AI can deliver a real return — but most of the spending isn't capturing it. This is an honest look at where the returns actually come from for a small business, what realistic results look like, and how to measure impact instead of chasing headlines.
The biggest, most reliable returns come from time saved on structured, repetitive work — the back-office tasks that quietly eat your team's hours. Free up that time and you redirect people toward higher-value work, which for a small business is often the difference between hiring another person and getting more from the team you already have. MIT's Project NANDA found the same thing at the corporate level: the highest-ROI AI deployments are in back-office automation, not the customer-facing tools that attract most of the spending.
The ceiling is real when the work is redesigned, not just automated. A 2026 Harvard Business Review analysis by a Bain and OpenAI team reported that companies which rebuilt workflows around AI saw 10–25% EBITDA gains — while bolting AI onto an unchanged process tends to produce activity, not profit.
Be skeptical of "10x" promises. The honest picture is meaningful, compounding gains on the right tasks — not a transformation overnight. MIT's Project NANDA found that 95% of organizations report no measurable P&L impact from their AI investments — not because the technology doesn't work, but because effort gets spread thin instead of going deep where it pays. Harvard Business Review calls a version of this the trap of optimizing isolated tasks without rethinking the workflow around them.
There's a subtler trap worth naming: AI doesn't automatically hand time back. In a 2026 Harvard Business Review study of a roughly 200-person company, researchers Aruna Ranganathan and Xingqi Maggie Ye found that AI tends to intensify work rather than reduce it — people take on more, multitask more, and let work seep into the day's quiet moments. As one engineer in the study put it, you "just work the same amount or even more." The return is real, but capturing it as genuinely freed-up time takes intention, not just a tool.
Pick a process, measure the before — hours, error rate, cost — and measure the after. Tie the result to the P&L. If you can't draw a line from an AI project to a number that matters, that's usually a sign the project was aimed at the wrong part of the business.
Let's find the one or two areas where AI pays off first.
Let's TalkThe question I open every engagement with: a two-minute read on the difference between seeing ahead and reacting.
The use cases that work — and the ones that waste money.
The failure rate is high — and the cause is rarely the technology.
A step-by-step way to get a first win without boiling the ocean.