> ## Documentation Index > Fetch the complete documentation index at: https://docs.weaveos.com/llms.txt > Use this file to discover all available pages before exploring further. # AI ROI > Measure the return on investment of your AI coding tools Weave measures the return on investment (ROI) of AI coding tools such as Claude Code, Cursor, GitHub Copilot, and Codex by dividing what you spend on them by the [code output](/metrics/code-output) attributed to them. That cost per output appears next to the pre-AI cost of the same output, one hour of expert engineering work at your hourly rate, so you can compare the two directly. Spend comes from provider billing data, list-price token estimates, and the seat costs you enter, and each dollar is labeled with how it was measured. ## Who can see it? Anyone with the **Team lead**, **Manager**, or **Admin** role can access the AI ROI report. ## Page layout The AI ROI page respects your date, team, and repository filters and defaults to monthly granularity. It shows: * **Cost by fidelity**: your full AI spend for the period, usage plus contracted seat cost, split by how each dollar was measured, from vendor-billed **Actual** to list-price **Estimated**. It's the only view with the full total. Click a source to drill into its spend. * **Team spend flow**: how that spend flows from your organization to its top teams. * **Projected annual AI spend**: your last 90 days of AI tool spend, seat costs included, annualized, with the 30-day run rate for comparison. Admins can use **Upload** to add a Copilot spend sheet or a ChatGPT Enterprise workspace report (both CSV). Cost per output and the per-person and per-tool breakdowns are on the [AI report](/reports/ai) overview: * **AI output & spend**: AI-attributed code output next to usage-based token spend for each period. Seat costs aren't included, so it can read lower than **Cost by fidelity**. * **Cost per output over time**: AI tool cost, including prorated seats, divided by AI code output, for all tools and for each tool, with a **Pre-AI cost per output** line at your hourly rate. Its toggle switches to **AI cost over time**. * **Cost per person**, **AI cost by tool**, and **Cost per output by tool**, described below. ## What's included ### Cost per person The **Cost per person** panel on the AI report overview ranks people by AI spend. Each person's bar is their total cost (the value of what they used, priced at the best available source), and beneath it a split shows how much is **billed** (vendor-confirmed) versus **estimated** (list-price approximation). Under the **Total cost for time-period** sort, the rows sum exactly to the **Cost by fidelity** total: spend the chart can't attribute to any person — usage with no linked member, plus contracted seat cost, which is billed per plan rather than per person — appears in an **Unattributed** row at the bottom. Click a person to open a drill-down of every usage row, each tagged with a **Fidelity** label (Actual, Estimated, Client-reported, Off-workspace usage, Modeled, Activity only). The footer shows **Total cost** (the sum of those usage rows) and, when applicable, a separate **Contracted seats** subtotal for day-prorated seat charges — seat cost is not folded into Total cost. See [AI tool costs](/metrics/ai-tool-costs#cost-fidelity) for what each fidelity label means and why a list-price total can differ from what's invoiced on flat-rate plans. ### AI cost by tool The **AI cost by tool** panel on the AI report overview ranks your AI tools by total spend for the selected period. Each tool's figure is collected per that tool's source, whether that's an actual provider invoice or an estimate (see [AI tool costs](/metrics/ai-tool-costs#cost-by-tool) for where each tool's number comes from). How a **flat-rate** tool (a fixed monthly seat plan) is shown depends on whether you've set an **included allowance** for its plan under **Settings → Financial** (see [Included allowance](/metrics/ai-tool-costs#included-allowance-for-flat-rate-tools)): * **No allowance set** — the tool shows its **list-price consumption**: the estimated value of the tokens run through it. It's an estimate and doesn't account for the plan's usage limits, so it can read higher than the flat fee you actually pay. * **Allowance set** — the tool shows its **seat cost plus overage**: the flat seat fee plus only the list-price usage above the allowance the seats cover. Usage-priced and hybrid tools show contracted seats (day-prorated to the date range) plus their metered/usage cost. Click a tool to drill into the users, days, and models behind its cost, including its seat / base-cost rows. Next to it, **Cost per output by tool** divides each tool's cost by the AI code output attributed to it. ### Adoption and impact For AI adoption and AI-attributed output, see the [AI report](/reports/ai) and [AI output percentage](/metrics/ai-output-percentage). For how well engineers work with AI tools, see [AI skills](/reports/ai-skills). ## Setting up cost tracking To track AI tool costs, configure your subscriptions: Navigate to **Organization settings** → **Financial** Enter your subscription tier and seat count for each tool To price the **Pre-AI cost per output** line without per-engineer salary data, enter your **annual engineering team cost** on the same Financial page. Weave derives an effective hourly rate from that total and your engineering headcount. When unset, ROI falls back to [HR salary data](/data-sources/hr#financial-metrics-and-hourly-rates) or your organization default hourly rate. Return to the AI ROI report for spend, and to the [AI report](/reports/ai) overview for cost per output ## Common questions ### How does Weave calculate AI ROI? Weave divides total AI tool cost, meaning usage plus seat subscriptions prorated to the date range, by the code output attributed to AI in the same period. One unit of output is one hour of expert engineering work, so the **Cost per output over time** chart on the [AI report](/reports/ai) overview draws a **Pre-AI cost per output** line at your hourly rate for comparison. ### Can I show that our Claude and ChatGPT spending is worth it? For coding tools, yes. Claude Code and Codex are broken out separately in the cost per output charts, so you can compare their cost per unit of output with each other and with your hourly rate. Chat usage is different, because code output is measured from pull requests: the ChatGPT Enterprise workspace report that an admin uploads through the AI ROI page's **Upload** menu counts messages per user, with no cost or code output. ### Can I compare our AI cost per output with other companies? Yes. The cost per output chart can show your percentile rank and benchmark lines drawn from other organizations that use Weave, either of a similar size or, if an admin selects it, in your industry. See [benchmarks](/annotations/benchmarks) for how peer groups work. ### Why is token spend lower than total AI spend? Token spend counts usage-based charges only. Seat and subscription costs you enter under **Settings → Financial** are added in **Cost by fidelity** on the AI ROI page, which is the only view with the full total. ### Are AI code review tools counted in cost per output? No. Review-only tools such as CodeRabbit, Cursor Bugbot, and Greptile are left out of cost per output so their spend doesn't skew the ratio. See [cost by tool](/metrics/ai-tool-costs#cost-by-tool) for where each tool's cost comes from. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.