Know What You're Buying - The Four Types of Enterprise AI, and Why Each One Is a Different Negotiation

// AI PROCUREMENT SERIES · PART 1 OF 8

The Four Types of Enterprise AI, and Why Each One Is a Different Negotiation

By Chad Johnson · Two decades running enterprise procurement, including CPO roles in Fortune 20 healthcare. Now SVP, Procurement Advisory at WNS Procurement.

Most enterprises treat AI as one thing to buy. A budget line, a category, a box on the technology roadmap. That single-category habit is the first mistake, and it is the one that produces the wrong contract, the wrong pricing model, and a return-on-investment conversation procurement cannot win six months later.

There is no such thing as buying "AI." An enterprise buys four distinct things, and each behaves like a different category with a different cost structure, a different risk profile, and a different negotiation. The evidence for treating them separately is already on the invoice. Blended AI token prices have fallen sharply year over year, by most estimates more than half, yet a majority of enterprises still report their AI costs came in over budget, and average enterprise AI budgets have climbed several-fold since 2024. Prices are falling while bills are rising. That only makes sense once you see that the four types behave differently and get bought differently.

The mistake that starts all the others: treating four different purchases as one category. The vendor across the table already knows which one they are selling you. Getting the type right is the first move that puts you back in control of the deal.

The AI Spend Lifecycle. This piece is part of an eight-part framework for bringing AI under the same procurement discipline as every other major category: buy it right, manage it as a category, fund it against the plan. The eight: (1) the four types of AI you buy, (2) the embedded AI in your renewals, (3) negotiating the token deal, (4) what belongs in every AI contract, (5) AI as your largest unaddressed spend category, (6) matching models to tasks, (7) why AI savings never reach EBITDA, and (8) governing AI spend against the operating plan. You are reading Part 1.

THE FOUR TYPES AT A GLANCE

→ Direct infrastructure. Token-priced model access. Negotiate the rate structure, not a license. Watch the committed volume.

→ AI-powered applications. A product with AI inside. Due-diligence the hidden model, and gate it on a 90-day finance-validated ROI.

→ Agentic systems. Autonomous, multi-step work. Negotiate on outcomes, and define the outcome in writing before you sign.

→ Embedded AI. Switched on in your renewals. The negotiation is your renewal, and the questions are about rights, not features.

Here are the four, and what changes about the negotiation for each.

This is buying raw model access directly from a provider: the large language models themselves, priced by the token. It looks like a simple usage-based software deal on the surface, and underneath it behaves like a cloud-compute contract, which is a very different animal.

The trap is that token pricing is unfamiliar terrain for most procurement teams, and unfamiliar pricing is where margin hides. Two features of the pricing catch buyers out. First, output tokens cost far more than input tokens, commonly three to ten times more, because generating text takes more compute than reading it, so a workflow that produces long outputs has a completely different cost profile than one that produces short ones. Second, reasoning-capable models bill for hidden "thinking" tokens that never appear in the visible answer, and on a hard task those can outnumber the visible output several times over. Consumption also compounds at scale in ways that stay invisible during a pilot. A workflow that costs a rounding error when three people test it can generate a serious monthly bill once it runs in production across the organization.

The scale of the difference between models is the part most budgets miss. Across major providers, the spread between a budget model and a frontier model runs from roughly 10 times to more than 100 times on a per-token basis. Frontier reasoning models reach $15 to $75 per million tokens and beyond, while efficient models sit under $1. Paying frontier rates for work a cheaper model would handle identically is the single most common source of overspend, and the contract structure either exposes that lever or hides it.

What changes in the negotiation: you are negotiating a rate structure, not a license. That means itemized pricing that lists input, output, and reasoning tokens separately, volume commitment tiers with defined discount thresholds, prompt-caching discounts written into the contract rather than assumed, and usage caps with the right to renegotiate if you exceed them. Treat the committed volume itself with suspicion. The most common 2026 failure is committing to a token budget that looks generous, then blowing through it months before renewal as adoption spreads, so the commitment has to come with the right to rebalance, not just a bigger number. It also means converting a price-per-million-tokens figure, which is operationally meaningless to a CFO, into a cost per unit of work the business actually does. Paying a rate per token tells you nothing. Paying a defined amount per contract reviewed, or per document processed, is a number finance can track.

Ask before you sign: can we express this deal as a cost per unit of work, not a cost per million tokens? If not, nobody in finance can govern it.

From the operator's chair. I was once handed a member communications platform to renegotiate, priced per member per month. PMPM is the industry-standard model, so it was the default nobody questioned. The problem was that we were managing to conversion, engaged users, and care gap closure, and PMPM tied to none of them. The cost scaled with membership while the outcomes we cared about sat entirely outside the meter. By the time it reached me there was no leverage to fix it, because the exit and the outcome-tie were never built in at signature. The only way out was moving to another provider. Token deals carry the same trap, committing you to a base that has nothing to do with the outcome, but with one difference worth knowing: with tokens you can change what you consume tomorrow through routing and caching, without touching the contract. PMPM gave me no such lever. The lesson that carries across both is the same. The standard commercial structure is the one nobody negotiates, which is exactly why it is the one that costs you.

This is a finished product with AI inside it: a contract-review tool, a spend-analytics platform, a sourcing-automation solution. You are not buying raw compute. You are buying a solution built on top of someone else's model, and the underlying AI cost is buried inside the vendor's margin where you cannot see it.

These deals feel familiar, which is exactly the risk. They look like ordinary software, so they get negotiated like ordinary software, and the AI-specific questions never get asked. Two of those questions matter more than the price. Which model actually powers this, and who controls that relationship? And what is the real value of the AI feature, separate from the software it is wrapped in? If the vendor's own model provider changes terms or pricing, your product changes with it, and you have no seat at that table unless the contract gives you one.

What changes in the negotiation: you treat it as software, and you add the AI due diligence a normal software deal skips. Ask the vendor to disclose the underlying model and who owns that dependency. Establish a baseline of the current-state process before go-live, so you can prove the tool actually improved something rather than taking the vendor's word for it. Tie payment to performance at go-live, not just to software delivery. And require a return-on-investment review with finance at 90 days, before anyone declares victory. This last point is not academic: by MIT’s 2025 estimate, roughly 95 percent of enterprise generative AI pilots never produce a measurable profit-and-loss return, and the ones that do almost always started with a baseline nobody could argue with.

This is the fastest-growing and least understood of the four. Agentic systems do not wait for a human instruction at each step. They take a goal, break it into steps, execute those steps across multiple systems, and complete a workflow on their own. An agent that runs a sourcing event end to end, or onboards a supplier, or processes invoices from receipt to payment, is agentic AI. Adoption has already left the pilot stage: one major vendor reported more than 29,000 agentic deals closed since launch in its most recent fiscal-year results, other large software vendors report hundreds of thousands of custom agents in use, and several have restructured their commercial models around autonomous tiers.

The commercial model is genuinely new, and that is what makes it dangerous to negotiate. The pricing model is fragmenting into per-seat, per-token, per-task, and per-outcome structures at once, with outcome-based pricing, charging only when the agent completes a defined result, as the notable new entrant. Real examples are already in market: about $0.99 per resolved support conversation from one vendor, $1.50 per automated resolution from another. Outcome pricing sounds buyer-friendly, and it can be, but it shifts the whole negotiation onto the definition of the outcome. If "resolved" is defined loosely, you pay for work that was not really done.

Two traps sit underneath the pricing. The first is what practitioners call agent-washing: existing chatbots and scripted workflows rebranded as agents, priced like autonomy but delivering none. The second is cost behavior under failure. A single failed step with automatic retries can multiply that step's cost several times over, and in a workflow with five or six system integrations, those retries compound. Cheaper per token rarely means cheaper per outcome once retries and growing context are counted.

The tell that a deal is not ready: if the vendor cannot define what a successfully completed workflow looks like, in writing, before you sign, they are telling you the outcome they are selling is not one they can stand behind yet.

What changes in the negotiation: you negotiate on outcomes, and you pin the outcome down precisely. Require a measurement framework agreed before signing, with a documented baseline, defined go-live criteria, and finance validation at 90 days. Define the successful outcome in the contract, including what does not count and does not get billed. Insist the scope of what the agent may and may not do lives in the contract, not just in a configuration screen someone can quietly change. Require human checkpoints for any action that commits money, selects a supplier, or touches a regulatory obligation. And build a pilot-to-production gate, so full funding releases only after the system proves itself on real work, not on a staged demonstration.

This is the type most procurement teams are not watching, and it is the one most likely to bypass procurement entirely. Your existing software vendors, the ones already running your productivity suite, your CRM, your service desk, are adding AI to products you already own. It arrives as a new module in a renewal, a terms-of-service update you clicked through, or a feature switched on by default after a product announcement.

The proof is concrete and current. In mid-2026 one major productivity vendor raised commercial list prices across its suites by roughly 5 to 43 percent depending on tier, explicitly tied to embedded AI, and folded its baseline AI assistant into most tiers so the opt-out disappeared. A major CRM vendor moved in parallel, raising prices across its AI and collaboration products for the same reason. Buyers describe a "double uplift" on these renewals: the AI line forces an upgrade of the base license tier, and then the AI charge lands on top of the enlarged base.

No new purchase order is generated. No new vendor relationship is created. No procurement trigger fires. The exposure lands in your renewal terms and your data-processing agreement, and most organizations discover it after the fact.

The quiet one: embedded AI is the only type that enters without a purchase decision. The exposure is not something you chose to buy, it is something that got switched on while you were not looking.

What changes in the negotiation: the negotiation is your renewal, and the questions are about rights, not features. Review the updated terms at every renewal for expanded data rights, especially any language letting the vendor use your data to train or improve their models. Require the vendor to quantify the value of any AI capability bundled into a price increase, separately from the base product, and treat the double-uplift pattern as two negotiations, not one. Confirm the AI feature's data handling meets the residency requirements you already have, because embedded AI sometimes runs on separate infrastructure. And negotiate the right to switch AI processing off for sensitive workflows without losing the base product or paying a penalty.

Line the four up and the pattern is clear. A token-priced infrastructure deal, a solution with hidden model costs, an autonomous system sold on outcomes it cannot always define, and a set of rights quietly changing inside contracts you already signed. Four purchases, four cost structures, four negotiations. One category label stretched across all of them, which is how procurement ends up with a contract built for the wrong thing, and how a market where unit prices are falling still produces bills that come in over budget three quarters of the time.

I sat on enterprise AI governance committees before I understood half the vocabulary in the room, and what pulled me back to solid ground was realizing none of this requires a technology background. It requires the same discipline procurement already applies everywhere else: know exactly what you are buying, price it against something real, and put accountability in the contract before anyone signs. The only new part is recognizing that the word "AI" hides four different deals, and that the vendor across the table already knows which one they are selling you.

The question for your next AI conversation: which of the four are you actually buying, and are you negotiating it like the category it is, or like the category the vendor wants you to think it is?

Get that right, and every AI negotiation that follows gets easier. Get it wrong, and no clause in the contract can fix a deal that was built for the wrong purchase.

Next in the AI Spend Lifecycle: Part 2, the embedded AI hiding in your renewals. For the full negotiation framework this series draws on, see the companion AI Negotiations paper.

Token pricing and consumption: industry analyses of enterprise LLM pricing 2025-2026 (blended token prices down more than half year over year; average enterprise AI budgets up several-fold since 2024; directional, not point estimates); FinOps Foundation 2026 State of FinOps (majority of enterprises exceeded AI cost projections); enterprise LLM pricing references (frontier vs efficient spread of 10x to 100x+; output tokens 3-5x input; reasoning-token mechanics). Specific rates vary by provider and move quarterly.

Model tiering: Optimum Partners / DataStorage analysis of 2.4 billion enterprise API calls (tiered architecture median $2.31 per million vs $18.40 for all-frontier routing, an 87% gap).

Agentic pricing and adoption: Pickaxe and Nevermined (outcome-based examples: Intercom Fin ~$0.99 per resolution, Zendesk $1.50 per automated resolution, Sierra outcome model); Salesforce Q4 FY26 earnings release (29,000+ Agentforce deals closed since launch); MarkTechPost and MightyBot (large-vendor agent adoption; agent-washing); CodeGiant (retry and integration cost multipliers); Deloitte (outcome-based pricing accounting treatment).

ROI and pilot outcomes: MIT NANDA 2025 (roughly 95% of enterprise generative AI pilots produced no measurable profit-and-loss impact); enterprise AI research on the correlation between pre-deployment baselines and positive returns.

Embedded AI at renewal: CIO Dive, Windows Latest, and SaaSRise (Microsoft 365 list-price increases ~5-43% on July 1, 2026, tied to embedded AI; Copilot Chat folded into most tiers); Redress Compliance (the 'double uplift' renewal mechanic); CIO Dive (parallel Salesforce increases across Agentforce, Customer 360, and Slack).

All figures drawn from third-party research published in 2025 and 2026. Token-pricing references should be re-verified against current vendor pages before external publication, since AI pricing moves quarterly.