As enterprise adoption of generative AI instruments accelerated by late 2025 and into 2026, expertise leaders started to hit a irritating wall. Whereas otheir rganizations poured tens of millions into AI tokens and mannequin subscriptions, company management and CFOs started urgent for onerous proof that this skyrocketing spend was delivering precise enterprise worth. Immediately, regardless of widespread integration of developer assistants and automatic instruments, organizations proceed to wrestle with figuring out whether or not their big investments in AI tokens interprets into significant product outcomes, or merely inflated operational prices.
Within the preliminary rush towards AI integration, engineering departments typically relied on uncooked utilization metrics—reminiscent of token consumption—to judge adoption success. Nonetheless, excessive token quantity shortly proved to be a poor proxy for real productiveness.
“As we did that, one of many issues that we noticed is our spend simply went by the roof as we adopted that,” mentioned Shams Chauthani, Chief Know-how Officer at Tempo.io. “And the query that our CFO began asking us is… ‘What are we getting for all these items that we’re doing?’”
Final result metrics don’t inform the entire story
As the constraints of “token maxxing” turned clear, the trade transitioned to monitoring output metrics, reminiscent of traces of code generated or pull requests submitted, utilizing engineering administration instruments like Atlassian DX and Jellyfish. Whereas these manufacturing metrics gave engineering managers perception into developer exercise, they didn’t reply government questions on enterprise worth. Producing code sooner didn’t routinely result in transport strategic options or enhancing software program high quality, and it typically penalized builders spending time on vital duties like resolving technical debt.
“In case you’re measuring what number of traces of code you wrote, AI is nice about writing tens of millions of traces of code very very quick. However ‘Did you really ship worth or not?’ was the query that was actually onerous to reply,” Chauthani famous.
Workforce Intelligence platform
To bridge this hole between engineering exercise and monetary accountability, corporations are in search of methods to attach AI spend on to strategic enterprise models of labor. Tempo lately tackled this problem with the launch earlier this month of its Workforce Intelligence (WFI) platform, to provide organizations granular visibility into how AI investments affect product supply.
Reasonably than token counts or uncooked code quantity in isolation, WFI correlates token spend knowledge from mannequin suppliers like OpenAI and Anthropic with GitHub code commits and maps them on to Jira tickets, epics, and initiatives.
“We mainly mentioned, what’s the unit of measure of productiveness and product supply that we’re ? And usually, what that’s is Jira in our case, or any ticket administration system,” defined Chauthani. “If we are able to tie the dots between what AI spend occurred and what ticket was it tied to, we are able to now impulsively get a visibility into [how] this AI spend actually drove this end result for you.”
This stage of attribution is changing into important as AI bills develop to characterize 20% to 30% of total R&D budgets. In keeping with the Tempo 2026 State of AI report, 91% of expertise leaders presently utilizing AI report that they’re unable to delegate work to AI and tie it on to tangible outcomes. By combining AI price monitoring with human labor monitoring—a site Tempo has addressed for twenty years—organizations can consider which fashions are most cost-effective for particular duties, whether or not refactoring technical debt or constructing new capabilities.
“Simply giving the AI spend is simply a part of the image,” Chauthani defined. “You want the human spend and AI spend collectively, and the power to roll that data up in a significant method, the place any person can really make selections off of that.”
