There are four ways to buy enterprise AI in 2026. Each of them assumes a customer who exists. Each of them is built for a different customer than the one who has the most to gain.
The four models and the customers they assume
The Big-4 transformation model assumes a customer with an in-house technology function and an offshore engineering capacity to take a strategic recommendation forward. The roadmap arrives at month six; the customer's internal team executes against it. McKinsey, Accenture, Deloitte, BCG, IBM, the firms that have been delivering this engagement shape for thirty years have all built their P&Ls around it. The model works when the customer can execute against the recommendation internally.
The AI boutique model assumes a tech-forward leader who can specify the problem precisely, has internal engineering to maintain the result, and can wait six months for the bespoke build to land. Faculty, Mosaic, Plan A Technologies, the class of fifteen-to-fifty person firms staffed by ML engineers will build something custom. The model works when the customer can specify what they want before they ask for it.
The bolted-on SaaS copilot assumes the work happens inside the application that gets the copilot. Microsoft Copilot for Microsoft 365. Salesforce Einstein. Glean, Writer, Notion AI. Pay a per-seat fee. The model works for documents, decks, and spreadsheets, where the work happens inside one application at a time rather than across the boundaries between applications.
The in-house AI team assumes the customer can hire ML engineers. The market rate for senior AI engineers in 2026 is fully-loaded compensation north of four hundred thousand dollars; the alternative employer for those engineers is a frontier lab that pays more and offers more interesting problems. The model works for the small number of customers who can compete in that labor market, recruit the people the work needs, and wait the eighteen months it takes to ship the first thing.
Each of these four models assumes a customer profile, and each of those profiles excludes the same business.
The business that gets left out of every model
The business that is excluded is running on QuickBooks or NetSuite or a legacy ERP installed in 2007 that the original implementer no longer supports. The business has a CFO, a COO, a head of operations, and a layer of people who do manual work between systems every Monday morning. Invoices get re-keyed from one system into another. Approvals get chased over email. The same spreadsheet gets rebuilt every week to compile the report that the executive team wants. The work consumes a substantial share of employee time, and the executive team knows it does, and has known it for years. The technology to fix it has existed in some form for almost as long. Nothing has shipped, because none of the four delivery vehicles fits.
Why none of the four delivery models fits this customer
This business cannot afford a McKinsey transformation. The price tag for a Big-4 AI engagement starts at half a million dollars and runs up from there, and the deliverable is a recommendation, not the AI. The business has no in-house technology function to execute against the recommendation, so the engagement either ends with an unused recommendation or with another half-million-dollar engagement to start the implementation. Neither option ends with the AP queue processed with one fewer person on the team.
This business cannot wait six months for an AI boutique to spec, build, and hand off a bespoke system. They need the work done this quarter. They cannot specify the problem precisely because the problem sits inside a workflow that nobody has documented at the level of detail the boutique would need. They have no internal engineering to maintain the result once the boutique team rolls off. The bespoke build that arrives at month six will be obsolete by month nine.
This business cannot get value from a horizontal SaaS copilot. The work that needs automating happens between QuickBooks and a spreadsheet and a vendor portal and an email thread. None of those have a copilot, and the copilot for the office suite handles the part of the work that was already easy. The horizontal copilot addresses roughly a fifth of the work and leaves the rest untouched.
This business cannot hire an in-house AI team. The labor-market math does not work. Even if it did, the eighteen-month timeline is unworkable.
Why the AI roll-up does not solve this either
The reasonable counter at this point is that the AI-enabled roll-up strategy will solve this. PE firms (General Catalyst's Creation Strategy, Thrive Holdings, Bessemer, Lightspeed) have committed more than three billion dollars to the playbook of buying fragmented services markets and deploying AI to compress costs. The strategy is real and well-funded and has produced operationally interesting outcomes. It does not, however, address the customer in the middle. The roll-up customer is the rolled-up firm, not the operating business. The customer in the middle is unaffected, while the roll-up happens to a service provider they may or may not have ever bought from. The gap remains.
The size of the unaddressed market in enterprise AI
The customer that is excluded is not a small or unsophisticated segment. Sequoia's services-as-software piece quantifies the labor TAM that none of the four models is currently addressing: insurance brokerage at $140B to $200B, recruitment and staffing at $200B+, supply chain and procurement at $200B+, IT managed services at $100B+, accounting and audit at $50B to $80B (US outsourced), claims adjusting at $50B to $80B, tax advisory at $30B to $35B, legal transactional at $20B to $25B. These are real budgets sitting in real businesses. The Bhargava framing from General Catalyst (sixteen trillion dollars in services versus one trillion in software) is the high-level version of the same point. The largest unaddressed market in enterprise AI today sits between four delivery vehicles that were never designed for it.
The shape of a delivery model that fits this customer is not the subject of this post. The subject of this post is to notice that no such model is currently being offered by the names that show up in the customer's RFP process. Every conversation about AI delivery that opens with "should we go with Accenture or Deloitte" is starting in the wrong frame. Both choices are wrong, in the same way and for the same reason. The model the customer is buying was not designed for the customer buying it.