Somewhere in most energy organizations, there is now a monthly technology invoice that has grown materially over the past eighteen months and that nobody in the building can fully decompose.
It arrives from a hyperscaler, or from a model provider under an enterprise agreement, or increasingly from a software vendor whose renewal quietly absorbed a set of capabilities that did not exist at the previous negotiation, and it blends subscription entitlements, seat licensing, and consumption-based charges into a total that reconciles cleanly against the contract while explaining almost nothing about what the organization actually bought.
Between them, they frequently cannot say which business capability consumed what, on whose behalf, or to what commercial end.
Expense is the more familiar of the relevant concerns, and expense is largely tractable, because an organization that knows a capability costs too much has at least established what the capability is and can then decide whether to negotiate it, constrain it, or stop it.
An unattributable cost forecloses all three of those options at once, since a capability that cannot be isolated cannot be evaluated, and a capability that cannot be evaluated will neither be scaled when it is working nor terminated when it is not. The spend continues, growing at whatever rate consumption happens to grow, defended by no one and questioned by no one, until it becomes large enough that someone asks a question the available data cannot answer.
This challenge is well documented outside the commodity sector, proving the problem is structural rather than a symptom of any particular industry's technology maturity:
"Is your AI providing value? No one can answer that question yet." - The FinOps Foundation
The structural reason this persists in refining, distribution, and retail fuels organizations is that cost and benefit are recorded in different parts of the enterprise and are never brought into contact with one another.
The people paying for the capability cannot observe what it produced, and the people benefiting from it have no visibility into what it costs. The conversation that would ordinarily resolve whether the thing is worth doing simply never takes place.
Large trading houses do not face this problem in the same way. In the trading organizations of Geneva, Zug, Singapore, and Houston, a desk carries the costs it generates against the P&L it produces, making market data subscriptions, analytics, systems, and support visible to the trader whose results they are meant to improve.
A refiner generally cannot replicate the arrangement because refining margin is not desk P&L and a scheduling function does not carry an income statement. Imposing desk-style allocation on a downstream operating structure will generate a great deal of internal argument and very little insight.
What IT organizations in downstream companies may have, and in some cases have had for a long time, is showback (vs. full chargeback). Most technology leaders in this industry have built or inherited some version of it: a periodic view that attributes infrastructure, application, and support consumption to the business functions that generated it, published to those functions without any transfer of budget and without any invoice.
Organizations that carefully attribute storage, compute, and application support to business functions are frequently carrying an AI line that is attributed to nothing, reviewed by no one outside technology, and growing faster than anything else they own.
The version of this that will cause the most difficulty over the coming years has barely begun to surface. Artificial intelligence is increasingly arriving inside software that organizations already own. CTRM and ETRM vendors across every tier of the market are actively embedding document intelligence, conversational reporting, and workflow automation into their platforms, delivered through ordinary product releases rather than through discrete procurement.
A capability that arrives this way has no acquisition cost to point at, no business case to review, and no owner outside the vendor relationship, rendering it invisible to review processes that would otherwise catch it.
Vendors currently absorbing inference costs to demonstrate value and drive adoption will eventually need to recover them; the mechanism for recovery is renewal.
That is a poor position from which to argue about price, and it is entirely avoidable by organizations that begin tracking usage of embedded capabilities well before the question becomes commercial.
A reasonable objection that all of this is transitional., The strongest version of it comes from Gartner, which expects the cost of performing inference on a very large model to fall by more than ninety percent for providers between 2025 and 2030. While credible, the assumption that falling prices will fix attribution fails because this industry already ran this experiment with cloud computing.
The original cloud proposition was explicitly utility-shaped. Consumption would be metered, organizations would pay only for what they used, capital expenditure would give way to operating expenditure, capacity would flex with demand, and the meter itself was offered as a transparency improvement over the capital budgeting process it replaced. Unit prices for compute, storage, and bandwidth then fell dramatically over the following decade, which is precisely the trajectory now projected for inference. FinOps exists as the industry's acknowledgment that metered consumption at falling prices produces bills organizations cannot decompose. Six annual surveys later, allocation remains the most prioritized capability across every technology category the discipline now covers.
The utility analogy also breaks down with AI in ways that should concern anyone budgeting for this:
Cheaper capabilities are also consumed more widely across more processes, with less scrutiny applied to any individual use, meaning falling unit costs increase the number of things an organization pays for while decreasing oversight. There is a further wrinkle in the survey data worth noting: many organizations report being asked to fund their AI investments from savings generated by optimizing existing technology spending. The efficiency recovered from the last illegible metered technology is being redirected into the next one, without the attribution discipline having been carried across in between.
The practical step is smaller than a chargeback program and considerably more useful than another budget line.
The Committee of Chief Risk Officers frames the underlying question as one that belongs to the selection process, on the reasoning that a capability whose cost and benefit have never been established is difficult to govern and easy to justify indefinitely.
That exercise requires no budget approval, no allocation methodology, and no negotiation with anyone about whose cost center absorbs what. It produces the one artifact that every subsequent decision about this technology quietly assumes already exists: a defensible account of what the organization is spending, on which capabilities, and on behalf of which parts of the business. Organizations that build this visibility will approach the next renewal cycle with leverage. Organizations that do not will approach it with a total.
When you choose Opportune, you gain access to seasoned professionals who not only listen to your needs, but who will work hand in hand with you to achieve established goals. With a sense of urgency and a can-do mindset, we focus on taking the steps necessary to create a higher impact and achieve maximum results for your organization.