# Measuring AI ROI Through Business Outcomes > Source: > Published: 2026-09-22 16:10:07+00:00 # Measuring AI ROI Through Business Outcomes In late 2025 and early 2026, development teams integrated AI coding assistants into daily workflows, resulting in rapid increases in operational expenditure. As AI spend expanded across multiple teams and projects, chief financial officers began demanding evidence of return on investment. Initial efforts to evaluate AI adoption relied on consumption metrics. Organizations tracked active user counts, license allocations, and token consumption rates. However, tracking token usage indicated computational activity rather than productivity or financial return. Relying on consumption data failed to answer the core financial question: What business deliverables did the token expenditure produce? To move beyond consumption tracking, engineering management turned to output-based production metrics. Engineering platforms began tracking metrics such as lines of code written, commit counts, and pull requests submitted. While output metrics provided visibility into developer activity and tool utilization, they introduced measurement distortions. Highlighting the shift in evaluation metrics, Shams Chauthani, CTO at Tempo, told SD Times, “My CFO was still going like, I don’t understand… what did we actually get from this? And so that’s where we started internally developing a way to tie the spend to the outcomes that business cares about.” He added, “The business cares about how many features and functions were shipped. What initiatives were we taking on that we were actually delivering on? How much did those initiatives cost? Did we actually get benefit by using AI in those initiatives versus not?” Determining AI ROI requires linking financial expenditure to business outcomes. In software engineering, work management systems like Jira define the structural units of product delivery. Strategic initiatives break down into epics, features, and task tickets. To achieve accurate cost tracking, organizations must connect token spend directly to work units. Connecting API cost data from model providers such as OpenAI and Anthropic to GitHub commits and Jira tickets allows management to calculate the precise expenditure required to complete individual features and initiatives. Mapping token costs to work items enables leaders to evaluate spend in relation to completed deliverables rather than raw activity. Solutions such as Tempo’s Workforce Intelligence (WFI), introduced following early access programs in mid-2026, resolve this tracking gap by embedding ROI measurement into project management workflows. By establishing connections between model spend, source code changes, and completed Jira tickets, organizations replace speculation with data-driven resource management. Tracking outcomes provides clear visibility into investment efficiency across engineering teams. Chauthani explained, “If we can tie the dots between what AI spend happened and what ticket was it tied to, we can now all of a sudden get a visibility into this AI spend. What really drove this outcome for you, and and our CFO can look at that and go, ‘Oh, I understand. Of the initiatives, we spend 25% on this one and 10% on this, so on and so forth, and actually get measurable results from it.” Evaluating R&D performance requires analyzing human and artificial resource allocation within a single system. Historically, tracking tools measured human hours spent on specific development tasks. Incorporating AI cost data into existing workforce tracking platforms creates a unified view of total engineering investment, where AI expenses represent 20 to 30 percent of overall R&D budgets. Unified investment tracking allows engineering leaders to compare cycle times, historical performance, and model efficiency across different task categories. Teams can evaluate whether tasks such as technical debt remediation are best completed using open-source models, premium proprietary models, or human effort. Industry data indicates that 91 percent of engineering leaders using AI cannot delegate tasks to AI while linking those tasks to verified outcomes. Addressing this limitation requires shifting focus from code production to outcome delivery. By establishing connections between model spend, source code changes, and completed Jira tickets, organizations replace speculation with data-driven resource management. ##### SD TImes Q&A ##### How do you calculate AI ROI in software engineering? Calculating AI ROI in software engineering requires linking model spend (API token costs from providers like OpenAI or Anthropic) directly to completed work units such as Jira tickets, epics, and features. This lets engineering leaders calculate the precise cost to deliver each feature or initiative, rather than relying on activity metrics like token consumption or commit counts. Comparing AI-assisted vs. non-AI-assisted cycle times and costs provides a financial basis for ROI reporting. ##### Why are token usage metrics insufficient for measuring AI coding assistant ROI? Token consumption metrics reflect computational activity, not business output. They show that an AI model was used but do not link that usage to delivered features, closed tickets, or completed initiatives. CFOs and engineering leaders need spend tied to deliverables — not raw usage — to make investment decisions. ##### What percentage of R&D budgets are AI costs expected to represent? According to the article, AI expenses are estimated to represent 20 to 30 percent of overall R&D budgets as development teams scale adoption across multiple projects and teams. A source citation for this figure was not provided in the original reporting. ##### How can engineering teams connect AI spend to Jira tickets? By integrating API cost data from model providers with source code commits and work management systems like Jira, teams can map token expenditure to specific tickets, features, or epics. Tools such as Tempo’s Workforce Intelligence (WFI) are designed to automate this linkage within existing project management workflows. ##### What are output-based metrics in AI development measurement, and what are their limitations? Output-based metrics include lines of code written, commit counts, and pull requests submitted. While they provide visibility into developer activity, they can introduce measurement distortions — for example, incentivizing volume of commits over quality of deliverables. They still do not directly answer whether AI spend produced valuable business outcomes.