What does an AI agent platform cost?

What does an AI agent platform cost? It is the first question every serious buyer asks and the one most vendors make hardest to answer. The honest answer has two parts, because there are two different products being priced: per-seat AI assistants, where the number is public and simple, and agent platforms, where the number depends on what you are automating. Here is both, with real figures.

The thirty-second answer

Per-seat assistants cost roughly £20 to £75 per person per month. A ten-person firm should budget £3,000 to £9,000 a year before anyone has automated anything. An agent platform costs a one-off fixed fee to put the first workflow into production plus a monthly platform fee that covers the infrastructure, governance and model usage. The platform route costs more up front and less per unit of work done, which is why the comparison is a total-cost question, not a price-list question.

What per-seat assistants actually cost

The public numbers, as of September 2026:

• Microsoft 365 Copilot: roughly £22 to £28 per user per month on top of your Microsoft licences.
• ChatGPT Enterprise: priced per seat annually, typically comparable for a similar headcount, with usage-based elements on some features.
• Claude for teams: from around £20 per seat per month, with heavier usage billed separately.

Those numbers look modest until you count the whole stack. Our analysis of UK SME AI spending found firms averaging about £23,000 a year on AI tools, with a true cost closer to £57,000 to £71,000 once overlap between tools, unused licences and compliance gaps are counted. See the UK SME AI stack tax for that breakdown. The per-seat model's hidden cost is not the seat. It is the fact that a seat assists a person rather than running a process, so the headcount cost stays on your payroll.

What an agent platform actually costs

A serious agent platform has three cost components, and you should ask any vendor to separate them:

1. A one-off production rollout. The first workflow is scoped, quoted as a fixed fee, and delivered in one to five weeks. This is the only piece that carries a full build fee, because it is the piece that pays for the de-risking: edge cases handled, integrated with your systems, full audit trail, your team trained.
2. A monthly platform fee. The infrastructure, access controls, governance and support that everything after the first workflow builds on. Odokai's starts at £500 a month on the shared platform and £1,000 a month for your own instance, ex VAT.
3. Model usage. Included up to a fair-use volume, with a stated rate above it. You should know this number before the month starts, not after.

After the first workflow, the economics change. The platform, governance and integrations are already in place, so the second workflow is a fraction of the first, and most firms ship it in days. Workflows three to five compound on the same instance. That is the structural difference: per-seat costs scale linearly with headcount forever; platform costs step down after the first build.

The maths that matters

Run this before any purchase. Take the process you most want gone. Cost the hours it consumes each month at loaded staff rates. Compare that annual figure against (a) the per-seat licences that would help people do it faster and (b) the fixed fee that would remove most of it. If the process consumes £30,000 of staff time a year, a £9,000 assistant subscription that makes everyone 10% faster is worth less than a platform that removes 70% of the process itself. We published a full comparison of the enterprise platforms with every figure sourced, including the scenarios where each one loses.

What changes the price of a platform

• Deployment mode. A shared instance is the lowest-cost option. A dedicated instance in your own AWS account costs more and is what regulated firms usually need, because the documents, index and audit trail never leave your environment.
• Volume. Fair-use model volume is sized to your workflows during scoping, so the monthly fee tracks what you actually run.
• Integration depth. Connecting to your document management, CRM or case system is scoped work. Shallow integrations are cheap; deep ones are quoted honestly rather than discovered mid-project.

How to get a real number

Any vendor who will not separate build, platform and usage costs is hiding something, usually in that order. The workable route is a scoped engagement on one real workflow: you see the fixed fee, the platform fee and the usage model against your own documents before committing to anything recurring. That is exactly how our discovery process works, and it is why we can tell a firm what their second workflow will cost before the first one ships.

Bottom line

Assistants: budget £20 to £75 per seat, expect linear costs forever, and buy them for people, not processes. Agent platforms: expect a fixed one-off fee, a monthly platform fee, and costs that step down after the first workflow. If a vendor quotes you only one of those numbers, ask about the other two. And if you want the comparison in full, it is published, gaps included.