|
AI, Applied
Some AI tools are now billed by the work they do. That changes who controls the spending.
|
AI, Applied
What changed. What matters. What to do about it.
Issue 07 · Fortnightly · 5-minute read
|
|
The Big Shift
|
|
The AI bill is changing. So is who controls it.
Business software has usually been bought by the seat: someone approves a fee per person, and the cost is known in advance. Some AI tools now add a second charge, for the work they do.
GitHub moved Copilot to usage-based billing on 1 June, saying that under its previous pricing a quick chat question and a multi-hour autonomous coding session could cost the user the same.
The consequence is a shift in spending authority. An employee can now initiate work that consumes a variable amount of AI usage. Depending on the organisation's settings, that spending may occur without a separate approval from finance or procurement.
A licence kept two things together: initiating spending and authorising it. Usage-based pricing separates them. The challenge isn't necessarily to approve every AI task. It's to establish who can initiate which kinds of work, within what limits, and how the resulting expenditure is assessed.
A related view comes from an essay by Jared Spataro, which Microsoft links: the cost of AI agents (tools that carry out tasks on their own) should be funded by the business function that creates it, not by IT.
|
|
Three suppliers, three different models
|
|
GitHub Copilot Business: usage included, extra on by default
In the seat: US$19 per user a month, including 1,900 AI credits per user, pooled across the organisation. On top: once they run out, extra usage is billed at published rates. Control: extra usage is on by default, and an administrator can switch it off.
|
|
Microsoft 365 Copilot: a flat seat, some features billed by use
In the seat: a flat per-user fee. On top: some experiences, such as the agent tool Cowork, need a Copilot licence but are also billed by use, in “Copilot Credits”. Control: administrators can set limits per user or group. Australian availability of this billing option isn't confirmed (see Worth Knowing).
|
|
Anthropic Claude Enterprise: access only, all use billed
In the seat: access only, at US$20 per seat a month (billed annually, 20-seat minimum), with no usage included. On top: every token (a small chunk of text an AI reads or writes) is billed at standard rates. Control: usage billing can't be switched off, but owners can set spend limits for the organisation and for individuals.
|
|
|
These are billing models in use today. If nobody changes the settings, the limit that applies is a default someone else chose. Cost also varies by task: GitHub estimates the average AI code review at US$0.25 to US$10, depending on the effort level chosen, before separately charged computing time. These are estimates only.
AI spending is already difficult to manage, whatever the pricing model. Elastic's June survey of more than 500 decision-makers at Australian organisations (50+ employees) using or piloting AI found that one in three said they had exceeded their AI budget last year. Separately, EY's July survey of 534 senior US leaders found that only 64% of those at organisations investing in AI said theirs both actively monitors token use and has clear budgetary guardrails. Both come from commercial organisations, and neither survey establishes that usage-based pricing caused the problems. But together they point to a broader challenge: businesses need better visibility over AI costs, and clearer ways to judge whether the spending delivers value.
Controlling cost is only half the question. A business also needs to know whether the result of a task justified what it cost, and who is accountable for that judgement.
Usage-based pricing doesn't necessarily mean AI will cost more. It changes how costs arise, who can trigger them, and how they should be managed. The financial decision doesn't end when the licence is purchased.
|
|
What I Would Do
|
|
Three practical actions
|
1
|
Sort every AI charge into one of three kinds.
A fixed fee, an allowance included in a fee, or usage charged on top. If you find only fixed fees, note when they renew and move on.
|
|
2
|
Check whether your spending limits actually stop spending.
An alert only tells you, a budget may only track spending, and only an enforced limit stops it. In GitHub, for example, an organisation or cost-centre budget only sends an alert unless an administrator switches on “Stop usage when budget limit is reached”. Ask for the dollar equivalent of anything quoted in credits.
|
|
3
|
Run one real task before you scale.
Before a wide rollout, run one real task and record what it cost and what it produced. Then decide who may start work like it, and who judges whether the result was worth the cost. A larger business may need an approval step; a small one just a simple rule.
|
|
|
|
Worth Knowing
|
|
A planned 2027 default on one Microsoft product, with Australia not yet confirmed
Microsoft's 3 November partner update says usage-based billing is planned to be on by default for new Microsoft 365 Copilot Business subscriptions bought through Cloud Solution Provider partners, from January to March 2027. This is a change to one billing arrangement, not the start of usage-based pricing. The default is a spending limit of 4,000 Copilot Credits per user a month: a cap on use that administrators can adjust, not included usage, with no published dollar value. Australia was excluded at first; the update says the timeline covers all markets including Australia but gives no Australian date, so ask your partner what will apply.
|
|
What one platform's card data shows about AI spending
Weel, a business card and expense platform, publishes an index built from anonymised spending by more than 4,000 of its Australian and New Zealand customers. In its latest edition (22 October), 32.0% were paying for AI in September, up from 30.8% in June. The average monthly spend per OpenAI customer rose from $298 to $407. For Anthropic it peaked at $1,172 in July and eased to $1,115.
Read it narrowly. Weel says its customers skew towards small and medium businesses, and that card data can't show why a business changed what it spends. Australia and New Zealand are combined and no currency is stated. Microsoft Copilot and Google Gemini can't be separated out.
|
|
|
Useful Resources
|
|
How included credits, shared pools, extra usage and budgets work.
|
|
Which services are billed by use and how administrators manage spending.
|
|
Seat fees, usage charges, prepaid credits and spend limits.
|
|
|
One Question For Your Business
Who can authorise AI spending in your business — and how do you know it's delivering value?
|
|
|
Mia Pursehouse · Applied AI Research & Analysis
|
|
|
AI, Applied is an executive briefing for leaders who don't have time to follow AI but need to know what it means for their business.
|
|
AI transparency: AI tools assisted with research and drafting. Final selection, analysis and editorial review by Mia Pursehouse.
|
|
Mia Pursehouse, Buttaba, NSW 2283
Sent to Reader
|
|
|
|
