Enterprise AI Returns Concentrate in Organisations That Redesigned the Work
MIT, McKinsey and Accenture reach the same finding independently. The gap lies in the work around the models, which few organisations have changed.
Research Findings
Four findings from three independent studies. Each links to its source.
of organisations report no measurable P&L return from generative AI
Against an estimated $30 to $40 billion of enterprise spend. MIT attributes the gap not to the models but to a learning gap: the difficulty of integrating them into existing workflows. The study draws on 150 leader interviews, 350 employee surveys and 300 public deployments.
report no tangible impact on enterprise-level EBIT from generative AI
Adoption is widespread but impact is limited. Eighty-eight percent of organisations now use AI regularly in at least one function, and only around a third report any enterprise EBIT effect at all, most of it below five percent.
have redesigned any workflow at all, and workflow redesign is the change most correlated with EBIT impact
Of all the organizational changes McKinsey tested, fundamental workflow redesign ranked highest in correlation with EBIT impact from generative AI. Few organizations have made this change.
of companies are ready to scale AI beyond pilots
Accenture finds a small group of front-runners pulling away, and the separation is driven by capability. Companies that scale a strategic bet are close to three times more likely to exceed their AI ROI expectations.
Underlying Finding
Of every organisational change McKinsey tested, fundamental workflow redesign ranked highest in its correlation with EBIT impact from generative AI.
Twenty-one percent of organisations have done it.
The shortfall does not come from the models underperforming. It comes from deploying capability into a process that was not changed, and then being unable to measure what it achieved.
Implications
Four steps, each following from the previous one.
The gap is organizational, not technical
All three studies land in the same place. MIT calls it a learning gap, McKinsey calls it a scaling gap, Accenture calls it readiness. None of them concludes that the technology underperformed. They conclude that the organization around it did not change.
Organizations that achieved a return redesigned the work
Workflow redesign is the single change most correlated with EBIT impact in McKinsey’s data, and only twenty-one percent of organisations have done it. This accounts for much of the gap between the five percent who report real value and everyone else.
Work must be measured before it can be redesigned
Most enterprises cannot state what a process costs today at the level an agent operates at. Workshops produce a target state assembled from what people remember doing, so the redesign often reproduces the problem it was meant to remove.
A return cannot be proven without a baseline
A pilot with no measured baseline cannot be shown to have worked. As a result, many pilots end in disagreement rather than a decision, and later funding does not follow.
Where Pyze Fits
Two of the four steps above are measurement problems. Pyze addresses both, and they determine whether the other two steps are achievable.
Redesign Based on Measurement
What the work costs today, measured across every person and every application, joined to the procedures and the application rules that govern it. The target state is derived from evidence rather than from what people recall.
Return Supported by Measurement
The same measurement method records what changed. Absorbed, assisted and retained work separated per person, released capacity traced to where it went, and the estate re-scored each quarter against the original baseline.
Pyze measures the return realized after deployment.
A business case is a forecast built before anything changes, and the research above largely reflects forecasts that were not revisited. Pyze is designed to measure the result after the change.
How the methodology worksExample Engagement
Order management at a Fortune 500 global CPG
Delivered with a global systems integrator. Work was measured first, the procedures were unified into one standard, and the capacity release was compared against the baseline.
Start With a Measured Baseline
Approximately six weeks from deployment to a priced opportunity register. The baseline is retained so that return is measured against actual results.