timeline
Some builds cross zero in month three. Some take until month six. The same four things decide which, every single time.
Documentation decides the first stretch. A team that can describe its client onboarding in eight steps usually has a working system in week two. A team whose answer is “it depends” spends week two writing down the eight steps. Both get there. The second one pays for an extra two weeks of build.
Then there’s how many workflows you point it at. The tools that made it in Project NANDA started at the edge of a workflow, proved value there, then scaled into core processes. We scope to what a team can actually absorb in two months. That’s the reason the ramp fits inside sixty days.
You matter more than either. These builds only work with you on the calls making decisions. The fastest timelines I’ve seen came from owners who treated the two months like a new hire’s first two months. They brought questions, corrected the outputs when something was off, then gave the system a few weeks to learn them before judging it.
The last one decides what happens once we’re gone. Gartner found that 33% of the I&O leaders with AI success had embedded it into the systems and processes people already used. A system your team owns and understands keeps compounding after we leave. One that only the consultant understands tends to stop improving around the time the invoices do.
Five questions to ask before you sign
Ask these of anyone you’re considering, us included. The answers tell you whether you’ll get a real AI consulting ROI figure out of the engagement or just a good feeling about it.
Who takes the baseline and when?
A named person, in week one, producing a dated sheet that both sides keep.
Which three numbers do we review, on which dates?
Hours per workflow, outside spend and response time, each with a date in the calendar. Anything vaguer produces a review nobody can act on.
What goes live in the first 30 days?
Something should, even if it’s small. Promising the moon in week one usually means off-the-shelf tools with a markup. Promising nothing until month four usually means a strategy deck.
Who is on the calls each week?
Plenty of firms sell on senior names and deliver with junior contractors. Ask to meet the people you’ll actually work with before you pay a deposit.
What do we own on the last day and what does it cost to run?
With us, all of it, at roughly the price of a Claude plan. If the answer includes a monthly retainer to keep things running, the ROI arithmetic changes. Put that retainer on the cost side before you compare anyone’s fee.
Where to start
If you run a real business and most of the work still lands on your desk, we map your build before you spend a dollar. Bring your hours. We’ll put the baseline sheet together on the call, and you’ll leave with a payback estimate whether or not you go ahead. The call is free.
If you’re pre-revenue, or you’d rather learn the systems and build them yourself, Labs teaches the same methodology, across 80+ courses and live builds twice a week, with 3,000+ students trained on it.
FAQ
What is a good ROI for AI consulting?
Anything above 200% in the first year is a good AI consulting ROI for a scoped, two-month build. That’s every dollar of fee coming back as three dollars of value across hours returned, outside spend replaced and revenue you can attribute. The IDC and Microsoft 2024 survey put the average generative AI return at $3.70 per dollar, with top performers at $10.30. That’s an average across companies of every size and every kind of project, so treat it as context.
How long does it take to see ROI from AI consulting?
On a two-month build, the first hours come back inside 30 days. The first figure you can stand behind lands between day 60 and day 90. Cumulative net crosses zero somewhere between month three and month six, depending on the fee and the hours involved. Company-wide AI programs take far longer. Deloitte’s October 2025 survey of executives across Europe and the Middle East found only 6% saw payback inside a year. Most reported two to four years.
How do you measure AI consulting ROI?
Take a dated baseline before the build starts, covering hours per workflow, outside spend per month, and response times. Measure the same three things at day 30, 60, 90 and 180. Direct savings are hours returned times the fully loaded rate, plus any outside spend that stopped. Keep attributed revenue in a separate column until you have at least a quarter of data.
Why do so many AI projects fail to show ROI?
MIT’s Project NANDA study put 95% of organizations at zero measurable return in July 2025. It traced that to tools that didn’t learn, held no context, or had been fitted badly to how work was actually done. Gartner’s 2026 survey added that 57% of infrastructure and operations leaders had had at least one AI failure, most often because they expected too much, too fast. Behind both findings there’s usually a missing baseline. You can’t show a return against a number nobody wrote down.
Does the cost of AI consulting change the ROI?
Yes, in both directions. A $15,000 two-month build that settles at $6,500 a month of net value pays the fee back in month four, counting the partial months while the systems were still going live. A $40,000 build has to take on more workflows or replace more outside spend to reach the same date. The running cost after handoff counts just as much. A retainer to keep the system alive eats into the return every month. Our guide to AI consulting cost covers the ranges.
The date is yours to set
Set the date, hold your consultant to it, and give the build until day 90 before you decide anything. That lawyer’s firm never set a date. $50,000 later, nobody there could say what the money had done.
