Autosegments

🌟

This is a Premium tier feature powered by Loomi AI.

Autosegmentation uncovers hidden opportunities by automatically generating high-value segments. Using Loomi AI to find unique combinations of customer properties and metrics, Autosegments remove the effort needed to analyze data and build hyper-targeted segments manually.

By targeting a more valuable micro-segment, you can expect to increase conversions within their campaign, driving first-time or repeat purchases. You can also tailor the content and recommended products within a campaign based on the qualitative insights about a segment provided by Autosegments.

How do Autosegments work

Autosegments are powered by Loomi AI to generate potential segments from unique combinations of customer properties and metrics.

Potential segments are presented using a visual, interactive chart that supports further exploration. Click into each segment for quantitative and qualitative insights, such as a favorite category, average order value, or revenue per visitor. This allows you to pick the best one and create it with a click to use in a personalized campaign.

Segments created via the Autosegments feature can be immediately used like any audience segment within Bloomreach Engagement - in Scenarios, Campaigns, Reporting, and more.

How to create and implement Autosegments

• You must enter at least one customer property and at least 2 metrics.
• You must have existing customer properties and metrics before using them in Autosegments. Learn how to define properties and create metrics.

Create new Autosegments

Go to Analyses > Autosegments and click on + New autosegment.

Choose a predefined use case in the form of a template or create a new Autosegment from scratch. You can also leave feedback if an important template is missing.

If you select one of the existing templates, the following user experience is the same as when selecting a use case from the Use Case Center.

Define your values

Select customer properties and metrics you wish Loomi AI to base the Autosegments on.

You can use a maximum of:

• 20 properties
• 10 metrics
• 5,000 segments (if more segments are generated, a random subset is returned)
• last 6 months of data

📘

Segments are mutually exclusive

Each segment within Autosegments feature operates independently, and does not influence another segment in any way. This allows the segments to consider all combinations of properties and filters.

Save and start your autosegment

Click on Save and then Start to generate Autosegment. You can see the process in the Results tab. It can take up to 24 hours to generate your Autosegment. The metrics are calculated while running the Autosegment and are not recalculated in real-time when working with Autosegment results.

📘

Only segments that meet the defined 'Minimum user count for each segment' are returned as a result.

View and analyze Autosegments

Once the Autosegment is generated, you can see the results in the Results tab.

The result is presented as a table with all the segments. The table could be sorted according to the metrics. The segments could also be filtered based on the metrics. The result is also visualized in the form of a diagram.

Click on any segment, either in the table or the diagram, and open a modal window with the segment details:

• segment name
• segment description: detail about used properties and values, portion from all customers (100% of customers is the number of customers after applying the customer filter from the setup screen)
• metrics for the specific segment compared to overall project performance
• action button: create segment - this opens a new definition of segmentation with one segment based on the clicked one

📘

Reading the segment metrics table

Please note the metrics in the segment table present values for the whole customer base within the particular segment. For example, the average customer lifetime value in the table above (avg_cltv = 48.64) demonstrates the value of the whole customer base within the Segment 177 (averaging the lifetime value of 33,753 customers), and not a value pertaining to each customer individually.

Create segmentations

Choose the segments that you find useful and create a new segmentation. You can then use this segmentation to personalize marketing campaigns or optimize your product offerings.

Use case examples

Try Autosegments in the following use cases:

Optimize email campaigns by removing a low performing segment and tailoring new content

In a given email campaign/flow, a brand might show different kinds of content: product-specific imagery, educational content about how to use the product, flash sales, and more.

After the email flow is over, marketers can find a low performing segment and remove them from the “overall” campaign. They can find insights about this segment and retarget them with new content.

For example, let’s say a cosmetics companies finds that a sub-segment of users really likes the educational content and they’ve never bought before. These users probably need education about how to use the product before deciding to buy! Marketer can use other insight, like their skin tone, to deliver educational content like how to use dark color foundation.

Example Autosegment dimensions

Property: Has bought before? = No
Property: Acquired email from TikTok influencer campaign
Property: skin tone = dark
Metric: CTR on overall email campaign = 2%
Metric: CTR on educational email = 60%

Target high value customers and refresh LTV targeting

A common best practice is to target high LTV customers, with the assumption that this group has a likely propensity to buy and buy again, given the track record of multiple and/or large purchases.

It’s a fine assumption, but targeting the same LTV group over and over again can lead to diminishing results over time. A marketer can do one of two things: find a sub-segment within the overall group of high LTV or use a different metric that indicates propensity for high LTV.

Let’s say a marketer chooses the first option. They might use Autosegments to find a segment with the example dimensions below. This sub-segment is a diamond in the rough: the LTV is not only way higher than the average, but they’re also likely to buy soon. But, they’ve also bought more than 90 days ago, making them eligible to buy items that are in need of replenishment or updating, like for the seasons.

A savvy marketer might target this sub-segment with messaging around returning to the brand and offer a conditional promotion if the customer buys 3 or more items.

Example Autosegment Dimensions

Property: LTV = $4000 (155% higher than average)
Property: Purchase Prediction = 75%
Property: Last Date Bought = +90 days
Property: Average Products Bought = 3
Metric: Add to Cart Rate = 95%
Metric: Bounce Rate = 12%

Deliver product recommendations for fans of a particular brand

Deliver a product recommendation to users via email, on mobile devices, with Nike products that cost more than $40, and are available to USA customers.

Find users with a high affinity for a specific product category, are mostly engaged on email, have high AOV, and are located within the NYC area.

In this case, you can include specific Nike products that are a bit more expensive, but will probably convert at a higher rate. This translates into incrementally higher revenue.

Example Autosegment dimensions

Property: favorite brand = Nike
Property: last category bought = shoes
Property: favorite channel = mobile
Property: location = NYC
Metric: Add to Cart Rate = 55%

Acquire more customers by refining lookalike audiences on social platforms

Acquisition marketers send valuable segments to social platforms like Facebook or TikTok, having them find potential customers who look like the given “seed” set. Oftentimes, these marketers will build these lookalike audiences off of a single dimension, like LTV.

Instead, acquisition marketers can “juice” the AI engines of the ad platforms by giving it a more specific audience to look for. For example, instead of just high LTV, they might send one that has the example dimensions below.

Example Autosegment dimensions:

Property: LTV = $4000
Property: AOV = $200
Metric: Bounce Rate = 2% (400% lower than average)
Metric: Conversion Rate = 40%

Optimize retargeting strategy and boost ad spend efficiency

Acquisition marketers create prospecting ads, and those ads will drive traffic to the website or app. Some of those users will engage, viewing products or even adding things to cart.

Of this group, acquisition marketers can find a sub-segment that is most likely to purchase, bidding relatively more money for this audience. Then, the marketer will retarget this group, bringing them back to the app to convert.

Example Autosegment dimensions:

Property: First touch source = Facebook
Property: In-session Prediction (Likelihood to purchase = 85%)
Property: First time buyer
Property: # of organic revisits = 3
Metric: Viewed Product = 2
Metric: Bounce Rate = 5%

Reengagement optimization

CRO managers care about their high value customers. Without any specific knowledge about this segment it is hard to personalized the message. With Autosegments they can analyze the high value customers that are at risk to get better understanding of specific subsegments.

Example Autosegment Dimensions:

Property: most visited category - shoes
Property: preferred brand - Adidas
Channel preference - email
Metric: CLTV - 156 USD (top 10%)
Metric: time since last purchase > 60 days

Limitations

• Number of properties: 20
• Number of metrics: 10
• Number of segments: 5000 (if more segments are generated, a random subset is returned)
• Number of considered events: 1000M (if more events are used, they are undersampled to 1000M)
• Number of considered customers: 50M (if more customers are considered, they are undersampled to 50M)
• Max history of considered events: 6 months
• Customer properties of type consents are not currently supported

All the segments and metrics are calculated only once from the longer-term data storage and are not updated until the Autosegments are run again. This data storage is populated once every 6 hours, which could lead to small differences between segments calculated from Autosegments and the real-time data storage that is used by standard Segmentation.