Most organisations can now see an AI bill. Far fewer can say what it bought. On 6 August, IBM announced the public preview of Apptio AI Value & ROI, a set of capabilities that links AI spending, including token costs, to measurable business outcomes. IBM planned general availability for the third quarter, so it is worth confirming where the product stands today.
IBM chief executive Arvind Krishna returned to the theme on the company’s third-quarter call on 21 October. He said clients ask two questions as AI scales: what will it cost, and who controls it. That pairing is a useful way to read the product, and a useful test of any AI contract.
What the product does
Apptio AI Value & ROI lets an organisation register each AI initiative and tie it to one or more business outcomes, using proof metrics the customer chooses, such as cycle time, cost avoided, conversion rate or incident volume. IBM says it tracks results across revenue, cost, speed, productivity and risk, and records a baseline, a target and the result actually realised over time.
It sits on top of two existing IBM products. IBM Cloudability provides visibility and attribution of token spend, and IBM Apptio AI TCO and Usage brings in modelled cost and consumption data, including technology, tokens, usage and labour. The preview is available to customers of Apptio Costing Standard or AI TCO and Usage, so it is not a product an organisation can simply switch on without an existing Apptio relationship.
Why tokens are a licensing problem
Tokens are becoming one more consumption unit that finance teams have to govern. Alongside them sit the credits that software vendors sell for their own AI features, each with its own rate card, allowances and expiry rules. A company that uses several AI products can end up with half a dozen meters and no single view of what any of them is delivering.
That is the gap IBM is selling into. Apptio’s own write-up of its AI cost work notes that visibility has not kept pace with AI investment and says IBM’s own CIO organisation uses the tooling for AI governance, citing fewer duplicated AI efforts and a 70 percent cut in software request processing time. Those are IBM’s figures about IBM’s use, and they should be read that way.
Visibility is not control
A dashboard shows what happened. It does not stop a team signing up for another AI service, and it does not change a contract that is already signed. Three things have to sit behind any tool of this kind.
The first is clean input. Attribution is only as good as the data feeding it, so check which AI services and meters the tool can actually read, and whether usage from vendors other than IBM is included. The second is a baseline from before the project started. Without a recorded starting point, a claimed saving cannot be tested. The third is ownership. Someone has to be responsible for each initiative’s outcome, and for deciding what happens when the numbers disappoint.
What to ask your vendors
Whatever tool you use, the underlying contracts decide how much control you really have. Ask each AI supplier for usage data you can export in a usable format, and how often it is refreshed. Ask how unused credits are treated, whether they expire and whether they can be moved between services. Ask what notice you will receive if a unit price or allowance changes. And ask whether the vendor’s own dashboard counts the same things your invoice does. When the two disagree, you want to find out before renewal rather than after.
A wider push into AI cost management
The August release is not IBM’s first move in this area. In June, Apptio introduced a conversational interface for technology spend, forecasting that uses IBM’s watsonx.ai models, and container-level cost visibility powered by Kubecost. Taken together, they show IBM building a cost and value layer for AI alongside its AI platform. That is a logical business, because the more AI an organisation runs, the more it needs to know what the AI costs.
It also means the tooling is part of IBM’s own commercial interest. A cost tool from a company that sells AI is not independent in the way an auditor is. That does not make it unhelpful. It does mean you should be clear about what you are using it for, and keep your own records of what you have bought.
Who should be involved
AI value tracking tends to fail when it is left to one team. Finance wants to know what the spend is and whether it is within budget. Technology teams know which services are running and how they are used. Procurement holds the contracts and the renewal dates. The business owners know whether the outcome was delivered. A tool like this works best when all four contribute, and when one person is accountable for the picture as a whole.
If you are starting from nothing, a spreadsheet is a perfectly good first step. List every AI service your organisation pays for, the unit it charges in, the contract end date, the owner, and the outcome it was bought to deliver. Many organisations discover they own more AI services than anyone realised, and that several of them overlap.
Measuring value is the hard part
Counting tokens is straightforward. Deciding what a token bought is not. A customer service assistant might be judged on resolution time, an engineering assistant on cycle time, a finance agent on cost avoided. Each needs a measure that the business agrees is fair, a starting figure recorded before the AI was introduced, and a way to separate the effect of the AI from everything else that changed at the same time.
IBM’s design reflects that by letting customers choose their own proof metrics and track baseline, target and realised results. That is the right shape. The weakness in any such system is human, not technical. Teams tend to pick flattering measures and to claim gains that would have happened anyway. A finance partner who challenges each claim is worth more than any dashboard.
A practical test for any AI purchase
Before approving a new AI service, ask five questions. What outcome is it meant to change, and by how much? What is the starting figure? Which meter will we be charged on, and what is the rate? Who owns the result? And what will we do if the result is not achieved after two quarters? If the answers are vague, the purchase is not ready, whichever tool you plan to track it with.
Conclusion
IBM’s Apptio AI Value & ROI is an attempt to connect AI spend to outcomes, at a time when most organisations can see the cost but not the value. It is a sensible idea, and it is still early. The preview launched in August, and general availability was planned for the third quarter.
Whether or not you use IBM’s tool, the discipline is the same. Record a baseline, pick measures that matter, know which meters you are paying for, and make sure your contracts give you the data and the notice you need to act on what you learn.
