AI features broke per-seat SaaS pricing, and mid-market buyers are the ones absorbing the cost
For roughly two decades, SaaS pricing ran on a simple and fairly honest assumption: the more people at a company used a product, the more value they got from it, and the more they should pay. Per-seat pricing was predictable for buyers and easy to forecast for vendors. That assumption is now broken, and the reason is AI features.
A single employee using an AI copilot inside a CRM, a support desk, or a coding tool can generate ten or a hundred times the backend compute cost of a colleague who never touches the AI feature. Inference isn't free, and it doesn't scale with headcount — it scales with usage intensity, which varies wildly and unpredictably between individual users. A vendor charging flat per-seat pricing for an AI feature is either eating that cost on power users or subsidizing light users at everyone else's expense. Neither is sustainable at scale, so vendors are moving the AI layer to consumption-based billing: price per API call, per token processed, per compute-minute, or per AI-assisted action completed.
The pricing structure buyers are actually seeing
The pattern showing up across CRM platforms, developer tooling, customer support software, and productivity suites follows a consistent shape: the base product keeps its familiar per-seat price, and the AI features that used to be marketed as included get unbundled into a metered add-on. Some vendors frame this as a new premium tier; others frame it as "AI credits" that get consumed and need periodic top-ups. Either way, the buyer's monthly bill now has a variable component that didn't exist in the previous contract generation, and that variable component is driven by how aggressively individual employees use a feature the buyer doesn't fully control.
This isn't vendors behaving badly — it's a rational response to a real cost structure that per-seat pricing was never built to handle. The problem is entirely in how the transition is landing on different sizes of buyer.
Why mid-market buyers absorb the worst of this
Enterprise buyers with six- or seven-figure annual contracts have negotiating leverage that smaller buyers don't: they can push for committed-use discounts, negotiate usage caps with overage protections built into the contract, or demand volume pricing that converts unpredictable metered costs into something closer to the old flat-rate predictability. Procurement teams at large companies also have the staffing to model AI usage projections before signing and to audit monthly bills afterward.
Small buyers, meanwhile, typically stay on self-serve or low tiers where usage is naturally capped by a small team size, limiting the blast radius of a usage spike almost by accident.
Mid-market buyers — companies large enough to have real usage volume and real AI-feature adoption, but not large enough to command a dedicated account team or negotiate custom contract terms — get the worst of both situations. They're big enough to generate unpredictable, meaningful usage spikes, and too small to have the contractual leverage to cap the financial exposure from those spikes. A 50-person company that adopts an AI-assisted workflow enthusiastically can see a SaaS line item triple quarter over quarter with no warning, discovered only when the invoice arrives.
What mid-market buyers should actually do about it
The fix isn't to avoid AI features — the productivity gains are usually real. It's to negotiate the contract differently than buyers have historically negotiated SaaS deals.
Ask for usage alerts and hard caps before signing, not after the first overage bill. Any vendor serious about consumption pricing should be able to configure a usage ceiling with an alert at 70-80% of a budgeted threshold. If a vendor can't offer this, treat it as a red flag about how mature their metering infrastructure actually is.
Model worst-case usage, not average-case usage, before signing. Ask the vendor for the actual unit cost (per token, per API call, per AI action) and multiply it against your most AI-enthusiastic power users' plausible usage, not your team's average. The gap between average and worst-case is where surprise bills come from.
Push for a committed-use discount even at mid-market volume. Vendors increasingly have tiered consumption discounts designed for enterprise accounts that they haven't advertised at smaller contract sizes. It frequently costs nothing to ask for the same discount structure at a lower commitment threshold — vendors would rather lock in predictable revenue at a discount than lose the deal entirely.
Demand unit-cost transparency, not just a final bill. A vendor who can show you cost-per-unit pricing is one whose metering you can actually audit. A vendor who only shows a bundled monthly total is one whose pricing you're negotiating blind.
The shift to usage-based AI pricing isn't going to reverse — the underlying cost structure makes per-seat AI pricing a dead end for vendors. The buyers who come out ahead will be the ones who renegotiate their SaaS contracts around that reality now, rather than discovering it in next quarter's invoice.