Your Microsoft account manager sends the renewal quote. Buried in the line items is Copilot for Microsoft 365: £25 per user per month, on top of what you already pay. For a 30-person firm that’s another £9,000 a year. The AI is bundled in, apparently. You don’t remember agreeing to that.
This is the moment a lot of UK founders discover that AI SaaS pricing has changed shape. The three things worth understanding before you sign anything: what you’re actually paying for, how the bill grows as you scale, and how hard it would be to change your mind.
What choice are you actually facing?
AI SaaS pricing has split into three distinct models that vendors often mix and layer. Per-seat licences charge the same regardless of how much AI any individual uses. Usage-based billing charges by the token, API call, or document processed. In between sit credit pools and outcome fees. Which model suits your firm depends on usage pattern, risk appetite, and how much negotiating room you have.
Around 65% of major SaaS vendors have added an AI usage meter on top of existing seat-based pricing, according to analysis by Tropic citing Bain research. That creates what buyers are calling an “AI tax” at renewal: forced migration to new pricing tiers that carry a 20 to 37% uplift, whether or not you asked for the AI features. Understanding which model you’re being sold matters before the contract lands.
The named operators across the spectrum illustrate the range. Microsoft Copilot sits at the flat per-seat end. OpenAI’s API is purely usage-based, billing per thousand tokens. Box AI uses a credit pool shared across users. Intercom’s Fin agent charges per resolved support conversation. Salesforce Einstein layers usage charges on top of existing CRM licences. Each represents a different answer to the question of who carries the cost risk when AI use turns out heavier, or lighter, than expected.
When per-seat pricing is the right call
Per-seat models like Microsoft Copilot make sense when the AI is woven into daily work for your whole team. The UK price for Copilot for Microsoft 365 is £25 per user per month, on top of existing licences. If staff spend their day in Outlook, Word, and Teams and will actively use the AI features, the fixed cost is predictable and the administration is simple.
The model works when three conditions hold: near-universal adoption across the team, a genuine fit between the AI capabilities and the daily workflows, and the bargaining position to cap year-on-year price increases. Without all three, the economics shift. If only a third of your staff end up as active users, you’re paying the full seat price for AI features that most of the firm rarely touches.
The negotiation points to press before signing include whether AI features can be unbundled from the core licence, what the usage distribution looks like in comparable firms, and whether you can lock the per-seat price for two or three years. Vendors pushing forced AI bundles at renewal are responding to exactly the margin pressure their pricing creates, and the CMA has flagged that committed-spend discounts and technical restrictions in cloud platforms can limit customers’ ability to switch or run services in parallel.
When usage-based pricing fits your firm better
Usage-based pricing charges by the work the AI actually does: tokens consumed, API calls made, documents processed, or conversations resolved. Around 83% of AI-native SaaS companies already use this model, according to Maxio data, and Gartner forecasts that 40% of enterprise SaaS spend will shift from seat-based pricing by 2030. For firms in the pilot phase or with uneven workloads, paying for consumption rather than access is generally the lower-risk approach.
The case for usage-based models is strongest when demand is uncertain or seasonal. A professional services firm running AI over client documents during busy periods but barely touching it in quieter months is a natural fit for pay-as-you-go. A firm that wants to test two or three tools before committing to one should not be on annual seat licences for any of them.
The risk is bill shock. A misconfigured AI agent, a developer with an uncapped API key, or a team running experiments without oversight can generate a significant invoice at the end of the month. Stripe, writing on pricing design for AI SaaS founders, recommends modelling the worst-case usage month before going live. As a buyer, the same logic applies: ask what happens to the bill if consumption spikes and whether the vendor throttles or simply charges.
Hard monthly caps, spend alerts, and the ability to attribute usage to teams or projects are the three controls to ask for. Without them, usage-based pricing shifts all the cost risk to you and gives you no early warning when something is running out of hand.
What it costs to get this wrong
Getting the pricing model wrong on AI SaaS creates costs that compound over the contract term. A 50-person firm buying Microsoft Copilot at £25 per user per month adds £15,000 a year before anyone measures a single productivity gain. If the AI features go largely unused, that’s straight overspend. On the usage side, a misconfigured agent or unguarded pilot can send a bill you weren’t expecting.
Three categories of cost show up when the wrong model is chosen. The first is direct overspend: paying for licences nobody uses, or usage bills without caps. The second is the AI tax at renewal, where Bain’s analysis suggests forced SKU migrations carry a 20 to 37% price uplift. The third is migration and dual-running costs when you need to exit. Switching AI providers when your data, workflows, and staff habits are embedded in one platform takes months and carries real cost in consultancy, retraining, and running two systems in parallel.
There is also regulatory exposure to price in. The ICO fined Clearview AI £7.5 million for unlawful use of UK personal data. For a small firm, enforcement at that scale is unlikely, but an ICO investigation or complaint triggers investigation time and remedial work that carries real cost even if no fine follows. If you have chosen a tightly bundled AI inside a wider suite and later discover it processes personal data in a way that needs to change, your ability to modify or suspend the AI without disrupting core operations matters more than it seemed at contract time.
What to ask before you sign
Before committing to any AI SaaS contract, three lines of questioning matter: what the pricing is really tied to, what cost protections you have, and whether you can exit without pain. Vendor conversations focus on features and headline price. The contract terms, the usage guardrails, and the data-export provisions are where the real risk sits, and these are the questions that buyers rarely raise.
On value: ask what business outcome the pricing is tied to. Time saved per person? Tickets resolved? Revenue influenced? Ask for case studies from firms of comparable size in your sector. Ask whether you can run a 60 to 90-day pilot with agreed KPIs before committing to full deployment.
On usage and cost: ask for the precise units and tariffs. Ask whether there are hard monthly spend caps and automated alerts. Ask what the vendor does if consumption spikes above a fixed level, whether they throttle or simply invoice. The NCSC’s cloud security guidance specifically recommends assessing access controls, logging, and supplier lock-in risk before committing to any cloud-based service, and this applies directly to AI SaaS running on shared infrastructure.
On lock-in and exit: ask how you export your data, in what format, and what high-volume exports cost. Ask what happens to AI-generated outputs such as summaries, embeddings, or fine-tuned models if you leave. Ask for the termination conditions and whether there is a price lock with a cap on annual increases. The ICO’s accountability guidance makes clear that you remain responsible for personal data processed by third-party AI tools, which means your ability to retrieve, correct, or delete that data on request is a compliance obligation, not just a commercial preference. If the vendor cannot tell you how to meet that obligation after contract end, that is the answer.
