# Exclusion groups

URL: https://docs.ablyft.com/guides/experiments/exclusion-groups/

> Make sure a visitor takes part in only one of several experiments that could influence each other.



When two experiments run at the same time, a visitor can be part of both. Often this is fine. But if the experiments
change the same area of a page, or test related ideas, they can distort each other's results.

An **exclusion group** solves this. You put the experiments that must not overlap into one group. ABlyft then makes sure
that each visitor takes part in **at most one** of them.

## How it works

For a visitor who qualifies for an experiment in a group, ABlyft decides as follows:

1. If the visitor already takes part in one of the experiments of the group, that experiment stays their experiment. They
   are not put into any other experiment of the group.
2. Otherwise, ABlyft picks one experiment of the group by chance. The **traffic share** you set per experiment is the
   weight: an experiment with 70 % is chosen more often than one with 30 %.
3. The chosen experiment continues with its normal checks, including its own
   [traffic allocation](https://docs.ablyft.com/guides/audiences-and-delivery/traffic-allocation/). The visitor is excluded from the other
   experiments of the group.

Things to know:

* Only experiments that are **running** or in **preview** count. Drafts and paused experiments in a group are ignored.
* The draw only considers experiments whose [audiences](https://docs.ablyft.com/guides/audiences-and-delivery/audiences/) match the visitor.
  A visitor who does not fit the audience of an experiment is not "used up" by it.
* The traffic share is a probability, not a guarantee. With few visitors, the real split can differ from the set share.
* In preview mode, exclusion groups are not applied, so you can always check your experiment.
* An experiment can be part of several groups. It then has to win in every one of them.

## Create an exclusion group

1. In your project, go to **Experiments → Exclusion Groups** in the navigation.
2. Create a new group (if the list is still empty, click **Create your first exclusion group**).
3. Enter a **Name** and optionally a **Description**.
4. In the section **Mutually Exclusive Experiments**, choose at least two experiments and set the **Traffic** for each.
   Click **Add another experiment** if you need more. The status of each experiment is shown next to its name.
5. Make sure the **Traffic Sum** is exactly **100 %**. The bar below turns green when it is correct.
6. Save.

By default, a new entry has a traffic share of 50 %. For two experiments that you want to treat equally, keep 50 / 50. To
give one experiment more of the visitors, for example 70 / 30, change the numbers.

| Field           | Rule                                                                |
| --------------- | ------------------------------------------------------------------- |
| **Name**        | Must be unique in the project.                                      |
| **Experiments** | At least two. Each experiment can be added once per group.          |
| **Traffic**     | A whole number from 0 to 100 per experiment. The total must be 100. |

If you add or remove experiments, adjust the traffic values so that they add up to 100 %. Changes to a group are
published automatically.

## Find out which experiments are in a group

* On the experiment page, the card **Status** shows a badge such as **Part of 1 exclusion group**.
* In the experiment list, filter with **Part of any exclusion group**, or show the column **Excl. Groups** with the
  column toggle.

## Delete a group

Select one or more groups in the list and use the bulk action to delete them. The experiments are not deleted, they only
leave the group.

## Plan your traffic

Because a group splits your visitors between experiments, every experiment in it receives only a part of your traffic.
Each experiment then needs more time to collect enough data. Use exclusion groups only where overlap would really harm
your results. Use the [pre-test calculator](https://docs.ablyft.com/guides/results/pre-test-calculator/) to estimate the runtime.

## Next steps

- [Traffic allocation](https://docs.ablyft.com/guides/audiences-and-delivery/traffic-allocation/): Distribute visitors across variations.
- [Audiences](https://docs.ablyft.com/guides/audiences-and-delivery/audiences/): Choose who sees an experiment.