IBM’s Apptio unveiled AI Value & ROI in public preview Aug. 6 — a tool tying AI spending, including token costs, to measurable business outcomes instead of auto-calculating ROI.
Enterprise AI spending has become a blank check with no memo line. Gartner reports that 84% of finance leaders have been unable to measure the ROI of their AI initiatives, and among those who tried, roughly two in five actually succeeded.
Apptio’s answer isn’t a smarter calculator. It’s an accounting discipline borrowed wholesale from Technology Business Management: register every AI initiative as a tracked object, merge its costs (cloud, labor, licenses, tokens) into one TCO model, then let the customer define what “value” means and watch the numbers argue it out over time: baseline, target, actual, realized.
What It Actually Fixes: Attribution, Not Prevention
Apptio doesn’t stop the bleeding. It totals the invoice afterward; baseline, target, actual, realized, tracked per AI initiative through the same allocation engine (cost pools, driver logic, Tower/Sub-Tower structures) Apptio has long used for cloud and IT spend generally.
IBM Cloudability supplies anomaly detection and token attribution; Apptio AI Value & ROI translates that into board-meeting language once the initiative is already scoped and running. IBM’s own documentation claims the underlying TCO layer can “detect cost anomalies early,” but early detection inside a monthly or quarterly reporting cadence is not the same instrument as a real-time governor; the mechanism that would have capped a runaway task before it, rather than months after, generated seven figures in unplanned charges.
One unnamed company reportedly burned through roughly $500 million in Claude credits in a single month after failing to set usage limits. Amazon engineers reportedly logged a routine listing-matching task that ran 860% over budget, generating $1.8 million in charges before anyone noticed, five months later. Neither failure mode is what Apptio’s launch is built to catch first. Both are exactly what it’s built to explain after the fact.
The Metric Still Has to Come From You
Apptio’s “proof metrics” (cycle time, cost avoided, conversion rate, incident volume) are customer-selected, not discovered by the platform. That’s a genuine structural improvement over raw usage counts, but it doesn’t eliminate the incentive problem it’s meant to fix.
IBM itself has already made this exact case elsewhere: in a separate IBM Think piece on “tokenmaxxing,” Neil Dhar, SVP for IBM Consulting, wrote that “in the absence of real metrics, organizations built usage leaderboards, which people quickly learned to game.” The same piece credits Marc Boroditsky, CRO of Nebius, with coining the counter-term “valuemaxxing” — shifting the question from how many tokens were spent to how much business outcome was produced, which is functionally the same pitch Apptio AI Value & ROI is now shipping as a product.
The risk repeats one layer down: a badly chosen proof metric is just as gameable as a token leaderboard, only with better formatting. The platform narrows how value gets measured. It doesn’t validate whether the chosen measure is actually load-bearing.
Everything Outside the Ledger
Reporting runs at the initiative and business-unit level; coarse enough that a single-task anomaly could still take months to surface through the aggregate, the same lag pattern that let the Amazon overrun run unchecked.
And IBM’s own framing cuts both ways: Bill Lobig, VP of IBM Apptio, said in the launch press release that one client saw a 50% reduction in costs, which “unlocked funding for new AI initiatives” — treating savings as straightforward reinvestment fuel. Read against IBM’s own cited scarcity problem, that’s closer to feeding the appetite than curbing it.
Beyond the Ledger: The Wider Visibility Gap
Apptio’s launch answers a cost-accountability question. It doesn’t touch two adjacent ones that are arguably just as consequential, and both are laid out in detail in our recent analysis, “Capability Without Visibility: What You Pay for Generative AI Workspaces Is More Than Money.”
The cost-accountability problem Apptio addresses is only one of at least three co-active “visibility gaps” (cost, retrieval, and session-state) all produced by the same underlying condition: capability shipped well ahead of the instrumentation needed to govern it.
On the cost side, we trace the same leaderboard-gaming dynamic IBM’s own Dhar names, citing reports that a single Disney employee interacted with Claude 460,000 times over nine days, that Meta’s internal dashboard awarded “Token Legend” tiers to its heaviest users, and that Uber’s engineers burned through the company’s entire annual AI budget in about four months without establishing whether rising token use correlated with any actual gain in output.
None of that is solved by better attribution. It’s a governance and incentive-design problem sitting one layer upstream of any ledger.
The other two gaps sit entirely outside Apptio’s scope.
Retrieval visibility covers what happens when an AI system’s citations resolve to dead, fabricated, or misattributed sources, or silently truncate a document’s most relevant section — failures no cost-allocation tool is positioned to catch, because they’re correctness failures, not spend failures.
Session-state visibility is stranger still: I personally documented a case where an agentic tool repeatedly overwrote its own standing instructions against explicit configuration, a bug with a public paper trail on GitHub. Even SOTA models today still can’t tell you what it’s doing wrong, even when you ask it directly, when it comes to these multi-layered, multi-faceted micro-infrastructures like session-state persistence. That’s not a line item any FinOps platform, including Apptio’s, is built to surface.
Is it still a good step? Yes, it’s a real, albeit baby, step.
Where Apptio’s launch treats the accountability gap as something a better ledger can close, our piece argues the deeper gap is structural: that an organization delegating work it cannot audit mid-task, in real time, has a governance problem no invoice, however well-attributed, actually resolves.
Aklatan’s news and analysis drills down to the structural mechanics, geopolitical shifts, and hidden constraints truly driving AI and Asian tech ecosystems and knowledge work.
See coverage span here: Aklatan’s News and Analysis
Generative AI Transparency:
This news article was written primarily with generative AI, specifically SupraGraphos’ A.C.E. News Module. Reviewed with human post-editing, all sources and claims are confirmed as of the time of writing.