Jason Pepper

by Jason Pepper

SAM Practice Head, SAM & License Management, Version 1

A couple of years ago, the conversation about AI tools in enterprise IT was mostly about access. Could teams get Copilot approved? Would security sign off on ChatGPT? The arguments were about governance in the policy sense: who was allowed to use what. 

That conversation has largely been settled. The tools are in. AI subscriptions have been approved, agents are running, and developers, finance teams and HR functions are all consuming AI at a pace nobody fully anticipated. Gartner projects end-user spending on GenAI models in Europe to grow 78.2% in 2026 alone.  

The new conversation is about money, and it is happening in finance directors’ offices, not IT strategy sessions. But how prepared are organisations to rein in these spiralling AI costs before they get out of hand?

The bills are arriving before the frameworks exist to manage them 

AI tools price differently to traditional software. You do not buy a server or a fixed AI subscription seat and know exactly what you will spend. You buy access to tokens, pay per API call, commit to consumption tiers, and watch agents burn through budget while workflows run in the background at times when nobody is watching. A single agentic workflow poorly configured can consume a month’s expected token budget in a day. 

The result is predictable: invoice shock. Finance teams are seeing AI vendor lines that are larger than expected, broken down in ways that are hard to interpret, attributed to no specific business unit, and impossible to challenge without understanding the underlying consumption model. 

This is not a new problem. It is a familiar one wearing a new outfit.  

Cloud infrastructure cost management went through exactly this cycle between 2015 and 2020. Organisations adopted AWS and Azure quickly, ran workloads at whatever scale felt necessary, and then discovered that unmanaged cloud consumption is expensive. An entire discipline, FinOps, grew up to address it.  

The FinOps Foundation now reports that 98% of FinOps practitioners are managing AI spend alongside cloud costs. That number was 31% just two years ago. The discipline has absorbed AI spend management because the problem is structurally identical: untracked consumption, misaligned purchase tiers, and no owner. 

Why this is happening 

Three things are driving it simultaneously. 

AI tool procurement has largely bypassed the normal software acquisition process. Where traditional software purchases go through IT, legal and procurement, AI tools have often been adopted at team level, sometimes on personal credit cards or departmental budgets, with limited visibility at the centre. Copilot AI subscriptions might be managed centrally, but OpenAI API access for a development team’s internal tooling probably is not. 

Gartner found 41% of enterprise AI usage occurs in tools not approved by IT, 
multiplying both cost and compliance risk 

AI subscription tier decisions are being made without analysis. Most organisations have not assessed whether their chosen AI subscription tier is appropriate for their actual usage pattern. E5 AI subscription for Copilot is expensive and not warranted for every user. Bulk token purchasing arrangements may significantly reduce unit costs for high-consumption workloads. These are decisions that require consumption data to make well, and that data is rarely being collected. 

Agentic workflows have no cost governance equivalent to what exists for human labour. When an employee runs a process, there is a salary cost attached to the time. When an AI agent runs the same process autonomously, at scale, repeatedly, the cost can be orders of magnitude higher, and nobody has set a budget for it because the concept of an “AI agent running budget” does not exist yet in most financial planning frameworks. 

What organisations should do about it 

  • Start with visibility. Before any optimisation is possible, you need to know what you are spending, on which tools, through which teams and agents. This is a data collection and analysis task, not a procurement conversation. Pull together API spend, AI subscription invoices, consumption logs where available, and map them against business units. Most organisations doing this for the first time are surprised by what they find. 
  • Right-tier your AI subscriptions. Once you have consumption data, audit your AI subscription structure against actual usage. Not every Copilot user needs E5. Not every developer calling an API needs a premium tier. The savings from right-tiering alone can be material, and the analysis is not complex once the consumption data is in hand. 
  • Establish ownership. AI cost governance needs an owner, and that owner needs both commercial and technical literacy. An IT director who understands AI subscriptions but not AI models, or an AI lead who understands the technology but not commercial terms, will each miss different things. The most effective arrangements combine both. 
  • Build ongoing monitoring, not point-in-time audits. A one-off baseline assessment tells you where you are today. It does not tell you where you will be in six months when usage has grown, new agents have been deployed, and a vendor has changed its pricing model. Consumption monitoring needs to be continuous, connected to financial reporting, and reviewed at least monthly. 

Where Version 1 can help 

We have established robust AI FinOps processes and governance frameworks to oversee AI and software asset management (SAM) costs within our own operations. This includes; 

  • continuous monitoring of consumption 
  • detailed analysis of AI subscription structures 
  • proactive identification of opportunities to optimise spend 

Our internal teams combine commercial and technical expertise to ensure that every aspect of AI tooling – from AI subscription tiering to API usage – is managed efficiently and transparently.  

By implementing ongoing reviews and connecting our monitoring to financial reporting, we maintain visibility over our AI estate and can swiftly adapt to changes in usage or vendor pricing.  

If you are looking to achieve similar outcomes within your organisation, our experience and proven methodologies in FinOps for AI can help you establish effective cost governance and maximise value from your AI investments.  

Contact us to learn more.