Viewpoints

The AI return is missing because the work hasn’t changed

Richard White ·

Firms are delaying AI spending because they can’t see the return. The models aren’t the problem. Most companies added AI to the old job instead of changing the job, and the return lives in the change.

What the numbers say

The gap between AI spending and AI results is now showing up in budgets. Forrester predicted last year that enterprises would defer a quarter of their planned AI spend into 2027. It found that fewer than a third of decision makers could tie AI to their organization’s financial growth, and that only 15% reported an AI-related increase in earnings over the previous year.

New research from MIT points the same way from a different source. Researchers at the MIT Initiative on the Digital Economy and MIT FutureTech read ten years of annual filings from S&P 500 companies. By the end of 2025, just 11% had AI deeply embedded in their business, while 45% were still running pilots. The early adopters often showed lower margins and productivity at first, which the researchers suggest may be a lag before the benefits arrive.

So the money is being spent, the tools are in place, and most of the value hasn’t appeared.

My view

This isn’t a technology problem. The models are good enough for a great deal of real work today. What’s missing is the change around them.

Most firms bought AI the way they buy software. They licensed a tool, ran a launch, trained people on the buttons and counted logins. The job itself didn’t change. The examiner still prepares the case the old way and now checks an AI summary as well. The analyst still builds the report and now reviews a draft too. Each person is doing the old work plus a new review step, and the firm is paying for both. Then the return shows up as a cost.

The return from AI comes from rewiring the work: deciding which steps a person no longer does, who reviews what the AI produces, what the new standard is and how fast someone can learn it. That’s change management in the plain sense. It’s owned by the business, not by IT, and it rarely appears in the AI budget.

Why it matters

For the industry. Expect 2026 budgets to separate firms that changed the work from firms that bought tools. The second group will cut, pause or restrict, and some already are. The first group will be able to show a number, because they decided in advance what would change and measured it. The deferral isn’t a verdict on AI. It’s a verdict on funding tools without funding the change.

How it connects to our work. The reasons the return goes missing are predictable, and each has a fix:

  • Nobody owns the outcome. A pilot with a sponsor but no business owner can show activity, never results. Name the person who owns the workflow and the number, before anything is built. Why AI stalls after the demo is about exactly this gap.
  • The job wasn’t redesigned. If review is added on top of the old process, effort goes up. Redesign the work, fund the review time and train managers to run the new routine. That’s the core of Adoption by Design.
  • There’s no baseline. Without the cost and time of the work before AI, there’s nothing to compare against. Measure one workflow first, then change it.
  • The answers need cleaning up. When an assistant doesn’t know what the firm means by a customer or an open claim, people spend their saved time correcting it. Give it approved meaning before it answers. See Context Before Query, or try demo 1.
  • Cost is counted per license, not per task. A seat price says nothing about value. Cost the task, including review and rework, and compare it with the old way. AI Token Economics shows how.

What I’d do in the next 90 days

  1. Pick one workflow that matters, with one accountable business owner.
  2. Measure it as it is today: time, cost, error rate and review effort.
  3. Redesign the work before choosing the tool: which steps go away, who reviews and what the new standard is.
  4. Run it with real users and real cases, and count accepted results, not logins.
  5. Decide on the evidence: expand it, fix it or stop it.

A firm that does this for one workflow learns more than a firm that rolls a tool out to everyone. The return is in the rewiring.

Sources

AI adoptionChange managementAI economics

Drafted with AI research assistance and reviewed by Richard White before publication. Linked sources are the record; check them for the full reporting.