Managing the unpredictable costs of AI tokenomics

Your AI bill arrived and it is nothing like last month. Same tools, same team, but the number has jumped. Welcome to tokenomics: the billing logic quietly reshaping what AI actually costs a small business.

This is not a future problem. As of August 2026, the shift from flat-rate subscriptions to consumption-based token pricing is already creating financial unpredictability for businesses deploying AI at scale. The BBC reported this week that the issue is acute enough to catch out even large organisations, with Uber reportedly burning through its entire annual AI budget within months of deploying agentic systems.

For a finance manager or SME owner, that should focus the mind.

What token-based pricing actually means for your business

Every interaction with an AI model costs tokens. A token is roughly a mathematical chunk of text, somewhere between a syllable and a short word. When your team writes a prompt, the model reads it in tokens. When it generates a response, that costs tokens too. The longer and more complex the exchange, the higher the bill.

Flat-rate personal accounts, such as a standard ChatGPT Plus subscription, absorb much of this variation for individual users. But the BBC's reporting makes clear that vendors are moving to clamp down on these accounts for business use. If your team is running business-critical AI workflows through personal subscriptions, that arrangement is likely to become untenable.

Business and API accounts expose the full token cost. That is not necessarily bad; it gives you genuine usage data. But without controls in place, it means your monthly AI spend can swing dramatically based on how your team writes prompts, which models they use, and whether any automated workflows are misfiring.

Why agentic AI makes this harder to predict

The shift from using a single AI tool to deploying connected, multi-model systems creates a compounding problem. In an agentic setup, one AI assistant may call another, pass outputs back and forth, and complete multiple reasoning steps before returning a result to the user. Each of those steps consumes tokens.

A single poorly structured prompt in an agentic workflow does not just cost a few extra tokens. It can trigger a cascade of additional processing across multiple models. The cost of one bad prompt is multiplied by every step in the chain.

This is the dynamic that caught Uber off guard. It is not that AI became expensive overnight. It is that the architecture changed, and the cost controls did not keep pace.

For UK SMEs beginning to explore automation and AI agents, this is the most important financial lesson from this news. The technical complexity of multi-model systems is manageable. The financial exposure from uncontrolled token consumption is the part that can genuinely damage a technology budget.

Why this matters for SME owners and financial managers

Many UK businesses arrive at AI through experimentation. A team member signs up for a personal account, finds it useful, and the practice spreads. That is a reasonable starting point. The problem is that the habits formed during experimentation, writing long, exploratory prompts with no particular discipline, become expensive when the business moves to a paid API or an agentic system.

For financial managers, the core difficulty is that token costs do not map neatly onto existing budgeting frameworks. There is no per-seat licence fee to approve. There is no predictable monthly ceiling. The cost is a function of behaviour: how your team writes, how often they iterate, and which models they use for which tasks.

That makes AI spend an operational management problem as much as a procurement problem. It sits at the intersection of technology governance, team training, and finance oversight. None of those functions typically owns it outright.

gecco's approach to AI Assistants is built around exactly this: structured prompt frameworks, using the GRAFT methodology, that make AI interactions consistent, purposeful, and measurable rather than open-ended and unpredictable.

For SME owners, the immediate concern is simpler: if your team is using AI tools in any business capacity, you need visibility into what that is costing. Personal subscriptions obscure this. Moving to a business account or API setup gives you the data, but only if you also establish basic controls before the bill arrives.

Three actions to take this week

1. Audit your current AI access. Ask every team member which AI tools they are using and whether those are personal or business accounts. This takes under 30 minutes and is the essential first step before any cost controls are possible.

2. Choose the right model for each task. Larger, more capable models cost significantly more per token. Not every task requires the most powerful model available. A financial manager reviewing a first draft of a supplier email does not need the same model as one analysing a complex contract. Identify two or three recurring AI tasks in your business and match each to the smallest model that gives a satisfactory result.

3. Draft a basic prompt structure for your most common AI tasks. Even a simple template, specifying the role you want the AI to take, the format you want the output in, and the length you expect, can meaningfully reduce token consumption per interaction. This does not require technical expertise. It requires one person to spend an afternoon testing and documenting what works.

The honest limitation here

There is no silver bullet for AI cost management at SME scale. The monitoring tools that enterprise teams use to track token consumption in real time are often expensive, technically complex, or both. Most small businesses will be managing this manually for the foreseeable future, which means the controls have to live in human behaviour and documented process rather than in software.

Prompt standardisation works, but it requires ongoing discipline. Workflows drift. New team members learn habits from colleagues rather than from documented guidelines. The Uber example is instructive precisely because Uber has sophisticated engineering teams and still lost control of its AI budget. For an SME without a dedicated technology function, the margin for error is smaller, not larger.

The most realistic near-term position for a UK SME is not perfect cost control. It is informed cost awareness: knowing which tools your team uses, roughly what each task costs, and where the biggest sources of variation are. That foundation makes better decisions possible.

Where to go from here

If your business is trying to get a clear view of what AI actually costs, and whether your current usage patterns are financially sustainable, that is a good starting point for the AI Readiness survey. Taking the survey gives you access to 65+ free resources and a custom AI Readiness report, followed by a free 45-minute call to walk through the results with someone who can help you read them.

Website · LinkedIn · Case Studies · Newsletter