# Enterprises can measure AI usage, but the hard part is proving that it actually delivered value > Source: > Published: 2026-08-25 10:01:00+00:00 Enterprises are accelerating their AI investments and deploying agents, budgets are ballooning out of control, and leaders are being asked to justify the cost. Yet insight into the return on investment (ROI) can be opaque. Tempo says its new Workforce Intelligence (WFI) offering can help product managers make the case for, and optimize, their AI spend. The collaborative workspace platform provider says that WFI is the first Atlassian Marketplace app that automatically connects AI tool activity directly to Jira work items, tasks, epics, and initiatives to help leaders understand AI use, cost, its productivity impacts, and where the tools actually deliver ROI. “The amount of money people are spending on AI is enormous, and a very large percentage of it is wasted,” said Tempo CEO [Vic Chynoweth](https://www.tempo.io/news/news-tempo-welcomes-new-ceo-to-lead-the-next-chapter-of-growth-and-innovation). “Being able to orient your investment toward outcomes you know are working is going to be a big lift for organizations.” According to IBM, only 29% of executives can confidently measure AI ROI, and just [25% of AI initiatives](https://www.ibm.com/think/insights/ai-roi) actually deliver expected ROI. And pressure is only increasing; Kyndryl reported that [61% of senior business leaders](https://www.cio.com/article/4114010/2026-the-year-ai-roi-gets-real.html) feel more burdened to prove AI ROI than they did just a year ago. Chynoweth describes the situation as being stuck between two rocks: “I’ve got to move faster and [deploy AI](https://www.infoworld.com/article/4209927/the-five-walls-standing-between-a-demo-agent-and-a-deployed-one.html)” and, at the same time, “I need to moderate my AI spend.” “Once companies started deploying, things got real expensive real fast,” he said. “I’ve now overspent my budget because nobody had any idea what it was going to cost, and the costs are only going up, not down.” Existing AI analytics tools measure prompt, token, and license usage, as well as code output, adoption percentages, and aggregated spend, yet they operate outside the system of work, Chynoweth noted. WFI, on the other hand, establishes what Tempo calls a “new layer of attribution” for [AI-powered work](https://www.infoworld.com/article/4209900/agentic-ai-in-the-enterprise-how-to-balance-autonomy-with-constraints.html). Now available to any enterprise in the Atlassian ecosystem, WFI measures AI contribution to provide verified attribution, rather than just estimated use. The platform is built around three layers: AI cost captured directly from the provider; attribution to the specific Jira work item it supported, not just a project or team; and a cost view that combines AI spend and human labor cost in the same record. “We call it workforce intelligence because it’s [people and AI](https://www.cio.com/article/4211674/when-ai-explains-its-decision-humans-may-stop-thinking-independently.html),” Chynoweth explained. “It’s not just agents. Work is delivered by people and agents today, so it’s optimizing the two of those.” Enterprises do not need Tempo products to use WFI. It is a standalone Atlassian Marketplace app for any team with an active Jira Cloud installation that uses one or more supported AI coding tool. These include OpenAI Codex, GitHub Copilot, and Claude Code. The platform ties human and AI activity to specific Jira issues and initiatives, and links inference API telemetry, combined human and AI effort, and cost rollup in one work record. Leaders get a “work item-level view” of AI activity, alongside recorded performance and cost data. “When you triangulate all that, you can actually connect ‘Here is what I delivered, and here is what I spent, both people and AI,’” Chynoweth said. WFI allows enterprises to answer questions like: What is AI actually costing per initiative? What tools are in use and how are they actually improving work? What does productivity look like in AI-assisted workflows versus non-AI-assisted ones? Tempo can turn WFI on for customers within hours, Chynoweth claimed. Once integrated, the platform has access to 90 days of historical data and can generate insights and recommendations based on prior usage and tasks. It can also recognize when data is incomplete, misaligned, or not tied to strategic initiatives, and can recommend ways to improve data quality. “It’ll have three months of history,” Chynoweth said. “This is what you got out of all of your LLM investments next to your people. Here’s what you then go do with that.” Enterprises have the ability to identify the teams and methodologies that are generating meaningful outcomes versus those that are lagging, he said. They can also catch scope drift before it compounds. Additionally, enterprises gain insight into their LLM use so they can optimize routing. Models are continuously evolving, and different LLMs are better for different work types; a frontier model can help scale new innovations, for instance, while a smaller version can be optimized for existing product maintenance, Chynoweth noted. “We are able to test the efficacy of the different models and choose not to deploy one that’s more expensive because it doesn’t garner a relative return,” he said. “It’s really about bifurcating, scaling, and continuing to manage at scale and continuing to experiment, but being purposeful.” WFI is priced to scale with organization size; pricing and packaging details are available directly through Tempo. The company plans to expand attribution coverage to additional source control and IDE integrations. Existing tools track how much LLMs spend, but not necessarily in real time, Chynoweth noted. For instance, they could report that an enterprise has an 100% adoption rate across its workforce. But, he said, “hold on a second, what does that even mean? That means 100% of people have touched an LLM. That doesn’t tell me anything.” Similarly, usage tracking platforms tell teams how much code they’re generating, or how many pull requests (PRs) they’re delivering. But that doesn’t provide much insight either, because AI has essentially commoditized code, Chynoweth said. “We generated code that may or may not have been useful.” “Initially it was a scatter shot: Let’s try it everywhere,” he noted. But enterprises now need to identify places where AI can be truly effective, scale those, and be purposeful about testing and experimenting. It is about deploying and scaling the things that are working, and “backing off from the scatter shot.” He added that, in his recent customer discussions, AI ROI “is the single most important, biggest talked-about item.” The prevailing opinion: Deploy AI or fall behind.