Reflections on the AI-Ready Companies Changing the Game
Only one in four organisations is ready for AI. The other three keep hustling with necessary conditions and hoping for sufficient results.
Necessary conditions you can buy. The sufficient ones you have to build.
Otto von Bismarck, the architect of German unification, is credited with saying: “Only a fool learns from his own mistakes. The wise man learns from the mistakes of others.” Although I have my share of blunders, I have always tried to understand where others failed, particularly when it comes to data projects and AI adoption.
I have written about this before, most recently analysing failure data released throughout 2025 and early 2026 by Gartner, MIT Sloan, RAND Corporation and McKinsey. Several of you wrote back with the same fair question: fine, so what works? I want to pick up right there, with the help of another giant of the industry: IBM. With numbers attached.
The IBM Institute for Business Value surveyed over a thousand C-suite executives and another 8,500 employees and consumers for its AI readiness research and the Five Trends for 2026 report. The headline finding: only one in four organisations is AI-ready.
Here is the part I find delicious. IBM’s own researchers, after all that surveying, landed on a sentence I could have written myself: “AI readiness is as much a mindset as a set of capabilities.”
The companies in this league are ten times more likely to deploy AI enterprise-wide. Ten times. That is not an edge. That is playing a different game.
Let that sink in. The largest enterprise technology company on the planet studied AI readiness and concluded that the differentiator is not technology.
I am a mathematician, so let me bring in a concept every first-year student learns: the difference between necessary and sufficient conditions. Necessary means nothing works without it. Sufficient means it works because of it.
The models, the data, the engineers, the architecture: all necessary conditions, every bit of them. The sufficient conditions live somewhere else. That distinction is what this piece is about.
Bringing forth the necessary conditions: the part I love
Nothing here diminishes the technical bar. IBM’s respondents named underpowered infrastructure, fragmented data and weak governance as their top barriers, and the projection that AI-ready IT systems will grow from 26% to 45% by 2026 still leaves more than half of organisations short.
Respondents also pointed to the importance of efficient, innovative machine learning and agentic AI. The AI-ready companies have mastered how to attract and keep top-notch mathematicians, statisticians, data engineers and data architects in their central data functions, and how to either develop or incorporate innovative data solutions.
This is what produces clean pipelines, high data quality and defensible governance. Skip it and nothing downstream matters.
These are all necessary conditions. None is sufficient.
Enabling the sufficient conditions: the ones hardly named
If you’ve read my writing for a while, you know I called my conditions the Three Ps two years ago. IBM just ran the experiment at scale, and the Three Ps came back wearing a lab coat. Let me restate them with precision.
Prowess. Deep STEM craft at the central data function, and deliberate skilling of everyone around it. Not just hiring the ten best engineers you can find, but raising the data fluency of the hundreds of people who will work alongside what those engineers build. IBM found the readiness leaders invest in exactly this: data skills for the wider pool of talent, the people without formal training in data who nonetheless consume it, transform it and deploy decisions on it every day. These are the unsung heroes who made Microsoft Excel a ubiquitous corporate presence. Give your people data prowess as a first-class capability, not a training afterthought.
Processes. Mapped, optimised, and ready to change. The silent killer in most organisations is what I call frozen Fordism: workflows designed for an assembly-line century, hardened into software, defended by habit. AI lands on these rigid processes and shatters them. IBM’s strongest correlation was between readiness and organisations whose IT strategy and business strategy are the same strategy. That alignment is only possible when processes are allowed to move.
People. Eager to build. Convinced that done is better than perfect. Learning from outcomes instead of litigating them. Collaborating inside safe spaces where a failed experiment is data, not a career event. This is not poetry; psychological safety is the operating condition under which the other two Ps compound.
Readiness, in other words, is a posture, not a purchase. You cannot buy it in a procurement cycle. You build it the way you build physical posture: deliberately, unglamorously, before you need it.
The flywheel
Strip away the frameworks and the whole game reduces to one loop, and it is a deeply mathematical one.
Read the data from your environment. Act on it to the best of your current ability. Gather the data your own actions generate. Feed it back. Act again, slightly better.
That is the flywheel of learning, and it is the only mechanism I have ever seen produce compounding success with data and AI. Every element of readiness exists to spin it faster: prowess reads the signal, fluid processes let you act on it, and people who learn from outcomes close the loop. Organisations that master this don’t need to predict the future correctly. They out-learn everyone who tried to.
The ROI paradox, or why the ready don’t count beans yet
This part of IBM’s research will make your CFO uncomfortable. About two-thirds of respondents, across functions and readiness levels, say proving ROI is not a top priority through 2026. The ready organisations are pouring effort into adoption and use cases instead.
On the surface this sounds like fiscal negligence. In practice, it is not. The flywheel I mentioned earlier explains why: the AI-ready organisations are buying revolutions of the loop. Demand twelve-month payback on a learning mechanism and you stall it before it compounds. IBM’s warning is blunt: treat AI as a cost center and you may never catch the competitors who took the longer view.
I partially agree with that take, but I want to be precise here, because this point gets abused. This is not a licence to skip measurement. My last piece showed what happens when projects launch with no agreed definition of success: 73% of failures.
The discipline is to define success before you build, and to measure capability gained, not just euros returned in the first year. The AI game is exponential, remember that. If you are doubling capability with every iteration, then reaching 1% of your goal means you are almost done: you sit just seven doublings from the finish line.
There is a canyon between “we don’t measure” and “we measure the right horizon.”
Trust is the entry fee
One more finding deserves a place here, because it confirms something I’ve repeated until my colleagues roll their eyes. Among consumers, 89% want to know when they are interacting with AI, and 80% would trust a brand less if it hides its use. Two-thirds would walk away entirely if they feel deceived.
Customers will forgive your AI for being imperfect. They will not forgive you for lying about it. Transparency is no longer a compliance checkbox; it is a commercial variable with a measurable price. For those of us operating under the EU AI Act, this is quietly excellent news: the regulation and the customer now want the same thing. Start with governance. Always. It stopped being just the safe choice; it is now also the profitable one.
So, do you want in?
Strip this piece down and it says: the AI-ready club has no secret handshake. The entry requirements are public. Necessary conditions: real STEM prowess, clean data, governed platforms. Sufficient conditions: processes free to change, people safe to build and learn, leadership willing to be a student again, honesty with your customers about the machines. Then spin the flywheel and let compounding do what compounding does.
Nothing on that list requires a frontier model. Everything on that list requires executive will.
The companies that join the AI-ready club in 2026 will spend the next decade compounding their advantage, because readiness, like posture, pays off on every movement that follows. The companies that don’t will keep buying impressive demos, overthinking complicated POCs, chasing single-project ROI to justify company-wide AI adoption, and wondering why the returns never arrive.
We already know that story. It ends at the 80% graveyard.
Sources: IBM Institute for Business Value, “AI Readiness” research (1,204 respondents, 10 countries) and “Five Trends for 2026” (1,000+ C-suite executives, 8,500 employees and consumers). Failure-cause figures referenced from Part 1, “The Secret Sauce Is Still a Secret.”
If you are working on getting your organisation into the AI-ready 25%, I’d be glad to compare notes. Find me on LinkedIn.