> For the complete documentation index, see [llms.txt](https://help.swarmia.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.swarmia.com/features/ai-tools/ai-impact-roi.md).
# AI impact: ROI
Compare the combined cost of your AI tools and engineering work against the stories, pull requests, and code your teams deliver.
Available at [AI tools → AI impact → ROI](https://app.swarmia.com/ai/impact/roi)
The ROI (return on investment) tab puts cost and output side by side, so you can see what your organization spends per unit of delivered work and whether that number is moving. It combines [AI token usage](/features/ai-tools/ai-cost.md#ai-token-usage) from your AI coding tools with the estimated cost of engineering time, then divides it by the stories, pull requests, and code your teams ship.
This helps you answer questions your finance and leadership teams tend to ask: What are we spending per story or pull request? How much of our engineering cost is AI tools now? Is the AI spend translating into more output per engineer?
Pick a timeframe and optionally a team to drill into. Everything is bucketed by calendar month, and all amounts are shown in US dollars.
### Total cost
The first chart stacks **developer cost** and **AI token usage** per month, so you can see how the two move relative to each other. Alongside it you get the timeframe totals: developer cost, AI token usage, AI token usage per FTE per month, and AI token usage as a share of total cost.
### Throughput and cost per unit
Each section pairs a measure of output per FTE per month with its matching cost per unit:
* **Issue throughput** shows completed stories or epics and their cost. Use the **Stories** and **Epics** tabs in this section to choose the issue type.
* **PR throughput** shows merged pull requests and their cost.
* **Code throughput** shows lines changed and their cost. Select **Exclude test files** in this section to measure non-test lines instead. Swarmia identifies test files using your [test file rules](/settings/organization/pull-request-data-quality/pull-request-test-file-rules.md).
Throughput is divided by [developer effort](/definitions/developer-effort-ftes.md) rather than headcount, so growing or shrinking teams don't distort the trend. The cost charts stay stacked into developer cost and AI token usage, which shows you whether a change in unit cost comes from AI token usage or from engineering time.
### Team breakdown
The table at the bottom breaks the same metrics down by team. Select a team to drill into its child teams, and download the table as a CSV.
## How the costs are calculated
Both cost components are estimates, and each has its own caveats:
* **AI token usage** is the list price of tokens consumed by your AI coding tools, based on the [AI coding tool integrations](/settings/integrations/ai-coding-tool-integrations.md). It isn't your invoice, and it doesn't include seat or subscription fees.
* **Developer cost** is estimated from [developer effort](/definitions/developer-effort-ftes.md) and an organization-wide average: a team's monthly FTE multiplied by one twelfth of your configured average yearly developer cost.
To get developer cost, an [organization admin](/settings/organization/managing-users-and-roles.md) needs to turn on **Show estimated cost** and set an **Average yearly developer cost** in [Settings → Organization → General](https://app.swarmia.com/settings/organization/general). The figure should cover salary, benefits, and overhead. Until it's set, the ROI tab shows AI token usage only, and the cost-per-unit metrics reflect AI token usage alone. Estimated developer cost is visible to admins only.
If you configure the average developer cost in euros, Swarmia converts it to US dollars at a fixed rate so both cost components share one currency.
## What good looks like
Start with the AI cost share, because it tells you where your effort is worth spending. If AI spend is a small fraction of developer cost, the upside is in getting more out of the tools, not in trimming the tool bill: shaving a few percent off a small AI cost barely moves the total, while a real gain in throughput per FTE does. Cutting AI cost is worth the attention once it makes up a significant share of your total cost.
From there, read the tab as a trend rather than as a single month.
### Is the AI spend paying off?
* **AI cost share climbing while cost per unit falls.** You're trading engineering time for tool spend and getting more out per unit of effort.
* **AI cost share climbing while cost per unit holds or rises.** The spend isn't translating into output yet.
### How has the shape of the work changed?
The three unit costs often move in different directions. Read together, they tell you what changed about the work, not just whether it got cheaper.
* **Cost per line changed falls while cost per merged pull request rises.** Your pull requests are getting bigger: AI produces more code per change, but each change still needs a review. Check [batch size](/guides/improve-pull-request-flow/analyzing-pull-request-batch-size.md) and review time in [AI impact: Code](/features/ai-tools/ai-impact-code.md) to see whether review has become the constraint.
* **Cost per merged pull request falls while cost per story holds.** The code is flowing faster, but there's no clear increase in customer-facing increments. That points at something outside coding, such as scoping.
### What the numbers don't tell you
Cost per issue is the closest thing here to a business outcome, so it's also the one worth questioning: does one story or epic still represent the same amount of customer value it did a year ago? Teams that change how they work with AI tools might change how they split and size issues too.
Treat the unit costs as directional rather than exact. They measure output, not the value of what you shipped, so pair them with quality signals like [DORA metrics](/features/metrics/track-dora-metrics.md). For the wider picture, see the guide on [understanding the impact of AI tools](/guides/understand-the-impact-of-ai-tools.md).
## Calculating ROI
If you need a single ROI (return on investment) figure on your AI spend, compare a current period with a deliberate pre-AI baseline:
`ROI = ((current throughput per FTE / baseline throughput per FTE − 1) × current developer cost − current AI token usage) / current AI token usage`
Use the same output measure (lines, PRs, stories, or epics), timeframe length, and comparable work in both periods. Use a baseline period where AI wasn't in use and AI token usage was zero (or negligible). Note that a baseline can become stale as seasonality, work mix, team composition, or processes change, and marginal ROI from small token-usage changes is noisy. Use a controlled rollout or cohort comparison where possible.
## Requirements and limitations
* [GitHub Copilot](/settings/integrations/ai-coding-tool-integrations/github-copilot-integration.md), [Cursor](/settings/integrations/ai-coding-tool-integrations/cursor-integration.md), or [Claude Code](/settings/integrations/ai-coding-tool-integrations/claude-code-integration.md) integrations are required for AI token usage.
* Issue throughput and cost per story require an [issue tracker integration](/settings/integrations/issue-trackers.md), with issues mapped to the story and epic issue types.
* Lines changed counts merged pull requests and excludes [auto-generated files](/definitions/batch-size.md#excluding-auto-generated-files).
* FTE follows the usual [effort rules](/definitions/developer-effort-ftes.md): effort counts the working days someone was active, and each person contributes at most 1.0 FTE per month. Authors need their Git and AI tool accounts connected on the [Authors](/settings/organization/contributors.md) page for their AI token usage and effort to land on a team.
* The current, incomplete month is included in the charts, so the latest data point moves as the month fills in.
## Frequently asked questions
### Why don't the figures match what we pay for AI tools?
AI token usage is the list price of the tokens your tools consume, including the harness. Seat and subscription fees aren't included, and on plans where heavy usage doesn't increase your bill, it won't match your invoice.
### Why is developer cost missing?
Either cost estimation isn't set up yet, or your [user role](/settings/organization/managing-users-and-roles.md) doesn't allow you to see estimated cost. Admins can turn it on in [Settings → Organization → General](https://app.swarmia.com/settings/organization/general).
## Further reading
[Measuring AI ROI takes more than one number](https://www.swarmia.com/blog/measuring-ai-roi-more-than-one-number/) by Miikka Holkeri, Product Manager · Sep 28, 2026
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.
Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:
```
GET https://help.swarmia.com/features/ai-tools/ai-impact-roi.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
