> ## Documentation Index
> Fetch the complete documentation index at: https://help.nops.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Scores

> What Velocity, Productivity and Efficiency measure, how each is calculated, and what moves them.

AI ROI sums up a quarter in three scores, each out of 100. You see them for the organization on the [Executive Summary](https://help.nops.io/ai-roi/executive-summary.md), for a team on the [Engineering Overview](https://help.nops.io/ai-roi/engineering-overview.md), and for a person on [Developers](https://help.nops.io/ai-roi/developers.md).

| Score | The question it answers |
| - | - |
| **Velocity** | How fast does work get from first commit to production? |
| **Productivity** | How much of our AI spend turns into shipped work, and at what cost? |
| **Efficiency** | How much of our AI spend was avoidable? |

## Bands

| Score | Band |
| - | - |
| 70 and up | **Good** |
| 40 to 69 | **Moderate** |
| Under 40 | **Poor** |

## How a score is built

Each score is a **weighted average of a few figures**. Each figure is placed on a **fixed curve** from 0 to 100, the same curve for every organization, team, person and quarter. That gives a score the same meaning everywhere: a 70 in Q4 is as good as a 70 in Q3, and a team's 70 means the same as the organization's.

### Velocity

| Figure | Weight | Meaning | 100 at | 0 at |
| - | - | - | - | - |
| **Lead time** | 50% | Median time from a PR's first commit to the release that carried it to production. | 2 hours or less | 30 days or more |
| **PRs / engineer** | 35% | PRs that reached production, per engineer per week. | 10 a week | none |
| **Output / engineer** | 15% | [Output points](https://help.nops.io/ai-roi/how-the-numbers-work.md#output-points) of PRs that reached production, per engineer per week. | 250 a week | none |

The curves are logarithmic, so early gains count more than late ones and a fast team doesn't max out. For reference, a lead time of 12 hours or less, 4.3 PRs a week or 47 output points a week each score about 70.

### Productivity

| Figure | Weight | Meaning | 100 at | 0 at |
| - | - | - | - | - |
| **Shipped** | 70% | The share of AI coding spend tied to PRs that reached production. | 80% or more | 20% or less |
| **Cost per PR** | 30% | AI cost per PR that reached production. | \$5 or less | \$500 or more |

Spend from the **last 14 days** of the range is left out of "Shipped", because it hasn't had time to reach a release yet.

The tile also shows the **median and 90th-percentile spend per engineer per month** (or, for one person, their own spend per month). These describe how spend is spread; they are not scored.

### Efficiency

Efficiency starts at 100 and loses **1.2 points for every percentage point of avoidable spend**.

Today, **avoidable spend is rework**: the AI cost of AI-assisted PRs that a later fix reworked or a revert undid within 14 days, plus the cost of those fixes. See [Rework](https://help.nops.io/ai-roi/how-the-numbers-work.md#rework).

The tile also shows the **rework rate**, the **avoidable dollars**, and the **cache read share** of tokens, for context.

## Definitions the scores share

* **Engineer**: someone with AI coding spend in the range who is [matched to a code host login](https://help.nops.io/ai-roi/teams.md#link-a-login-to-a-person). People who aren't matched can't be tied to PRs, so counting them would make every per-engineer figure look worse.
* **Reached production**: a PR that merged and was then carried by a release (a later release PR from the branch it merged into), or merged straight into a release branch. In a repository that has no release PRs, merging counts as reaching production.
* **Comparison**: for the organization, the same number of days into the previous quarter (**+6 vs Q3**). For a range longer than a quarter, the period of the same length just before (**vs before**). For a team or a person, the organization's score in the same range.

## When there is no score

A tile shows **—** and a reason instead of a number when:

| Reason | What it means |
| - | - |
| **Too few PRs** | Fewer than 5 PRs reached production in the range. Medians and per-PR figures say little below that. |
| **Not enough data** | The figures behind the score have no data yet, for example no AI spend or no code host connected. |
| Hidden for a small team | A team with fewer than 5 engineers shows no scores to people outside it. See [Who sees what](https://help.nops.io/ai-roi/access.md). |

## How to raise each score

| Score | Levers |
| - | - |
| **Velocity** | Release more often, so merged work doesn't wait. Keep PRs small. Remove review bottlenecks. |
| **Productivity** | Finish and merge AI-assisted work instead of leaving it in branches. Use smaller models for small tasks. Look at the costliest PRs in [Engineering & PRs](https://help.nops.io/ai-roi/engineering-overview.md#engineering-and-prs). |
| **Efficiency** | Review AI-written code before merging. Give the AI tools good repository instructions (an `AGENTS.md` or `CLAUDE.md`) and tests to run, so fewer changes need a fix later. |

On the [Developers](https://help.nops.io/ai-roi/developers.md#why-this-score) tab, **Why this score** shows exactly which figures moved a person's score and by how many points, with tips chosen for their weakest figure.

## Privacy

No score reads prompt text, AI responses or code. The figures come from your code host's PR metadata, your AI tools' billing and, if installed, the counts the [AI ROI Agent](https://help.nops.io/ai-roi/agent/overview.md) sends.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.