Ninety five percent of corporate AI pilots produce no measurable return, MIT researchers found, and small businesses have the least runway to absorb it. A study released July 22, 2026 by ManpowerGroup Talent Solutions and Everest Group puts a number on why: just 3% of organizations say their leaders are fully prepared to run an AI enabled team. The gap between buying AI and being ready to run it is where small business owners are losing money right now.
Most owners blame the software when a pilot flops. A pilot that fails for lack of a documented process leaves something else behind: a paper trail showing exactly how dependent the business is on one person. That is the same record a lender or buyer goes looking for when the owner tries to borrow against the business or sell it.
Leaders Are Buying AI Faster Than They Can Run It
Companies have poured $30 billion to $40 billion into generative AI over the past two years. MIT’s Project NANDA, a research initiative tracking enterprise AI deployment, reviewed more than 300 public AI initiatives, ran 52 structured interviews and gathered 153 survey responses from senior leaders at four industry conferences between January and June of last year. The finding: 95% of pilots show no measurable effect on profit or loss. Only 5% are extracting real value, and it is concentrated in a narrow slice of back office work.
Leadership readiness looks just as thin. ManpowerGroup Talent Solutions and Everest Group surveyed 80 senior leaders across the US and UK in healthcare, life sciences, manufacturing and technology for a report titled The New Talent Equation: Activating Workforce Confidence at Scale. Only 3% called their leaders fully prepared to manage AI enabled teams, even though 78% of employees say they are worried about how AI will change their jobs. Just 17% of organizations describe their AI use as advanced enough to show up in business results.
McKinsey’s 2025 State of AI Survey landed on nearly the same gap: 1% of organizations report full maturity in their AI strategy, even as 88% have deployed some form of AI already. The lag between deployment and mastery is not confined to small operators. Microsoft’s own security AI recently made a similar admission, saying human analysts alone can no longer keep pace with attack volume without machine help. Record enterprise tech budgets are running into the identical wall; a stubborn AI payoff gap now shows up at $6 trillion in scale, just with more zeros attached.
| Study | Who They Measured | Key Finding |
|---|---|---|
| MIT Project NANDA | 300+ AI initiatives, 52 interviews, 153 leaders | 95% of pilots show no P&L impact |
| ManpowerGroup / Everest Group | 80 senior leaders, US and UK | 3% call leaders fully prepared for AI teams |
| McKinsey State of AI Survey | Global organizations | 1% report full AI strategy maturity |
| Boston Consulting Group | AI transformation programs | 70% miss leaders’ own results targets |
Read together, the four studies describe the same company from four angles: one that bought the tool before it built the habit.
What the Successful 5% Did Differently
NANDA’s researchers found a clear pattern behind the pilots that worked. They clustered around repetitive tasks that already had an established process: invoice processing, scheduling, back office functions with a known right answer. None of them involved dropping a general purpose chatbot onto an undefined workflow and hoping it adapted.
Buying from a specialized vendor or forming a partnership succeeded roughly 67% of the time in NANDA’s sample. Building the same capability in house succeeded about a third as often. A flexible looking chatbot on a sales page is not a substitute for a business that has already mapped its own workflow, and a small business copying a Fortune 500 company’s software stack without that groundwork is copying the wrong half of the equation.
Boston Consulting Group has quantified where the effort should actually go: 10% on algorithms, 20% on technology and data, and 70% on people and process, meaning the roles, workflows, governance and change management around the tool. BCG also found that companies that dedicate at least 10% of budget to training are one and a half times more likely to succeed than those that skip it. That mirrors what already went wrong in one documented small business back office function: an automated bookkeeping tool that traded one error for a sneakier one once it ran unsupervised.
The Bill Arrives After the Pilot Ends
Most small business owners skip a step before buying AI: writing down who else could make the call if the founder disappeared for a week. The National Association of Insurance Commissioners, an organization that coordinates state insurance regulation, found that 71% of small businesses depend on one or two people for their overall success. Nationwide’s research on the topic cites the same NAIC figure.
- Key person risk – the financial exposure a business carries when its results depend on one or two people whose absence, departure or death would disrupt daily operations, a category insurers and lenders both price separately from the business itself.
Businesses built around one person do not just run riskier. They sell for less. Acquirers treat founder dependency as a single point of failure and discount the price accordingly, and Nationwide’s own guidance on key person risk walks business owners through exactly that exposure. Small Business Administration lenders run their own version of this test during an acquisition. A firm can pass it and still receive less funding than expected, face a longer transition period with the founder kept on payroll, or get told to carry life insurance on the person the whole operation still runs through.
An AI pilot that fails for lack of documentation is not a separate problem from that lending test. It is a preview of it, produced months or years before the owner ever sits across from a loan officer.
A Historic Wave of Business Sales Is Right on Schedule
This is not theoretical timing. A tidal wave of retiring business owners, stronger government backed lending and a growing pool of acquisition minded buyers are converging on what industry trackers describe as one of the largest ownership transitions in US small business history. Roughly 62% of owners say their retirement timelines have accelerated over the past five years.
The lending market is already moving to meet it. SBA loan approvals reached $37.3 billion across more than 78,000 loans through May of fiscal 2025, an 11% jump from the same stretch a year earlier. Change of ownership loans, the ones that finance a buyer acquiring an existing business rather than starting one from scratch, are among the fastest growing categories in that total. Trackers put the peak transition year at 2027 to 2028, with elevated volume extending years beyond that.
Every business swept into that wave gets assessed the same way: can it run without the person currently running it. An AI pilot built on a documented process answers that question in the owner’s favor. One built on improvisation answers it the other way, right as the buyer pool sizes up the business for sale.
The Founder Who Closed the Gap Alone
Anthony Godley has lived both sides of that gap. He founded Logix BPO, a business process outsourcing company, with a single client. He grew it into an operation employing more than 1,000 people and became its chairman in 2025.
The biggest barrier to AI adoption isn’t technology. It’s founder dependency. If every important decision still comes back to the founder, AI will only expose that bottleneck faster.
Godley said that in a written statement describing what separates companies that scale from those that stall. The businesses that succeed with AI, he said, already have disciplined processes and leaders who do not have to make every decision themselves, because AI amplifies operational maturity rather than replacing it.
His prescription runs in a specific order: make the decisions first, then bring in the software. Write down every approval and escalation that currently has to come back to the founder. Then write down how each of those calls actually gets made, including the judgment calls a spreadsheet or an AI model will never be able to render on its own.
Which Employee Could Run Your Business for a Week?
Pick the person on staff with the least written down about their job, the one whose tasks live mostly in their own head, and cross train someone else on that role before the month ends. Companies that beat the 95% failure rate already knew who could step in before they bought a single AI seat, not because a vendor told them to check, but because the habit was already part of how they ran the place.
Godley frames it as three questions worth answering before the next AI purchase.
- Who else in the organization has the authority to make this decision about how AI gets used?
- Is there a written definition of success for this task, one detailed enough that someone other than the owner could read it and know if the job got done?
- Would the process keep functioning if the owner were out of town for a week?
A firm that cannot answer those three questions today will not answer them tomorrow just because it added a subscription. The software only buys more tools for longer, without forcing the questions that determine whether those tools pay for themselves. Once the answers exist, measuring AI’s return on a single documented process, and weighing that against its monthly cost, becomes the easy part of the project rather than the whole of it.
Frequently Asked Questions
What Loan Structure Do Buyers Use to Acquire a Small Business?
Buyers are commonly financing acquisitions through SBA backed loans layered with seller notes and other hybrid structures, letting a buyer take over a cash flowing business with as little as 5% to 10% down while preserving working capital. More than a third of Gen Z and Millennial owners surveyed say they plan to acquire a business from a retiring owner rather than start one from scratch, and Gen X currently holds the largest share of search fund operators using this kind of financing.
What Does Key Person Insurance Pay For?
A key person policy pays the business itself, not the employee’s family, if a critical person dies or becomes disabled. The payout is meant to cover the cost of replacing that person, which NAIC data cited by Nationwide puts at 100% to 300% of their annual salary depending on their expertise and responsibilities, buying the business time to recruit, retrain or restructure around the loss.
Why Do General Purpose Chatbots Fail More Often Than Task Specific Automation?
A chatbot has no built in definition of what success looks like for a given task, so it improvises against an open ended request with no scorecard behind it. Invoice processing or scheduling already has a clear before and after, paid or unpaid, booked or not, which gives the AI a fixed target instead of a guess. That is why the successful 5% cluster around narrow, already documented processes rather than broad conversational tools.
Is Founder Dependency the Same Thing as Key Person Risk?
They overlap but are not identical. Key person risk can apply to any critical employee, including one who could plausibly be trained, insured against or replaced with enough lead time. Founder dependency is the harder version of that problem, since a founder’s judgment, relationships and unwritten decision rules are often the least documented part of the entire business.
What Is the First Process a Small Business Should Document Before Buying AI?
Pick one process with a quantifiable result already attached to it, such as hours spent per invoice or days to close a scheduling request, and write down every decision point in it before adding any software. That documented baseline is what makes it possible to measure AI’s actual return afterward, rather than guessing at whether the subscription paid for itself.
