AI is moving to pay-as-you-go: how to control the bill before December

The big AI vendors are changing how they charge. From 1 December, new Microsoft Copilot Business licences bought through resellers will be billed on usage by default, Anthropic has released a cheaper model for high-volume work, and Google's Gemini can now act as an agent with its own business identity. For a small or medium business, that makes it easier to move AI from trial into real use, but also easier for costs and access to drift unless you set limits now.

What has actually changed in how AI is charged?

For the last couple of years, buying AI for your team mostly meant paying a fixed monthly price per seat. You knew the bill in advance, even if you weren't sure everyone was using what you paid for.

That model is shifting towards pay-as-you-go. Over the past fortnight, three announcements point the same way:

• Microsoft is making usage-based billing the default for new Copilot Business licences from 1 December.
• Anthropic has launched a cheaper small model aimed at high-volume work and added monthly API credits to some plans.
• Google has turned Gemini into an agent that can carry out tasks across business apps.

None of this is a reason to panic. It is a reason to decide, before December, who controls spending and what your AI tools are allowed to touch.

What does pay-as-you-go Copilot mean for your bill?

According to Microsoft's October Partner Center announcements, from 1 December pay-as-you-go becomes the default for new Copilot Business licences bought through Microsoft's partners. There is a default spending limit of 4,000 Copilot Credits per user per month, and admins can adjust it.

Two other details matter. Copilot will also be available without first buying a qualifying Microsoft 365 base subscription, which lowers the barrier to getting started. And, as reseller Sherweb notes in its October round-up, no action is required for existing subscriptions.

So if you already have Copilot seats, nothing changes overnight. The change bites when you add new seats, typically through your IT provider. At that point your monthly bill could vary with how heavily people use it.

The practical questions are simple:

• Is the default limit right for your team, or should it be lower?
• Who in your business, or at your IT provider, is allowed to change it?
• Who gets told if spending runs higher than expected?

Agree those answers before 1 December rather than discovering them on an invoice.

Why do cheaper models matter for a small business?

Anthropic's release notes describe Claude Haiku 5.5 as its cheapest, fastest and most capable small model, designed for high-volume, cost-sensitive tasks. The same notes say Max and Team plans now include monthly API credits for running your own apps and agents.

Separately, TechCrunch reports that Anthropic's startup programme now offers a free year of Claude Team with up to five premium seats, plus $1,000 in API credits.

Why does this matter if you're not a tech startup? Because the jobs that suit AI best in a small business are often the dull, repetitive, text-heavy ones:

• sorting and routing incoming emails
• pulling supplier, date and amount details out of invoices
• turning meeting notes into a list of actions

These tasks happen hundreds of times a month. When the model behind them gets cheaper, the sums for automating them change. A job that wasn't worth it a year ago might be now.

It's also worth checking whether you're already paying for a plan that includes API credits you haven't used. Unclaimed credits are the closest thing to a free trial you'll get.

What changes when an AI agent gets its own email address?

Google is turning Gemini into an AI agent that can plan, carry out tasks and work across business apps and systems. It can hand work to subagents, use several AI models, and gets its own workplace identity, complete with an email address.

Google's Workspace updates have added more everyday features too, including asking questions of business data in plain English from Google Sheets, publishing interactive apps from Gemini to Google Sites, and meeting recaps in Google Chat.

An agent with an identity is a bit like a new member of staff. You wouldn't give a new starter access to every shared drive, the finance system and the whole customer database on day one. The same thinking applies here:

• Decide which systems and folders the agent can see.
• Decide what it can do on its own and what needs a person to approve.
• Keep a record of what it has done, so you can check its work.

The Commons Library briefing of 14 October confirms the UK has no AI-specific regulation or legislation covering AI as a technology. That doesn't mean anything goes: your existing obligations around data and customer information still apply. It does mean the access rules are yours to set.

Is this the moment to move beyond the pilot stage?

NatWest's recent research into an emerging AI growth gap surveyed 1,400 UK SMEs and mid-market businesses. It found 44% currently use AI and a further 41% expect to adopt it within five years. But only 6% have reached the most advanced stage, and 29% of those using AI are still piloting.

Cheaper models and flexible billing remove two common excuses for staying stuck in a pilot: the upfront cost and the commitment to seats nobody uses. What they don't remove is the need for a clear job to do, a way to check the results, and someone who owns the outcome.

The businesses that get value tend to start narrow. One process, measured before and after, with a person checking the output until everyone trusts it. If you want help picking that first process, our AI workflow automation work starts exactly there.

What about automations you've already built?

Pay-as-you-go and new models are a good prompt to look at what you already run. Automations and small apps built a while ago, sometimes with AI coding tools, quietly depend on a specific model and a specific account. Vendors do replace and retire older models, and when that happens an automation can simply stop working.

Keep a list of each tool, the model it uses, whose account pays for it and who looks after it. If something you built yourself has become business-critical, our AI app launch checklist is a free way to see what it might be missing, and we can take AI-built prototypes to production properly.

What should you do next?

A few quick wins you can do this month:

1. Set your Copilot limit. Before 1 December, agree a monthly spending limit for any new Copilot Business licences and decide who can change it.
2. List your AI tools. Note every AI tool and automation, which model it depends on and who owns it.
3. Claim what you've paid for. Check whether any plan you already have includes API credits.
4. Pick one job. Choose one repetitive, text-heavy task and run it on a cheaper model with a person checking the results.
5. Limit agent access first. Before any AI agent gets its own identity, decide exactly what it can see and do.

If you'd like a second pair of eyes, our AI strategy and implementation and consultancy services can help you set sensible limits and choose where to start.