Auto-targeting / AI-Audiences

Auto-targeting saves time by identifying optimal customer groups and the items to target them with. This section provides an overview of the process, followed by a deeper analysis of the various settings and adjustments available.

Why use auto-targeting?

You can use Sift Labs auto-targering AI to answer the below questions:

What products should I use in my targeted campaign to a specific customer group?

Sift Labs AI ranks the products based on what products are most likely to trigger a purchase given the selected group of customers.

Which customers should I send a campaign with specific products to?

Sift Labs AI ranks the users based on who are most probable to make a purchase given the products in your campaign. Using this information you can avoid sending email to customers that would consider it spam by only selecting customers that show the highest affinity to the product you want to promote.

Step-by-step guide to create an AI-based auto-targeting group

Step 1: Create auto-targeting

1. To create your first auto-targeting group: go to the left menu and click Auto-targeting. You can either start a new target group by clicking the plus (+) icon in the menu or the green plus (+) icon in the auto-targeting screen.

2. Name your auto-targeting group based on the target audience you want to create. This name will appear in your marketing/external system when you export your audience as the first part of the audience name.

3. Choose Audience based on:

AI-predictions: Choose this option if you want Sift Lab to use its AI to find the best audiences based on different criteria.

Filter: Choose this option if you want to create a basic target group and filter data manually without AI functionality and scoring possibilities.

Step 2: Settings

Click Next to proceed to the settings page - here is where we start to filter and do the different audiences settings.

Schedule dynamic audiences (optional)

You can schedule your target groups to update dynamically daily, monthly, or on specific weekdays and send them to the chosen system. If you are creating a one-time target group for a specific campaign, send-out, or other use case, you can skip this step.

Toggle Schedule On, then choose:

How Often: Choose how frequently your target group should be re-run with new data.

At What Time: Select when you prefer the data to be updated. Ensure this is scheduled after data has been loaded into Sift Lab.

Time Zone: Choose your preferred time zone.

Destination: Select where the data should be sent. Available options depend on your specific setup.

Advanced Settings (optional)

The advanced settings are generally not recommended to change. With ongoing improvements, they have been replaced by more effective internal optimizations and are expected to be phased out in future updates.

Trend Impact Slider

Adjust how much weight trends in product sales should have. A high trend value increases the influence of seasonal sales patterns (e.g. Christmas-themed products).

Block Users

Choose whether to exclude users from advertisements featuring products they have already purchased. For example: Products sold rarely (like a new sofa): Avoid retargeting users who have already bought the product. Consumables (like toothpaste): Include past buyers, as they may repurchase. Obs. the block will only be if a certain product is chosen in the auto-targeting, not if it is used as a part of a larger selection (like a whole category of products).

Targeting Users Without Transactions

Choose whether to target customers without previous transactions. These users will receive recommendations based on general high-scoring products.

User Settings

This section allows you to filter out the audience/customers for the auto-targeting you are creating. If you do not apply a filter, you will target your entire user base. For example, in the case below you have 71,573 total users.

To filter your audience, click on the Add Filter icon .

Filters can be applied based on:

User metadata (e.g., age, gender, country, region).

Interaction metadata (e.g., sales price, discount, store, date for purchase).

Item metadata (e.g., product name, category, color, size, cost).

For example, in the picture below: Here we have filtered out customers who live in Stockholm and have a email adress. Those are the only ones that will be part of our target group.

As you apply filters, the total user count updates dynamically, allowing you to verify filter effectiveness. When you are done with your overall filters continue the next section.

Step 3: Creating an Audience

Click Add Audience to create your audience.

Naming Your Audience This name will appear in your marketing system/external system after export (alongside the auto-targeting name of the whole auto-targeting set in first step). Choose a clear and recognizable name.

Different AI-Predictions

You can choose to:

1. Find the best products for customers (e.g., identifying likely purchases for a specific customer segment).
2. Find the best customers for products (e.g., targeting users who are likely to buy a certain brand or category).
3. Combine both methods for more refined targeting.

1. Finding Products for Customers

Select a customer segmentation of your chooise (e.g., Lapsing customers).

Add Auto Items to let AI choose the most relevant products. Add at least six auto items to allow meaningful predictions based on items, and for you to be able to see good suggestions. BILD Click Save and Run. (continue to ”Evaluating your target group”)

On the results page you will now see different suggestions of products that these people (lapsing customers) are most interested in buying right now. (More on evaluating your results in next section)

2. Finding Customers for Products Select a product to predict against.

You can either choose the product/products you want to predict upon or you can use the filter options to instead filter on a certain group: like a category level, a product, a brand.

To choose a certain product you type in the name of your items and they will popup in the list and you can click on the products to add them.

To choose a certain group of products: Start by Adding auto items so that there is a “spot” for the prediction to take place. Add at least 6 auto-items and then start to filter upon the certain criteria for this prediction. Note: It is not before you add the auto-items the filter on Items will appear.

For example, in the picture below: we have added a filter on Shoes and Brands: Lacoste or Fila. The prediction will be who is most likely to purchase the shoes from the brands Lacoste or Fila. A total of 2 products out of 261 available.

Click Save and Run. (continue to ”Evaluating your target group”)

On the results page you will now see the predictions on the selected items and the score distribution of the users.

3. Combining both methods You can also do a combination of the two examples above: Add a customer criteria and adding filters on certain products/categories/brands.

For example, in the picture below: we have a combination of customers in the Lapsing phase of their Customer Lifecycle, and we want to predict which of those are most likely to buy shoes of brand Lacoste or Fila.

Min and Max number of users (optional) The minimum number of users and maximum number of users gives you the opportunity to limit or force a certain number of users to be part of your audience. This can be used after you evaluate your target group (see next section) or if you want to force a fixed number of users into an audience you are creating. If you for example are doing different audiences where you want at least 10 000 users per audience, you can set 10 000 as minimum in each of them to make sure you have users scored in all audiences.

Step 4: Evaluating Your Target Group

After clicking Save and Run, Sift Lab processes the prediction, which may take between a couple of seconds to up to 5 minutes depending on the amount of data and filters. Your results will then be shown:

Total Users: Displays the number of users in the target group.
View Results: Shows current and past predictions.
Edit Audience Names: Allows direct name adjustments without returning to previous screens.

Evaluating Your Scoring

The score distribution graph on the right side shows the likelihood of customers purchasing specific products. If the graph has a long tail, it indicates many uninterested customers.

For example, in the picture below: Adjusting audience size to 10,320 users will represent 80% of future orders. Then you want to exclude those from being a part of your target audience (they are not likely to buy.

To adjust your target audience: Click Edit on the right upper side of the screen.

Set a maximum number of users for your audience.

Click Save and Run again to update results.

Your target group is ready to be exported.

Step 5: Exporting Segments

Click Export Segments to send your audience data to connected platforms. Exports take 0–30 minutes, depending on system setup. If you have scheduled audiences, the audience is sent automatically.

More information about Exports is found under the Exports section.

More advanced filter opportunities

Filter Settings Disable Filter: Temporarily deactivate a filter group. This can be used to easily see the effect a filter has on the customer base. Exclude: Invert filters (e.g., include users who have not made a purchase in a category).

Match Any: Changes filter logic from AND to OR. By activating Match Any users that fill the criteria of ANY filter is included

In the data model you can assign different roles to a field depending on how it should behave.

In the datamodel you can assign different roles to a field depending on how it should behave.

A category behaves like a string (text) this means you can create a filter that include all fields that contains a certain text.

With a Number you can create filters that are larger or smaller than the field.

With a Timestamp you can use the date format to filter on dates larger than ex. 2022-10-10.

Number options

Equals
Not Equals
Less than
Greater than
Less than or equals
Greater than or equals
Between
Not between

Timestamp options

Between
Not between
Equals
Not equals
Before
After