Buying Software in 2026: Procurement Lessons from AI Vendors
Enterprise software procurement developed around a reasonably stable set of conditions: a product that changed slowly, pricing per seat, and a vendor whose costs were largely fixed.
Those conditions have weakened considerably, and contracts written on the old assumptions are producing unwelcome surprises.
Per-seat pricing does not fit consumption
Where a product’s cost to serve depends on usage rather than headcount, per-seat pricing is unstable. Vendors have responded with consumption components, credit systems and tiered allowances layered onto seat licences.
The result is that few buyers can predict next year’s bill. Insist on a modelled cost at your expected usage, and at double it, before signing anything.
The product you buy is not the product you keep
Capabilities change materially within a contract term, and not always in your favour. Features are deprecated, models are replaced, and behaviour shifts in ways that affect your processes.
Contracts should say something about material change: notice periods for deprecation, and what happens to your commitment if a capability you relied on is withdrawn.
Data terms deserve more attention than pricing
Negotiating effort concentrates on price, where the achievable gain is a percentage. The terms governing your data carry considerably more risk.
Establish what is used for training, what is retained and for how long, what happens on termination, and where processing occurs. These are increasingly regulated questions and the standard terms are frequently drafted in the vendor’s favour by a wide margin.
Exit is a design requirement
Ask concretely what leaving looks like: what data you get back, in what format, how long it takes, and what it costs. Vague reassurance is not an answer.
Where the honest answer is that exit is impractical, that is a strategic dependency and it should be approved at a level appropriate to that, not buried in a software renewal.
Pilot with a real exit
The most useful protection is a genuinely time-boxed pilot with pre-agreed success criteria and a real possibility of stopping.
Most pilots are not that. They are implementations with a trial label, and by the end the organisation has integrated the tool sufficiently that discontinuing is no longer practical. Deciding the criteria before you begin, and being willing to act on them, is worth more than anything you will win in the pricing negotiation.
