Research
The memory gap
A research report on why AI deployment keeps failing. Most of what it recommends is work a company can do on its own.
The figure from the report's cover: seven links, with the connection between the fourth and the fifth missing.
Nearly nine in ten companies now use AI regularly in at least one part of the business, but only 37% report any effect on earnings, the same share as the year before.
95% of enterprise GenAI pilots are showing little to no measurable impact on profit and loss.
In 2025, 42% of companies abandoned most of their AI initiatives, up from 17% the year before.
Inside the report
The report takes the problem from the evidence to what you can do about it.
Chapters one and two describe the problem and give the evidence for it. Chapters three, four, and five publish the lens we use to examine it, in enough detail that you can apply it to your own company. Chapters six and seven cover what the evidence says to do about it. The last page says who we are.
Most companies are seeing little real benefit from AI
MIT, BCG, S&P Global, McKinsey, Gartner, and KPMG each measured what happened when companies tried, with different samples and different methods. Their findings agree with each other.
The tools do not learn the business
The pilots failed the way they did because what a company knows is scattered, versions conflict, some of it is stale, and everything carries the same weight. A capable tool reads all of it and returns superseded and conflicting answers with the same confidence as current ones.
Seven links decide whether knowledge reaches the point of use
The chapter goes through each link and the way it fails, and it draws the loop the links form. A search tool, however good, works on only one of the seven.
The other half of the problem is how the workflow runs
Knowledge only matters inside a piece of work, so the chapter asks four questions about the workflow itself: whether there is one way it runs, what moves between the steps, what numbers exist about it, and whether anyone ever decided who does each task.
The right target is level three
The five levels are given by the sentence you would hear inside the company. The chapter also makes the case against aiming for the top of the scale.
AI pays back first in the same few places
Customer service, back-office document work, drafting against known rules, and internal question answering are the four categories with a measured record behind them. The chapter also gives the five things the successful deployments have in common.
Most of the first work needs no outside help
The chapter gives the method for reading one workflow with your own people, in enough detail to act on, and a worksheet that works on paper. It ends with what to fix first and what to refuse to buy.
About this report
The last page says who we are and how every figure in the report was verified.
Sources
Every figure is credited to its source on the page where it appears.
None of the numbers originate with us. Where we could check a figure against the primary document, we did: the McKinsey 2026 report, the Gartner data-readiness release, the Gartner agentic-AI release, the KPMG trust study, ISO 30401, and the NIST AI Risk Management Framework were verified against their primary publications. The remaining figures were verified across multiple independent published accounts of the named study, and the methods note in chapter eight discloses which is which. The lens in chapters three through five is ours, and the same note states that its descriptors have not been calibrated against a statistical sample. The worksheet in chapter seven works on paper, and you can copy it freely.
Next step
The free assessment applies the chapter seven method to one of your own processes.
Pick one process that is costing you something, fill in a short form, and spend thirty minutes on a call with us; within two business days you get a written assessment of where that process stands and what we would do first.