The workflow fallacy

A pattern I have seen often enough to name it. A company runs a six-month AI pilot, declares it a success, and a year later looks structurally identical to the company that started the pilot. The slide deck shows a 38 percent cycle-time reduction. The quarterly report shows the same revenue per employee, the same gross margin, the same growth rate. Nothing has compounded.

This is not a failure of execution. The pilot worked. It is a failure of altitude.

The workflow fallacy is the assumption that a company is the sum of its workflows, and therefore that improving each workflow improves the company. Both halves of that sentence are wrong. A company is not the sum of its workflows. A company is the set of decisions it makes about which workflows to run, for whom, in what order, at what price. The workflows are the medium through which decisions become outputs. Optimize the medium and you have optimized nothing strategic. You have made the same decisions arrive at the same place, faster.

The reason this matters is that AI's actual leverage on a company is at the decision layer, not the workflow layer. Workflows are bounded. The contract review workflow has a ceiling: the cycle time can drop, the accuracy can rise, the cost per review can fall, but the workflow is still doing the same thing it was doing. Decisions are not bounded that way. A pricing decision can become five times more granular. A hiring decision can incorporate ten times more signal. An inventory decision can move from quarterly to weekly. The shape of the decision changes. The shape of the workflow does not.

Once you see this, the explanation for why so many AI projects feel hollow becomes obvious. They are pointed at the wrong layer. The team is trying to do the same things, faster. Their competitors are starting to do different things, possible only because the cost of judgment dropped.

There are three reliable signs that an organization is stuck in the workflow fallacy.

The first sign is that the AI roadmap is a list of processes. "We are doing AI for support tickets, AI for sales emails, AI for invoicing." Each line item is a workflow. None of them is a decision. Ask the person who built the roadmap which decisions in the company are about to change shape, and you will get a polite blank look.

The second sign is that ROI is measured in time saved or cost avoided. These are workflow metrics. They tell you nothing about whether the company is becoming a different and better company. A real strategic AI project will often look bad on workflow metrics in the first year because it is reorganizing the underlying decision and workflows feel disrupted in the meantime. If your dashboard cannot tolerate that, your dashboard is filtering for the wrong projects.

The third sign is that the org chart is unchanged. AI projects that are doing real strategic work tend to require new accountabilities, because the underlying decision rights are moving. If a year of AI projects has produced no change in who owns what, the projects were workflow projects, regardless of what they were called.

I will tell you how to escape the fallacy in one sentence and then take the rest of the essay to defend it. Pick one decision, not one workflow, and reorganize a year of AI work around making that decision better. Start at the altitude where the strategic value lives.

A decision in this sense is not a single moment of choice. It is a recurring class of choices a company makes that shapes what it produces. "How do we price this customer" is a decision. "How do we route this incoming issue" is a decision. "Which campaigns get more budget next week" is a decision. Each of these is made hundreds or thousands of times a quarter. Each is downstream of a workflow stack that is, currently, the company's main lever on it. Each is about to become, for the first time in industrial history, directly addressable by software.

The instinct, when you tell a leader to pick one decision, is to refuse to pick. They want to do AI everywhere. They have a list of forty processes. The CFO has been asking for an AI strategy and the strategy has to look comprehensive. So they say "all forty, ranked by ROI." This is the moment the workflow fallacy reasserts itself. If you cannot tell which decision you are trying to change, you are not doing strategy. You are running a procurement exercise.

The companies that will compound are the ones that pick the decision deliberately. They will look unimpressive in the first year because they are spending the year reorganizing one decision. They will look unrecognizable in year three because the decision they reorganized was load-bearing and the rest of the company had to bend around it.

This is the unglamorous truth about strategic AI work. It is slower than workflow work. It is harder to measure in the early quarters. It rarely produces a clean before-and-after. It tends to make things look worse before they look better, because reorganizing a decision means breaking some things on purpose. The leaders who can tolerate that are rare. The companies that get it right are rarer.

But they exist. I have watched two of them go through the transition and emerge as different companies, with different growth rates, in markets their workflow-fallacy competitors had assumed were mature.

The good news is that you can choose, today, which side of this you want to be on. You can pick the decision. You can give it a year. You can refuse to be measured, in the first six months, on cycle-time savings.

The bad news is that almost no one does. The pull of the workflow fallacy is strong. It feels productive. It produces nice slides. It leaves the org chart undisturbed.

If your next AI project is a workflow project, fine. Run it. Just do not tell yourself, or your board, that you are doing strategy.