# AI features

URL: https://docs.ablyft.com/guides/experiments/ai-features/

> ABlyft Lumen generates experiment ideas, interprets results and answers questions about your experiment.



ABlyft Lumen is the AI assistant in ABlyft. It helps you to find ideas for new experiments and to understand the
results of experiments you already ran. AI features are optional and have to be enabled for your team first.

## Enable AI for your team

AI features are off until the **owner** of the team decides about them.

1. Open the team settings (**Team Settings** in the navigation).
2. In the section **ABlyft Lumen (AI Settings)**, click **Enable AI features**. If you do not want to use AI, click
   **Disable AI permanently** instead; ABlyft will not ask again.
3. Read the information about data processing, tick **I confirm and enable AI features for my organization** and click
   **Enable AI features**.

The field **AI status** then shows **Enabled**, including who enabled it and when. Other roles see the note "Only the
owner of the team can change this setting."

> **Data processing:** Only limited input data is sent to the AI provider, no personal data is intended to be processed, and requests are not
> used to train AI models. The confirmation dialog names the sub-processor and explains how it relates to your data
> processing agreement. If you need a formal amendment, contact ABlyft support.

Without enabling AI, the AI buttons show **ABlyft Lumen not enabled** with a link **Go to team settings**.

## AI credits

AI features use *credits*. Your plan includes a number of credits per month. Each use of a feature costs credits:

| Feature                                                                            | Credits |
| ---------------------------------------------------------------------------------- | ------- |
| Generate new experiment ideas (one run creates up to 5 ideas)                      | 100     |
| Interpret results (per message you send in the chat)                               | 50      |
| Generate variation code with AI (through [MCP](https://docs.ablyft.com/developers/reference/mcp-server/)) | 50      |

The credits reset at the start of each month. If your credits are not sufficient, ABlyft shows **Insufficient AI
credits** and the feature is disabled until the next month or until you get in touch with ABlyft support.

To monitor your usage, go to **Team Settings → Plan Settings → AI Usage**. The page **AI Usage** shows:

* **AI Credits**: used credits compared with the monthly allowance (turns yellow at 70 % and red at 90 %),
* **Available**: credits remaining,
* **Period Reset**: when the credits reset,
* the table **AI Credit Usage** with every transaction of the current period: date, type (**Usage**, **Refund**,
  **Adjustment** or **Purchase**), amount, description and user. You can filter it by type and user.

## Generate experiment ideas

Open **Experiment Ideas** in your project and click **Generate new ideas**. The form, the fields and the result are
described in [Generate new ideas](https://docs.ablyft.com/guides/experiments/experiment-ideas/#generate-new-ideas). Ideas you rejected earlier
with a reason and ideas that already exist are taken into account, so you do not receive duplicates.

> **Usage limit reached:** Your plan limits how many ideas your team can hold. When the limit is reached, ABlyft shows **Usage limit reached**.
> Contact ABlyft support to upgrade your plan.

## Interpret results with AI

1. Open the results of an experiment, see [Interpret results](https://docs.ablyft.com/guides/results/interpret-results/).
2. Click **Interpret**.
3. A chat opens: **Lumen Result Interpreter**. Pick one of the suggestions (for example "Which variation is the best?",
   "Is this statistically significant?" or "Any differences by device?") or ask your own question.

The assistant reads the current results of the experiment. You can ask it to narrow the analysis to a date range or to
device types, or to compare the statistical methods (Frequentist, Bayesian, Legacy). Answers are short and end with a
recommendation, for example to ship a variation, to keep running or to collect more data.

> **AI is not a substitute for judgement:** AI answers can be wrong. Always check important decisions against the numbers on the results page and wait for enough
> visitors, see [Interpret results](https://docs.ablyft.com/guides/results/interpret-results/).

## AI and MCP

If you connect an AI client such as Claude to ABlyft through the MCP server, it can create experiments and variations,
read results and use the same AI functions. These calls use credits as well. See the
[MCP server](https://docs.ablyft.com/developers/reference/mcp-server/).

## Next steps

- [Experiment ideas](https://docs.ablyft.com/guides/experiments/experiment-ideas/): Collect and prioritize test ideas.
- [Interpret results](https://docs.ablyft.com/guides/results/interpret-results/): Read the results of your experiment.
- [MCP server](https://docs.ablyft.com/developers/reference/mcp-server/): Connect AI clients to ABlyft.