When Every Team Asks the Same Model the Same Question

Enterprise AI carries a quiet risk: the answers may start to look alike. Give a dozen strategy teams the same market facts, prompt templates, vendor dashboards and frontier models, and their outputs will differ in wording while often converging on substance. The recommended markets end up next to one another, the risk lists take the same shape, and customer segments get the same labels. Board decks become more polished as the judgments underneath them grow more alike. I call that convergence the AI herding problem.

Executives spent the past two years testing whether AI speeds up knowledge work. Once it does, a harder question follows. If analysis gets cheaper and more uniform, the advantage shifts to people who can tell when to depart from the consensus the machines produce.

The same tools can make a company's judgment more independent or more crowded, depending on how they are used.

What investors have noticed about correlated models

Goldman Sachs Exchanges recently examined AI's effect on market efficiency, and the argument applies to firms that hold no stocks at all. When many participants rely on similar models, the tools can improve discovery where coverage is thin. They can also crowd positions, because comparable users asking comparable models reach comparable conclusions. Inside a company the same thing happens: routine analysis improves while strategic moves become more imitative.

A model can review more evidence, list more options and flag more anomalies than a tired team. It also standardizes language and normalizes assumptions, and a recommendation that reads cleanly gives the impression it has been tested when it may not have been. So each piece of local analysis gets better even as the organization's decisions drift toward everyone else's.

Adoption is running ahead of management practice. McKinsey's 2025 State of AI report shows companies starting to redesign workflows and address risks, with measurable financial impact still uneven. Boston Consulting Group (BCG) research from 2026 finds only a small share of firms qualify as leaders, and what separates them is talent, workflow redesign, reinvestment and results, far more than how they talk about the tools. Deloitte's enterprise studies describe the same pattern, with many generative AI experiments failing to scale even as spending and interest rise.

How herding enters the company

Herding rarely announces itself. It usually arrives looking like ordinary process improvement.

Template herding starts when a central team issues prompt libraries, scoring rubrics and board formats. These materials help, but they also force varied strategic questions into one frame. Over time, a distinctive local fact starts to register as an exception to the template, when it ought to be a reason to change the conclusion.

Vendor herding comes from enterprise platforms that embed assumptions about how sales, finance or legal review should run. A buyer takes the software and gets a management approach bundled with it. If competitors adopt the same approach, any distinction has to come from the operating model built around the software.

Benchmark herding appears when leaders track what peers are doing because every board asks the same question. Consultants surface patterns across clients and analysts reward visible activity, so effort goes into whatever peers and analysts will recognize as progress.

Confidence herding comes from the tidy structure of model output, with its pros, cons, scenarios and next steps. That surface can hide weak premises. A human view that still shows its uncertainty can lose out to a smoother model-assisted version, even when the human spotted the detail that decides the matter.

Deciding which calls stay with people

The sensible response is to give human judgment more weight where correlated analysis is most dangerous, which a firm can do at any pace of adoption.

The National Institute of Standards and Technology (NIST) AI Risk Management Framework and its generative profile help here, because they push organizations to govern, map, measure and manage their systems. But governance that simply formalizes the same patterns leaves the herding problem untouched. The gap is closed by deciding in advance which calls get AI assistance, which get formal challenge, and which remain under explicit human ownership. Human ownership can still involve the model. It requires a named owner, a record of the dissenting evidence, and a stated reason for accepting or rejecting the machine-supported view.

Five decisions deserve that standard.

Decision	Why AI Herding Is Dangerous	Human-Owned Test
Market entry	Models overweight public evidence and common category language, which can make crowded markets look safer than neglected ones.	What proprietary reason do we have to win that a model using public information would not see?
Customer promise	AI can optimize messaging toward familiar benefits and erase the awkward promise that actually makes the business distinct.	Which customer truth are we willing to state more sharply than competitors?
Risk appetite	Standardized risk registers can make rare, local, or reputational risks look secondary because they do not fit the common taxonomy.	What failure would be uniquely damaging to us even if it scores low in a generic model?
Capital allocation	Model-assisted business cases can converge on the same productivity assumptions, payback periods, and vendor narratives.	What evidence would make us stop funding this even if peers keep spending?
Operating doctrine	Tools can standardize how work is described before leaders decide what work should become more human, more automated, or more controlled.	Which decisions must remain accountable to a person because trust, legitimacy, or strategy depends on it?

Adding a dissent pass before approval

Good AI management adds deliberate review at a few points in the process, and the aim is scrutiny, with as little delay as possible. Before a major AI-assisted recommendation goes to an executive committee, require a dissent pass. One person uses the same evidence to argue for the opposite conclusion. Another identifies what the model could not have known, because the relevant fact is private, political, tacit or recent. Finance separates productivity claims from cash impact. The workflow owner states what changes on Monday if the recommendation is approved.

The extra step can shorten the overall timeline, since approving a polished consensus and later discovering that nobody owned the key assumption usually costs more time. The 2025 CEO study from International Business Machines (IBM) is a warning on this front: leaders are pressing ahead with AI while naming disconnected technology, data architecture, funding balance, incentives and expertise as continuing constraints. Firms pushing ahead without a clear process for making decisions risk spending heavily to reach the same conclusions as their competitors.

Where AI should disagree

A mature organization asks its models for structured disagreement as well as answers. One pass makes the strongest case for acting. A second builds the case against. A third sets out what would have to be true for the recommendation to fail. The accountable owner then writes the final decision in plain language.

The purpose is to stop the company from outsourcing the shape of its thinking, and opposition manufactured for its own sake does nothing for that. A process that cannot explain why it rejected a plausible model answer is weaker than the model it overrode. If the company accepts the answer without stating what it knows that the model does not, it has stopped doing strategy.

A herding check for this week

Take the three most important AI-assisted decisions now moving through the company and attach a one-page herding check to each:

Which parts of this recommendation could any competitor generate from the same public facts?
Which assumptions came from vendor material, benchmark language or standard prompt templates?
What private fact, customer context, operational constraint or founder judgment changes the answer?
What is the strongest case against the recommendation?
Who owns the final judgment, and what evidence would make that person reverse course?

Keep using the tools for speed, breadth and memory. The check exists to find the part of each recommendation that rests on the company's own facts and judgment, and to put a named owner next to it before it reaches the committee.