Enterprise AI Has a People Problem
The market measures seats. It's missing the people who matter.
The enterprise AI market measures seats. It should be measuring influence.
A seat count tells you how many people have access. It tells you almost nothing about who set the terms of that access, whose risk assessment decided what was allowed, or who will decide a year from now whether the deployment grows or stalls.
Those answers belong to a group the market has mostly chosen to ignore.
The layer between the user and the purchase
In most risk-sensitive organizations, an AI adoption decision doesn’t start with the end user, and it doesn’t end with IT procurement. It passes through a layer in between. Legal counsel weighs the liability exposure. Compliance sets the boundaries of acceptable use. Governance advisors build the policy structure that makes deployment defensible. The CISO decides which data can never touch which systems. Fractional executives and board advisors frame the decision for leadership before a vendor ever gets in the room.
Call it the advisor class. These people aren’t in the seat count, and they’re rarely in the CRM. Most enterprise AI distribution strategies treat them as friction to manage rather than infrastructure to build.
That’s a serious miscalculation, and the market has already produced evidence of how deep it goes.
Protection priced by headcount
Look at how the major AI vendors handle data protection agreements. A DPA is the contract that governs what a vendor does with your data: whether your inputs train its models, how long they’re kept, and what controls you have. For the chat and workspace products most professionals actually use, that contract typically comes with a business plan, and business plans typically come with a seat minimum.
I’ve run into this myself. To get contractual protection for my own intellectual property, I have to buy a minimum number of seats. My organization doesn’t need them. The vendor’s pricing model simply has no way to protect one person’s IP outside a commercial structure. The protection isn’t scoped to what I put into the system. It’s scoped to how many people I pay to use it.
A privacy toggle in the settings menu isn’t the same thing. A setting can change with the next product update. A contract can be enforced. And while some vendors will sign a DPA for their developer platforms, that helps the engineer building an application. It does nothing for the author working in the chat window.
The problem reaches well past me. Writers are putting unpublished ideas into these tools. Authors are working through arguments that took years to develop. Researchers, strategists and independent advisors are running the frameworks that make up their professional IP. The vendors handling that material have often put basic contractual protection out of their reach. No one weighed the risk and said no. The pricing architecture just never imagined that one person with valuable intellectual assets would need protecting.
This isn’t an edge case. It’s a design failure, and it lines up exactly with the market’s blind spot about the advisor class.
The advisor’s credibility problem
The people shaping how organizations adopt AI are often personally unprotected by the systems they’re evaluating. A governance advisor helping a company design its AI policy may have no enforceable data handling agreement with the vendor whose tool they’re using to build that policy. That’s a personal risk. It’s also a credibility problem for every organization relying on the advice.
We’ve seen this before
Other industries learned this the hard way. Drug makers learned that prescribing physicians, and the specialists those physicians listen to, shape adoption as much as any hospital purchasing office. Security vendors learned that a CISO’s informal veto could kill a deal that had cleared every other gate. The same dynamic now shapes enterprise AI, and most of the market is still optimizing for access while the real decisions get made one layer up.
What fixing it takes
The fix isn’t complicated.
For vendors: decouple data protection from seat minimums. IP protection should follow what goes into the system, not how many people use it.
For go-to-market teams: treat the advisor class as a distribution channel. Build the frameworks they can put their names behind, the governance structure that makes adoption defensible, and the trust that turns resistance into recommendation.
For organizations: bring the advisor class into the adoption decision as a structured input, not an afterthought.
For advisors: use the leverage you already have. Most of you are underusing it.
The vendors who see this first will be built into the influence layer before their competitors notice it exists.
Access is not permission
The enterprise AI market is selling access. The next wave of deployment will be governed by permission. Those are different decisions, and different people make them.
If your organization is working out where its real AI decisions get made, and who should be making them, that conversation starts at book.fellowshipintelligence.com.
