How do we make sure our AI project actually moves the numbers?

Define the metric before you build - hours saved, errors prevented, faster response time - then redesign the one process around the agent rather than bolting it on. Keep human approval on anything customer-facing or financial, measure for a few weeks, and only expand once the number has moved. Scope, ownership and redesign decide ROI, not the size of the AI budget.

Record AI spend, no earnings bump: the 2026 gap and your business

Q: What does workflow redesign actually mean for a small business?

It means picking one recurring process - order intake, invoice handling, quote turnaround - and rebuilding the steps so an agent owns the routine path end to end, with a person approving the exceptions. It is not a company-wide transformation. It is one bounded process, redrawn so the human touches it only where judgment is needed.

Q: Does a smaller company have an advantage in closing this gap?

Yes. A smaller company can redraw a whole process in a single afternoon because there are fewer stakeholders, fewer sign-offs and less legacy politics. Large enterprises struggle to redesign because every step touches another department. That structural simplicity is why a well-scoped small deployment can land on the right side of the earnings gap faster than a Fortune 500 program.

The AI earnings gap is the distance between what companies spend on AI and what actually reaches the bottom line. In late August 2026, McKinsey's State of AI survey put a hard number on it: despite record investment, only 6% of organizations turn AI into meaningful earnings, and just 39% report any profit impact at all. The striking part is why. The gap is not caused by weak models or thin budgets. It is caused by bolting AI onto processes nobody redesigned - and that is precisely the trap a smaller, focused business is built to avoid.

Key takeaways. McKinsey's State of AI 2026 found only 6% of companies are AI high performers and only 39% see any EBIT impact, even as spending hits records. The single clearest differentiator is workflow redesign: high performers are about 2.8 times more likely to have fundamentally rebuilt a process around AI (55% versus 20%). For a small or traditional business the lesson is freeing - the constraint was never the budget or the model, it is whether you redraw one process instead of decorating the old one. Smaller companies can redesign faster, so they can land on the right side of this gap sooner.

The number that should stop you: 6%, not 60%

Most AI coverage celebrates adoption, and adoption is genuinely high - 62% of organizations are at least experimenting with agents and 23% are scaling one somewhere, according to McKinsey's State of AI 2026. Yet the earnings picture is far starker: only 39% of companies report any measurable profit impact, and a mere 6% qualify as high performers who attribute real value to AI. Put the other way, for the overwhelming majority record spend has not moved the needle. That disconnect, first reported in coverage of the survey, is the story of AI in 2026 - not the model race, but the value race, and most companies are losing it.

The one variable that separates the 6%

Here is the finding that matters most in practice. The high performers are not the biggest spenders; they are the ones who redesigned their work. McKinsey's data shows they are about 2.8 times more likely to report fundamental workflow redesign - roughly 55% of high performers versus 20% of everyone else. In other words, the companies that see earnings impact did not point an agent at an existing process and hope. They rebuilt the process so the agent owns the routine path from start to finish and a person handles only the exceptions. This is the same mechanism behind why so many pilots stall short of value, which we covered in why most AI agents never reach production. The technology was ready; the operating model was not.

Why bolting AI on top produces nothing

Consider what "bolting on" looks like day to day. A team gets an AI assistant, uses it to draft a few emails faster, and the individual saves ten minutes here and there. Those minutes are real, but they never aggregate into earnings because the surrounding process - the approvals, the handoffs, the re-keying between systems - is untouched. The work still flows through the same bottlenecks. This is why broad, shallow deployment produces thin returns while a single redesigned process produces measurable ones. It also explains the abandonment wave we wrote about in the AI project graveyard: programs that spread AI thinly across everything quietly deliver nothing anyone can point to, and eventually get cut.

Why a smaller business is structurally advantaged here

This is the counterintuitive good news. Redesigning a process is hard at a large enterprise because every step touches another department, another system owner, another sign-off - which is exactly why the 6% club is so small. A traditional mid-market company or a small business faces none of that friction. You can redraw an entire process - order intake, invoice handling, quote turnaround - in a single working session, because the people who own each step are in the same room. That structural simplicity is a genuine advantage. It means a well-scoped smaller deployment can reach the workflow redesign that defines high performers far faster than a Fortune 500 transformation program, and at a fraction of the cost, as we lay out in the small-business playbook.

What to do this quarter to be in the 6%

Pick one process, not a strategy. Choose a single recurring workflow with a clear start and finish. Ignore the company-wide "AI transformation" framing - it is what produces the 94% result.
Redesign it, do not decorate it. Map the steps and rebuild them so the agent owns the routine path end to end and a human approves the exceptions. This is the variable that separates the 6%.
Define the number before you build. Hours saved, errors prevented, response time cut. If you cannot name the metric, you cannot land on the right side of the earnings gap. Our payback guide shows how to set a realistic target.
Keep humans at the gates. Anything customer-facing or financial waits for approval. Redesign does not mean removing oversight; it means putting it only where judgment is needed.
Measure, then expand. Run for a few weeks, read the real numbers, and only then add the second process on the same foundation, as described in our back-office automation guide.

The headline of 2026 is not that AI failed to deliver. It is that spending and delivery turned out to be two different things, separated by one habit most companies skipped. For a smaller business without a research department, that is the most encouraging finding of the year: the thing that decides your return is not a budget you cannot match, it is a redesign you are uniquely positioned to do.

Frequently asked questions

Why does record AI spending not show up in earnings?

Because most organizations add AI on top of an unchanged process instead of redesigning the process around it. Only 39% report any profit impact and just 6% are high performers - and the high performers are about 2.8 times more likely to have redesigned a workflow.

What does workflow redesign actually mean for a small business?

Picking one recurring process and rebuilding it so an agent owns the routine path end to end, with a person approving the exceptions. It is one bounded process redrawn, not a company-wide transformation.

Does a smaller company have an advantage in closing this gap?

Yes. Fewer stakeholders and less legacy friction mean a smaller company can redesign a whole process in an afternoon - the exact move that large enterprises struggle to make.

How do we make sure our AI project moves the numbers?

Define the metric first, redesign the one process around the agent, keep human approval on sensitive actions, measure for a few weeks, and expand only once the number has moved.

Record AI spend, no earnings bump: what the McKinsey 2026 gap means for your business