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

# How to set up Seed Signals on Criteo

> This page explains how to send Seed Signals to Criteo and use them as Lookalike sources in your prospecting ad sets.

## Recommended setup

Start with three Seed Signals. Each one becomes the source of its own Lookalike and runs in its own ad set, so you can compare them.

| Seed Signal | Size |
| :- | :- |
| Seed 1 | 1% to 10% |
| Seed 2 | 11% to 20% |
| Seed 3 | 21% to 30% |

Seed 1 holds the users with the highest incremental lift and gives the most selective Lookalike. Seed 2 and Seed 3 give Criteo more users to expand from.

## Choose what the Seed Signals learn from

Every Seed Signal learns from its Treatments. The scope decides which Treatments these are:

| Scope | What the Seed Signal learns from | When to use it |
| :- | :- | :- |
| `campaign` (default) | The complete campaigns, including existing campaigns you select as a learning basis | Most setups. The Seed Signal learns from all ad sets in these campaigns from day one. |
| `treatment` | Only the ad sets in which a Lookalike of the Seed Signal runs | When the Seed Signal should learn only from the ad sets it is used in. Innkeepr then adds each new ad set that uses the Lookalike. |

You set the scope when you create the Seed Signal. How the learning works is explained in [How Audience Signals learn](https://docs.innkeepr.ai/concepts/audience-signals.md#how-audience-signals-learn-from-your-campaigns).

## Set up Seed Signals

**1. Create the Seed Signals in Innkeepr**

For each row in the table, create an Audience Signal with the purpose **Seed**, select your Criteo connection as the destination, and choose the scope. Innkeepr creates the audience in your Criteo account and keeps it updated. Criteo needs 24–48 hours to match the users.

**2. Create the Lookalikes**

In **Criteo** > **Audiences**, create one Lookalike per Seed Signal, using it as the source.

**3. Add the Lookalikes to your prospecting ad sets**

Add each Lookalike as the audience of its own ad set. Keep creatives, bid strategy, and product set the same across these ad sets, so differences in results come from the audience.

**4. Check the Treatments**

In Innkeepr, open each Seed Signal and check its Treatments: the campaign with scope `campaign`, or the ad set with scope `treatment`. Innkeepr adds new ones automatically; add one manually if it is missing.

## Test against your current prospecting audience

To measure what Seed Signals add, run them next to the audience your prospecting runs on today, for example a Criteo In-Market audience or a Lookalike of your own customer list. That ad set serves as the baseline.

| Ad set | Audience |
| :- | :- |
| Baseline | Your current prospecting audience |
| Seed 1 | Lookalike of Seed 1 |
| Seed 2 | Lookalike of Seed 2 |
| Seed 3 | Lookalike of Seed 3 |

* Give every ad set the **same daily budget**, so spend does not shift between them during the test.
* Keep creatives, bid strategy, and product set identical, so the audience is the only difference.
* Leave budgets, bids, and creatives unchanged for the first one to two weeks while the ad sets learn.
* After two to four weeks, compare CPA, ROAS, and new-customer share of each Seed ad set against the baseline. Shift budget step by step, by 20–30% at a time, to the ad sets that beat it.

## Troubleshooting

> [accordion] Audiences stay empty
>
> | Cause | Fix |
> | :- | :- |
> | Sync still in progress | Allow 24–48 hours after creating the Seed Signal. |
> | Wrong Criteo account connected | Check that the Criteo connection in Innkeepr uses the advertiser account of your campaigns. |
> | Few users matched | Send email addresses to Innkeepr through a shop integration such as Shopify or through [Identify](https://docs.innkeepr.ai/connections/spec-identify.md). |

> [accordion] Lookalike ad sets perform below expectations
>
> | Cause | Fix |
> | :- | :- |
> | Not enough data when the Objective model was trained | Check that the Objective in Innkeepr shows as ready and has enough conversion history. |
> | Seed too small | Use Seed 2 or Seed 3 as the source, so Criteo has more users to expand from. |

> [accordion] Performance drops after a sales event
>
> During and after sales events, such as Black Friday, audiences and conversion patterns shift for a while. Innkeepr adjusts its models to these phases. If performance stays lower for more than one to two weeks after the event, contact the Innkeepr team.


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