Greg Holmes on Why Modern AI Demands a New FinOps Discipline

The global enterprise software market is projected to reach US$1.4tn by 2030, according to data from Grand View Research, fuelled by a relentless shift toward cloud-native architectures and the Gen AI boom. 

Yet, as technology footprints expand, the complexity of managing these investments grows. 

Today’s leaders must focus on mastering the business of technology. This is difficult because 44% of UK IT leaders cite in Apptio’s Technology Investment Management Report a mistrust in data as their main barrier to making decisions. 

Since its inception in 2007, Apptio has positioned itself as the definitive platform for solving this friction, helping over 60% of the Fortune 500 translate fragmented technical telemetry into clear business value through FinOps and automated technology business management. 

Now part of IBM, Apptio’s reach spans 22 countries, providing the architectural discipline required to manage the volatile consumption patterns of the AI era.

Here, Greg Holmes, EMEA Field CTO at Apptio, discusses how leaders can bridge the gap between black-box AI expenses and measurable unit economics to ensure every pound of cloud spend drives a quantifiable strategic edge.

Recent Apptio research reveals that 44% of UK IT leaders say the biggest barrier to making technology decisions is a mistrust in data. What are your thoughts on this?

Low confidence in data accuracy is typically the result of fragmented systems and a lack of real-time insight. 

When cost information – like cloud bills – and portfolio management is siloed in different places, the process of manually piecing it all together is a common source of friction. 

This slows down critical decisions, and ultimately stops teams being able to make informed decisions. 

Leaders need integrated and accurate data, contextualised for all stakeholders, to create a credible, shared understanding of value. 

In practice, this means creating an integrated data architecture that connects the numbers in finance systems (like ERPs) with real-world consumption from operational tools (like cloud usage telemetry and portfolio management software). 

You’ve spoken about the importance of unit economics, like cost-per-inference or cost-per-API-call. Why is this more critical for AI than it was for traditional SaaS or cloud infrastructure?

While traditional cloud and SaaS costs are often more predictable, AI workloads are incredibly dynamic and resource intensive. 

Without granular unit economics, such as cost-per-inference, you could be flying blind. This is because you cannot accurately price your AI-powered products, determine the profitability of a new feature, or even decide which model is the most cost-effective. 

It is the difference between treating AI as a black-box expense and managing it as a value-driving part of the business. 

By connecting every pound of cloud spend to a specific business metric, AI can be managed with the same discipline as any other major business investment.

What are the most common hidden costs of AI that leaders tend to miss in their initial forecasts?

The elements of AI spending that often go untracked or unnoticed are data storage costs, licensing fees and the spend required to build a team of skilled engineers to keep everything running smoothly. 

This is because these costs are often scattered across departments and are tricky to identify within wider IT budgets. 

These costs might not look like AI costs, but they are part of the AI cost chain and need to be factored in.

As a result of this low visibility, leadership lacks a precise understanding of the true financial impact and overall profitability of a project or specific technology. This is why a single, shared view of spending is becoming business critical – to keep everything in check. 

When a CFO asks for the business value of an LLM project that hasn’t hit production yet, what are the key metrics you recommend CTOs use to defend that spend?

A CTO should be able to clearly articulate how the new model will be better than the current process. This can be done by modelling the future state, so it’s clear to see forecasted gains from team capacity, reductions in process friction and accelerated delivery cycles.

Important things to track are outcome related metrics in terms of units of work executed (e.g. sales driven, customer requests handled, items manufactured), consumption metrics, which resources are used by which processes and cost metrics. 

By joining up these data sets, you get an understanding and a model to plan future costs when projects go into production.

Once this is then presented in a format that everyone at the table can understand, the way to secure buy-in is to connect the investment to strategic imperatives like securing a competitive edge or unlocking future revenue streams. 

By translating a technology initiative into the language of business outcomes, you move the conversation from a cost justification to a shared strategic objective with your CFO.

With the rise of Gen AI, cloud spending has become more volatile than ever. How is the discipline of FinOps evolving to handle the unpredictable, consumption-based nature of AI workloads compared to standard compute?

The rapid adoption of Gen AI has created significant tech sprawl. Our data reveals a critical gap: only 30% of UK organisations have a clear view of their AI-specific spend, with 63% of FinOps teams still relying on manual processes. This isn’t sustainable. 

A Gen AI-led world demands a more dynamic, data-driven evolution of FinOps. At Apptio, we see this happening in three keyways. 

First, through enhanced visibility and forecasting, moving from monthly reports to real-time dashboards that track AI cost drivers and anticipate demand. 

Second, FinOps must foster deeper cross-functional collaboration, creating a shared language for engineers and finance teams to innovate responsibly within financial guardrails.

Finally, the focus must shift to aligning spend with business value. The conversation has to mature from tracking costs to measuring value. 

By instituting granular, AI-native metrics like cost-per-inference, and even AI-cost per sale/transaction, organisations can apply the necessary financial discipline to confirm that every pound of AI spend produces a clear, quantifiable impact on business performance.