AI by accident, or AI on purpose

Every CEO I talk to is being pitched AI. Most have tried something. Most of what they tried is not working, and they cannot tell whether that is because the technology is overhyped or because they picked the wrong thing to try.

It is almost always the second one.

Two companies in the same industry will buy comparable tools in the same year. A year later, one of them has changed how it operates and the other has a subscription. The difference is rarely budget and it is rarely talent. It is whether anyone decided, in advance, where AI was supposed to pay.

The noise is the problem, not the technology

“Our competitors are doing AI. We need to figure out how to as well, but where do we even start?” That is what I hear in the first five minutes, and it is not the question of someone who has not been paying attention. It is the question of someone who has been paying close attention to a very loud room – a new capability every week, a new vendor in the inbox every day, and no clear signal about what actually moves the needle for a business their size.

The numbers say the same thing. RSM surveyed 966 middle-market executives for its 2025 Middle Market AI Survey and asked how prepared they were when they adopted AI. Fifty-three percent said they were only somewhat prepared. Another ten percent said they were not prepared at all.

That is not a story about companies being slow. Nearly two thirds of them moved anyway, because the alternative was standing still while competitors moved. It is a story about companies moving without a framework, which is a different problem and a fixable one.

Two patterns

Watch enough of these and you stop seeing individual decisions and start seeing two patterns. I call them AI by accident and AI on purpose, because the distinction really is intent rather than sophistication.

AI by accident looks like this. Tools get bought because a vendor was convincing that week. Pilots run, demo well, and never scale past proof-of-concept, because no one ever said what scaling would be worth. The technology strategy ends up owned by whoever sold you the last thing, since leadership never claimed it. Every new announcement produces a small amount of anxiety and no decision. And a year on, the operational drag is exactly what it was, plus some licenses.

AI on purpose is less exciting and works better. Capabilities get deployed against named business priorities, because someone did the work of deciding which priorities AI actually touches. Implementations are tied to outcomes you can measure, so it is possible to be wrong and to find out quickly. Leadership owns the strategy, which mostly means leadership owns the sequencing. And because there is a framework, the next capability that arrives gets evaluated in an afternoon rather than becoming another open loop.

One of those compounds. The other just costs more.

What is actually on the other side

Technology is never the destination. It is worth being specific about what the destination looks like, because “we did AI” is not an outcome.

Decisions get made faster, because the information is available when the decision is in front of you rather than at month-end or after somebody does a manual pull. Your best people spend their time on judgment, relationships, and the problems that need a person, instead of compiling data and rebuilding the same report. Revenue grows without headcount having to grow at the same rate, because the routine work runs on its own and growth stops being purely a hiring problem. And when the next capability lands, you evaluate it against a framework you already have instead of reacting to it.

How the work runs

Most consultants deliver the map and move on. The map is the easy part. The guide is what makes the difference between reaching the destination and getting lost in the weeds, which is why the engagement model has three steps rather than one deliverable.

It starts with a conversation. Thirty minutes on your workflows, your team, your revenue model, and your priorities is usually enough for me to say which AI capabilities matter for your business and which are noise. That call is free, and it carries no vendor bias, no technology shopping list, and no obligation to go further.

If the opportunity warrants it, the next step is the Technology Roadmap Workshop – it produces a business decision document rather than a technology document. Where AI pays in your operation and where it does not, the data and process gaps that would block each opportunity, and a sequenced roadmap you own outright. It is $2,499, credited in full toward a retainer if you engage.

Then there is seeing it through. Technology initiatives rarely fail because the technology was bad. They fail because change management, alignment, and sustained direction were absent. That is what the fractional CIO and CTO retainer is for – ongoing technology leadership including your AI roadmap, with specific automations run against specific processes, and outcomes you can point to inside the first engagement. Three levels of presence, from $4,500 a month.

Why I can say this

I spent twenty years as a technology executive at a global investment manager, running the business-side systems across seven offices with a forty-person technology team and a $5M annual budget. The AWS migrations we ran cut cloud costs by thirty percent, measured rather than estimated. One CRM implementation covered all seven offices.

None of that transfers directly to a company of fifty people, and I would be suspicious of anyone who claimed it did. What transfers is the judgment about sequencing – which problems are worth solving first, which are symptoms of something upstream, and which are expensive distractions that look urgent. That is the part that is hard to buy and hard to hire at your scale, and it is most of what I do now.

The companies that pull ahead over the next couple of years will not be the ones with the most tools. They will be the ones that decided where AI changes the economics of their specific business, and then deployed it there. Intentionally. On purpose.