The companies I talk to are not short on AI activity. Most of them have a Copilot pilot running somewhere, a few teams using ChatGPT in ways nobody formally approved, a vendor or two who walked through the door last fall with a deck about transformation, and at least one executive who read something in the Journal and asked a question at the last leadership meeting that nobody was ready to answer.
What they are short on is AI progress. The gap between how much is happening and how much is actually moving is the most consistent thing I see across middle-market companies right now. By the time a CEO calls me, they usually have a word for it already: spinning our wheels. We have been talking about this for twelve months and I cannot point to anything that changed.
Five patterns make up almost everything I see. None of them are unusual, and none of them mean the company is unsophisticated. They are the predictable failure modes of trying to move fast in a market that is moving faster.
The most common version of this goes like the following. A senior leader attends a conference or takes a vendor meeting. The vendor has a demo. The demo is impressive. The vendor has a roadmap for what implementation looks like, a pricing structure, and three case studies from companies that look like yours. The next thing you know, you are in a procurement conversation before you have asked whether this is the right problem to solve.
Vendor-led middle market AI strategy is not malicious. The vendors are doing their job. The problem is that a vendor's roadmap is designed around what they sell, not around where your highest-leverage AI opportunity actually is. A process automation vendor will find an automation opportunity. A data vendor will find a data problem. A generative AI platform will find a use case for generative AI. They are not lying to you. They are solving the part of your problem that their product can solve.
The companies getting this right start from the other direction. Before any vendor is in the room, they have a working answer to the question: where in this business does improving speed, accuracy, or unit cost by 30 percent change a number that matters? That answer comes from your operators, not from a sales deck. Once you have it, vendor evaluation becomes a procurement exercise instead of a strategy exercise, and you hold the leverage.
This one requires honesty. Most middle-market companies running AI experiments are running too many of them. Six pilots across four business units with three different vendors and no shared definition of what success looks like is not a portfolio strategy. It is a hedge against having to make a decision.
Pilots are comfortable because they are bounded. Nobody loses their job over a pilot that does not work. The uncomfortable truth is that a pilot that does not produce something in production is not a learning; it is a sunk cost with a debrief attached.
The pattern I keep seeing is that companies run pilots in sequence or in parallel, never move one to production, and then conclude that AI does not work for their business. The problem is almost never the technology. It is that no one is accountable for getting from proof-of-concept to production workflow. That is a change management problem, not an AI problem, and it requires someone senior enough to move an organization, not just someone technical enough to build the thing.
The companies getting this right pick one thing, define what done looks like before the project starts, and do not start the second project until the first one is in use by real people doing real work. One working deployment is worth ten impressive demos.
This pattern sounds like a quirk of large enterprises, but I see it constantly at the middle market. AI strategy gets handed to IT, or to the data team, or to whoever hired the most recent data engineer. The executive who gets the AI assignment treats it as an infrastructure project, because that is what IT projects look like. And then everyone is surprised when twelve months later there is a lot of new tooling and no business outcome.
AI is not an IT project in the same way that pricing strategy is not a finance project. Finance does the math. Someone who understands the business makes the call. The AI version: technology does the building. Someone who understands the business decides where it matters.
The leader who should own the AI agenda at a middle-market company is almost never the CTO. It is the CEO, or the COO, or the executive who owns the P&L line where AI has the most leverage. In most mid-market companies that means someone is running AI strategy who did not sign up for it, does not have the budget authority to make it real, and is measured on something else. The structural fix is simpler than it sounds: make one person accountable for the business outcome, give them the authority to make tradeoffs, and make sure they report to whoever approves the budget.
This is the shadow AI problem, and it is further along than most leadership teams realize.
By the time a CEO starts thinking seriously about AI strategy, their employees have usually been using AI tools for a year or two already. Customer success teams using ChatGPT to draft responses. Finance analysts using Copilot to clean up spreadsheets. Engineers using Cursor or GitHub Copilot on production code. Marketing teams using image generators. Sales teams using AI notetakers that are recording every call.
None of this is necessarily bad. Most of it is good. People found tools that made their jobs easier and used them. The problem is that you cannot build on what you cannot see. If you do not know what your organization is already doing with AI, you will spend money solving problems that are already solved and miss the gaps that actually need attention. You will also run real risk: customer data in tools that are not covered by your vendor agreements, IP in training pipelines you did not consent to, compliance exposure your legal team does not know about.
The map comes first. Before strategy, before vendor selection, before any new spending, you need an honest inventory of what is already running. Not a survey with leading questions. A real conversation with operators in each business unit about what tools they are using, on what data, for what purpose, and whether it is working. Most companies find that more is already happening than the leadership team knew, and some of it is surprisingly far along.
The fifth pattern is the one people are least likely to name out loud. Many middle-market companies are waiting. Not intentionally. Not with a plan to start in Q3. They are waiting the way organizations wait when no one has made a decision: scheduling more discovery, asking vendors to come back next quarter, forming a working group that meets monthly, adding AI to the next strategy offsite agenda.
The case for waiting is not unreasonable. The technology is moving fast. What is true today may not be true in six months. The costs are coming down. The models are getting better. Why commit to a platform or an approach that might look outdated in a year?
The problem with this logic is that the cost of waiting is not zero, and it compounds. Your competitors are not all waiting. The talent market for operators who have actually shipped AI projects is thinning as those people get absorbed by companies moving fast. And the organizational capability you need to use AI well, clean data, documented processes, a leadership team that understands the tradeoffs, takes time to build regardless of when the technology is ready.
The companies getting this right are not moving recklessly. They are making one concrete bet, learning from it, and using what they learn to make the next bet better. The path from "we are thinking about AI" to "AI is part of how we operate" is not a planning exercise. It is a series of production deployments, each one smaller and less glamorous than the deck you showed the board, and each one building the capability to do the next one faster.
If any of these patterns sound familiar, the starting point is the same in almost every case: build the map before you build the strategy. Know what is already happening in your organization, where the highest-leverage opportunity is, and who owns the outcome. That takes a few weeks and a few honest conversations. It is not glamorous work. It is the work that makes everything else go faster.
The AI Opportunity Sprint is the two-week version of this exercise. It produces the map, the ranked opportunity list, and the 90-day action plan your team can actually execute. If you are reading this and recognizing your organization in one of these five patterns, it is a reasonable next step. If you want to think through where you are before committing to anything, a scoping call takes twenty minutes and costs nothing.
Twenty-minute scoping call. No slide deck, no pitch. We talk about where you are and whether a Sprint or a Fractional engagement fits.
Book a scoping call