Why Most AI Strategies Fail Before the Technology Does

Usage is up. Returns are not.

Almost every organization now uses AI somewhere. Writing in Fortune on October 1, McKinsey’s David Pralong noted that 89% of organizations use AI in at least one function, up from 88% a year ago, while the share earning 5% or more of EBIT from AI has stayed at 6%.

A BearingPoint survey of 1,050 executives, published October 8, found the same gap from another angle. Only 13% scaled AI as planned, and almost three quarters either narrowed their plans or got less than they expected. The technology is rarely the reason. The work around it is.

Start where the work is already measured

Pralong points out that the clearest gains so far came in contact centers and software teams, where processes and metrics were standardized before AI arrived. Begin with a recurring task that takes too much time, creates errors or leaves useful knowledge hard to find. If nobody can say how long it takes today or what a good result looks like, fix that first.

Redesign the workflow, then add the tool

Organizations that redesigned workflows were 5.3 times as likely to report enterprise-level value (32% against 6%), according to the McKinsey research Pralong cites. BCG’s Applied AI Index, summarized by Consulting.us on October 6, makes a similar point with its 70-20-10 rule: most of the effort belongs with people, organization and process, and the smallest share with the algorithm.

Map who performs the work, what inputs they need and where the result goes. Then decide which steps change, which disappear and where a person reviews the output.

Measure growth, not only hours saved

The 2026 AI Adoption Report from Wharton and GBK Collective, released October 13, found that 35% of leaders who track at least one growth measure report strongly positive returns, against 13% of those who track efficiency alone. Hours saved is the most common measure (40%), and net-new revenue the least used (27%).

BearingPoint saw the same pattern. Among the organizations it classed as leaders, 70% tie more than half of their AI projects to financial targets, compared with 34% of the rest.

Run a bounded pilot with a named owner

Give a small team a clear use case, approved inputs and a review process. Record a baseline before you start and compare results with the existing approach, including the new burdens of checking output and training people. If the pilot works, one person should own the redesigned process, not just the tool.