Advertiser Lookalike Clean Room - Onboarding and usage tutorial

Dataset preparation

The lookalike data clean room requires an input file: A CSV table with two columns:

1. The identifier (e.g. email, phone number) agreed with the publisher
2. The audience type. You can use this feature to use different seeds and further optimize your activation AI model: this column groups the identifiers for the insights and the AI model analysis, which will treat differently each audience type. If you do not want to subgroup your seed audience, add a constant value per each row

Below an example:

The CSV file needs to have the following characteristics:

• No duplicated rows: the same user cannot appear more than once in the same audience type
• UTF-8 encoded (can be exported from Excel as shown below)

Registration to the platform

If you do not have an account yet and you have been invited to a clean room, you need first to register.

1. If your organization has a license, contact your Decentriq admin to get a Decentriq seat. If you don’t and you have been invited, ask to the organisation that invited you to provide you an ‘external user’ seat
2. Then, you can sign up at platform.decentriq.com setting up your account with an email and password. The email you register with needs to be the same one stated in the Decentriq seat
3. Confirm your email clicking on the link you will receive by email
4. Log in to the Decentriq platform at platform.decentriq.com . At the first log in, before being able to use your account you will need to:
   1. Accept the end user conditions
   2. Create a keychain password. This keychain password is different from the account password as it manages your encrypted datasets and is used to interact directly with the confidential computing technology leveraging its security features
5. Now, you can access the clean rooms. If you have been invited by another user, you will find your clean room in the sidebar under ‘shared with me’

Data upload from the Decentriq UI

Follow this tutorial to upload a dataset to a lookalike data clean room from the advertiser seat and be able to get inisghts and AI-optimized lookalike audiences to target.

1. Log in to the Decentriq platform at platform.decentriq.com - if you do not have yet an account follow the above steps

2. If you created or you have been invited to a lookalike audience, you can open it from the sidebar

3. Under data tab, click on provision data set in the advertiser box. If you expand the box, you can see what the expected schema of the file is.

4. We have here two different options. One is picking the CSV file with ‘import from my computer’ in case you did an extraction from, for instance, your CRM. The second option ‘choose from my stored dataset’ In this other case, you can use previous data sets that you already made available to the platform, or use files you imported, for instance, from a cloud storage or a CRM integration. Click on ‘import from my computer’ to follow this tutorial

5. Browse the file system or drop a file in the box. The platform will check if it is a valid CSV and will parse it

6. Then, you will see our upload helper tool, which will highlight inconsistencies and help you in cleaning the input file.

   

   1. First, the separator is not always a comma. In continental Europe, it is common to have CSV with semicolons. Other times we have tabs and pipe. On the top right you can select the correct option so that the tool is actually able to parse correctly the CSV file

   2. If your file has a header, click on the option on the top right

      

   3. Make sure that the columns have been mapped correctly to the title. The identifiers should be labelled as matchingID, and the audience type as audienceType

      

   4. Then, we can start looking for errors. Invalid rows are highlighted in red. You can hover over the red rows to see what the issue is. As far as there are invalid rows, you will not be able to upload the file

      

      

   5. HASHING FUNCTION: If the agreed identifier is a hashed email, a hashed phone number or another hash, you can use the highlighted function to hash your matchingID column with the function SHA256 with no salt. rows will then turn from red to black being a valid hash

      

   6. AUTOFIX FUNCTION: If we have still some rows that are red, it is sometimes due to some special characters that are not supported, like trailing whitespaces, quotes or e.g. uppercased letters in emails. If you have some cases like this, you can switch on the AUTOFIX: it will correct automatically these issues. Corrected rows will turn from red to blue

   7. FILTERING FUNCTIONS: If you still have emails with with validation issues, you can filter the invalid rows with the filters in the bottom left, ad filter by ‘invalid rows’.

      

   8. DROP FAILED ROWS FUNCTION: Once the upload helper identified the remaining issues, you can either decide to clean the original file and redo the process, or simply drop the invalid rows.Dropping failed rows mean that the uploaded file will not contain the rows with the highlighted issues, and will contain just valid rows

7. Once everything has been cleaned and you can proceed clicking on Encrypt and Provision.

8. In the INSIGHTS tab or the ACTIVATION tab, you might see that the UI is loading. This is normal: the statistics might take several minutes to compute, and the AI model might take up to one hour to prepare the audience.

Common issues during the data upload

1. If there was an issue with the data or the analyses, once you go to the INSIGHTS tab you will see an error message. Here below the common ones:

   1. Validation error: when uploading a dataset for the first time and such an error manifests, it means there might be formatting issues: most likely matching ID formatting or an invalid file has been selected. Please check if the validation report hints whether issues have been found, by navigating to the DATA tab and clicking over your dataset menu under validation report. Please follow Dataset preparation step above to clean it.

      

   2. Maximum execution time: The AI model might have taken too long to converge and provide a result. Please try again by de-provisioning and re-uploading your dataset again from the DATA tab, and refresh the INSIGHTS page.

      

   3. Can't show insights, overlap too small: Not enough users in the overlap (less than 150 users). If this error appears, it is probable that there was an issue with the matching ID. Please check again that the ID is the one agreed with the publisher, and if the hashing was required that it has been applied as expected.

      

note

For any other issues, or if the issues above still persist, please feel free to contact your person of reference or support@decentriq.com.

How to activate

Once the input data is provided and are valid, the AI model will run. It can take up to 60 minutes, then the data clean room will allow you to create optimized audiences based on the expected reach you want to get.

1. Go to the ACTIVATION tab
2. Select the audience type you want to optimize the audience for. If there was only one in your input data, it will be chosen by default
3. With the slider, you can decide how many users to reach. The number is the estimated number of users that can be potentially targeted by the publisher in 30 days. The smaller the audience size, the more precise the campaign. The bigger, the broader the reach will be
4. Once you decided the desired reach, you can click on ‘Generate’ and an audience item will be created. Once you want to activate it, you can click on ‘make available to the publisher
5. Once the audience is available to the publisher, it will be able to export it from the clean room and activate the audience

Please, note that no notification is triggered automatically when you make an audience available to the publisher.

Multiple audiences can be created and made available to the publisher, they are all independent one from the other