The Technology Worked. The Value Didn't Arrive
By Brad Ferris · 6 June 2026
This week Bain & Company put a number on something a lot of us have suspected for a while.
Bain surveyed 951 companies, all with revenue above US$100 million, about the returns on their AI programmes. Bloomberg covered the findings on 1 June, and Bain published the underlying analysis under a heading that does most of the work: Your AI Budget Is Growing. Your Returns Aren't.
The numbers are blunt. Per the Bain survey, 40 per cent of companies achieved AI cost improvements of 10 per cent or less. Only 4 per cent cleared 30 per cent. That sits against expectations: 37 per cent of the same firms had banked on cuts in the 10 to 20 per cent range. Bain's summary of the pattern, as reported by Bloomberg, was that the technology worked and the value didn't arrive.
Read that again. These are companies with nine-figure revenues, dedicated IT teams and consulting budgets. The models did what the vendors promised. The savings still failed to land on the P&L.
Where the value leaks out
When a capable technology produces a weak return, the leak is almost always in the deployment, and it tends to happen in three places.
Pilots that never touch the process. A team runs a proof of concept, the demo impresses everyone, and then the tool sits beside the real workflow instead of inside it. Staff use it when they remember. The old process keeps running underneath. You have added a cost line without removing any work.
Savings claimed in slides, never banked in budgets. "This saves each analyst four hours a week" only becomes money when someone decides what happens to those four hours. Redeployed to revenue work, absorbed into faster turnaround, or removed from the cost base: any of those is a real outcome. Left unmanaged, the hours evaporate into the general noise of the week and the saving exists only in the business case.
Nobody owns the number. Ask who is accountable for the return on the AI budget and watch the room. In most organisations the honest answer is nobody. IT owns the tooling, the business units own the workflows, finance owns the spend, and the return falls between them.
None of these are technology problems. All of them are operating problems, which is uncomfortable news for anyone hoping the next model release will fix their numbers, and genuinely good news for anyone willing to do the unglamorous work.
Why a 50-person business should read this differently
The instinct is to look at a survey of US$100 million companies and conclude it has nothing to do with you. I would argue the opposite.
The reasons big companies leak value are structural: long distances between the person who approves the spend and the person whose workflow changes, layers of change management, competing programmes, political ownership of tools. A mid-market business has almost none of that. If you run a 50-person firm, you can baseline a process on Monday, change it by Friday, and see the result in the same month's numbers. The execution advantage sits with you.
The same week Bain's numbers ran, Anthropic launched Claude for Small Business, embedding its models into QuickBooks, HubSpot, Canva, DocuSign, Google Workspace and Microsoft 365. The tools are now arriving inside the software mid-market businesses already run. Access is being commoditised at speed. Which means access will not be the differentiator. Bain's 4 per cent club was not separated from the rest by better software; every firm in that survey could buy the same models. The separation came from how they deployed.
What I would do on Monday
If the survey teaches anything, deploy narrow and bank the result before you scale. In practice:
- Pick one process, not a platform. Choose a single workflow with real volume: quoting, first-draft client reporting, invoice matching, inbound enquiry triage. Resist the enterprise-wide programme.
- Baseline it before you touch it. Hours per week, cost per transaction, turnaround time, error rate. Whatever the honest current numbers are, write them down. Without a baseline you will be in the 40 per cent, guessing.
- Put the tool inside the workflow. If using it requires anyone to remember it exists, it will not survive a busy fortnight. Rewire the step so the AI path is the default path.
- Decide in advance what happens to the saving. Redeployed, absorbed or removed. Name the choice before you start, and name the person who owns the number.
- Scale only what has paid. One process that demonstrably returned its cost buys you the credibility, and the internal knowledge, to do the next three.
The Bain data will get quoted for the rest of the year, usually as evidence that AI is overhyped. That reading misses the point of the survey. The technology cleared its bar. The 4 per cent prove the returns are available. The rest is discipline, and discipline is free.
Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.