For CEOs, AI ROI is a board-level concern – with a new solution

What does AI attribution actually look like in practice? It’s a question that can have different answers depending on where you are sitting.

From engineering's side, it's about visibility: Which AI tools are actually earning their spot, is output improving, and how do we know where we need to improve?

From finance's side, it's an attribution problem: A bill arrives, and there's no credible way to connect it to the initiatives it supported and point to actual data, projects, and results that come from that specific spending. 

For the CEO, the battle is how to chart a clear path in an unmapped ocean. There is a demand to ramp up productivity, adopt more AI tools like your rivals are, and be a cutting-edge company. At the same time, you’ve got a new and rapidly growing line in your budget that needs justifying – and no hard data to do that. 

This article is written for chief executives who approved an AI budget, told the board what it would do, and now have to show what it did. Here at Tempo – we’ve been developing the answer: Workforce Intelligence. 

Why this matters beyond the board deck

Forty-seven per cent of large organizations lack full visibility into which AI tools their employees are even using, according to Protiviti's 2026 AI Pulse Survey. 

However, only 7% of tech leaders have established ROI reporting on their ROI investments according KPMG’s Global AI Pulse Survey.  Andrew Macdonald, Uber's COO, put the problem plainly when discussing AI productivity statistics: "That link is not there yet… It's very hard to draw a line between one of those stats and, 'Okay, now we're actually producing 25 percent more useful consumer features.'"

One of the common solutions has been a cap – a hard per-engineer token allowance, a frozen budget, licences pulled back. Stories of that kind circulate widely enough that most executives have heard one: A nine-figure annual AI coding bill that could not be defended, an entire annual AI budget consumed and then capped per engineer, an "unlimited tokens" pool exhausted in 38 days.

Those are market anecdotes rather than customer results, and the same reasoning runs through all of them. The AI kept working. The buyers capped it because they could not tell what it bought.

Bryan Catanzaro, who runs applied deep learning at Nvidia, put the scale of the problem in plain terms: "…For my team, the cost of compute is far beyond the costs of the employees." Adam Mosseri, who runs Instagram, framed where the broader market is heading: "…the cost of using AI tools may soon reach a level comparable to the cost of hiring the engineer themselves."

When a category of spend approaches headcount economics, it deserves headcount-grade accounting. You know precisely what each engineer costs and which team they serve. AI arrived with none of that discipline attached to it, and a cap is the only instrument available to a leader with no attribution.

What AI attribution means for a chief executive

Attribution is a finance word that has spent most of its life in marketing. Applied to AI, it means something narrower and more useful: Every token, every compute charge, and every seat licence gets assigned to the piece of work it helped produce, and that assignment survives the trip upward through your hierarchy.

The differentiator is the roll-up. AI attribution travels through every level – epic, initiative, OKR, strategic objective – rather than stopping at engineering output. Your CFO reads the initiative view. Your CTO reads the epic view. You read the objective view, and all three are the same record at a different height.

That changes the conversation you have with your board. Adoption statistics describe activity. A quarterly bill describes cost. A director hearing either one still has to ask whether the money moved the company toward the outcome you promised.

We know the importance of being a company with an accurate picture of itself, because without that, you don’t have a full understanding of what your organization is capable of.

Workforce Intelligence gives that picture inside Jira, where the work already lives, without asking anyone to change how they work.

What is Workforce Intelligence?

Tempo Workforce Intelligence optimizes AI productivity across your workforce by giving you traceable ROI tied to work delivered. 

The design idea behind it is straightforward: Workforce Intelligence records AI cost against the same Jira work item that already carries human effort, which makes it the first application to put both on one record.

That is AI attribution in practice. A dashboard that can show you what tools are in use, by what team, and what it actually delivered, with figures. 

The product organizes around three moves.

See it: What is this spend really delivering?

Native connectors reach your AI tools, Git, and Jira, with token and compute cost pulled straight from provider APIs. Claude Code, GitHub Copilot, and Codex go on one side; every dollar comes out traced to a ticket, an epic, and an initiative. A verified receipt.

The traced total includes the spend tied to no work at all. That figure tends to be the one a chief executive reads first, because unattributed spend is the portion of your AI budget with no defence available in a board meeting.

Steer it: How is AI affecting delivery?

Human time and AI activity land in one work record, which lets you deploy people and agents deliberately rather than by default. What did this kind of work cost with a person, and what did it cost with an agent? Which objectives are getting genuine AI lift, and which are being funded on the assumption that they are?

That second question belongs to the chief executive alone. Your product organization sees its own roadmap. Your engineering organization sees its own throughput. The view of which strategic bets receive AI investment and which go starved spans both, and it determines where you reallocate next quarter.

Evolve it: Which AI tools should be in use, by whom?

Verified cost rolls up from work item to portfolio, and comparison becomes defensible across five axes: Team against team, quarter against quarter, tool against tool, model against model, and AI-assisted work against everything else. Each comparison runs on the delivery data your organization already trusts rather than a vendor's proprietary score.

The practical consequence is that a lagging team gets diagnosed rather than dismissed. When Team A outperforms Team B, attribution tells you whether AI adoption explains the difference or whether something else does. 

What each side gets

Your CTO gets an architecture built for a permanent multi-tool reality. Session cost from Anthropic and OpenAI at the source. GitHub Copilot and 21 other AI tools detected from commit metadata. Developer machines untouched. Attribution built into the Jira issue itself – and no engineer 

Your CFO gets attribution that holds up under audit. Per-ticket traceability, exports at the work item, story, epic, initiative, and OKR level, and a mechanism for moving the capitalizable share of AI spend from an estimate on the P&L to a traced number on the balance sheet.

Both sides read the same record. When your CTO argues for a new AI tool, your CFO can answer with cost per outcome. When your CFO asks whether AI is producing what it costs, your CTO can point to the initiatives it moved and the ones that stayed flat. The conversation stops being a negotiation.

Where this goes next

Whether you're the one presenting to the board, the one signing off on the invoice, or the one accountable for what the whole organization delivered, the underlying need is the same: Numbers that agree with each other, and a leader who can trust what's underneath them.

If you're an engineering leader, a finance leader, or the person who ends up owning both questions at once, Workforce Intelligence is open for early access to a small group of organizations working through exactly this problem. 

We'd rather show you what it looks like against your own data than tell you more about it here.

Workforce Intelligence is available today with a free 30-day trial against your own Jira and AI tools.