Event Based Retargeting

What it is & the business problem it solves

Event Based Retargeting (EBR) is a module of Inspire that reacts to something a customer did — left a cart, browsed products, applied a category filter, bought something — and sends that customer personalised product recommendations that fit the moment: alternatives for products they did not buy, popular products from a category they were looking at, or matching products after a purchase.

Where the Behavioral Trigger module only reminds a customer of the products they abandoned, EBR adds recommended products. One event per customer contains both the trigger (the products or category the customer did something with) and the recommended products, so one email can say "Still thinking about this? You might also like these".

There are two modules in the cockpit (the front-end tool where you adjust the business rules of your predictions):

  • Event Based Retargeting — the standard (batch) version.

  • Real-time Event Based Retargeting — a near-real-time version with the same business rules. Enterprise customers can choose near-real-time instead of batch for this module.

How it works (plain language)

  1. Learn which products belong together. EBR builds on the Base Item Item Recommender, which finds substitutes (alternatives) and complements (products that go with each other) from the behaviour of all customers.

  2. Find trigger moments. For each customer, EBR looks for the chosen Behavior Trigger within the last Trigger Event Lookback Days (default 30). For abandonment triggers, a trigger is cancelled when the customer later removes the product, buys, etc. (the same rules as the Behavioral Trigger module).

  3. Wait and do not repeat. Optionally waits a number of days (or hours) before sending, and skips triggers already sent within Trigger Event History Days (default 90).

  4. Pick recommendations for the trigger products: substitutes or complements depending on the trigger (see table).

  5. Apply your rules — exclusions, boosts, diversity limits — and top up with a fallback (a backup list of popular products) when there are too few.

  6. Deliver exactly Number of recommendations products per customer, together with the trigger, as one event per customer to your customer data platform: the CDP or the CXP, depending on the delivery destination that was set up for your account. From there, workflows, segments and MMC Email Campaigns use it.

When does it run? EBR has no schedule of its own. The batch module runs every time the Base Item Item Recommender finishes, so the schedule you set on the Base Item Item Recommender (a cron expression in Amsterdam time) determines how often EBR sends. If that module has no schedule, EBR never runs. The real-time module instead starts after each incremental data load (a small, frequent load of only the newest events).

How near-real-time works: with the Enterprise package you can choose near-real-time instead of batch: the module then runs after every incremental data load of the Data module (roughly every 15 to 30 minutes, illustrative: depends on the configured schedule) instead of on the schedule of the Base Item Item Recommender. Prerequisite: incremental loading must be switched on in the Data module; your CM contact enables it for you.

Behavior triggers

Behavior Trigger

Fires when the customer…

Recommends

abandoned cart

added products to the cart and did not remove them, start checkout or buy

Substitutes

abandoned browse

viewed products and did not add them to the cart or buy

Substitutes

abandoned checkout

started checkout and did not buy

Substitutes

abandoned category

applied a category filter on a listing page and did not remove it or look at, add or buy a product from that category

The most popular products of that category (see below)

after purchase cross-sell

bought products

Complements

purchase intent

added products to the cart or bought them (no cancel rule)

Substitutes

The "abandoned category" trigger

Use this trigger when the customer showed interest in a category (for example "sofas") by filtering a listing page, but did not look at any product from it afterwards.

  • What fires it: the customer applies a category filter on your website (a "filter applied" event). The category counts as abandoned when the filter has not been removed again and the customer has not since viewed, added to the cart or bought a product that belongs to that category. A purchase in a different category does not cancel it.

  • Only real categories: the filtered value must match a category in your Product Feed. Filter values that are not categories (for example the checkout page) are ignored. Category matching uses the URL-slug form of the category name.

  • What the customer gets: the trigger is the category itself (not a product). The recommendations are the most popular products of that category (popularity is derived from the Base Item Item Recommender), up to Number of recommendations. A product that is top in several abandoned categories appears once. These recommendations are not fallback recommendations.

  • What does not apply to this trigger: the inactivity delay (Trigger Event Type Delay Hours), the waiting days delay, recommendation exclusions, boosts and the fallback are not used for abandoned category. If a category has too few products, up to 20% of the customers may get fewer products than requested without failing the run (those customers are left out; for all other triggers the limit is below 1%).

  • Settings that do apply: Number of recommendations, Trigger Event History Days (the same category is not sent again), User Inclusion, Email Filtering and Limit Similar Recommendations.

  • What you need to set up first: website tracking of the category-filter events, Enable Filter Events and the Filter Attribute Mappings in the Data module (mapping the filter name on your site to the product category column), and Product Feed enabled. Without this the module runs but sends nothing.

Example use cases

  1. Abandoned cart with alternatives

  • Scenario: a fashion retailer emails customers who left a jacket in their cart, showing the jacket plus three similar jackets.

  • Channel: MMC Email Campaigns, started by a workflow in your customer data platform.

  • Required products: CDP or CXP, Webtracking, Product Feed, MMC Email Campaigns; Inspire Base Item Item Recommender (with a schedule) and EBR (abandoned cart).

  1. Post-purchase cross-sell

  • Scenario: a DIY retailer emails customers 7 days (illustrative) after buying a drill, recommending drill bits and safety glasses.

  • Channel: email.

  • Required products: as above, trigger after purchase cross-sell, delay enabled with 7 waiting days; complements switched on in the Base Item Item Recommender.

  1. Browse abandonment

  • Scenario: a pet-food webshop reminds customers of viewed products with alternatives in stock, excluding products that are sold out.

  • Channel: email or a customer segment for advertising.

  • Required products: as use case 1, trigger abandoned browse, recommended item exclusion on out-of-stock.

  1. Abandoned category

  • Scenario: a furniture webshop filters its "sofas" category page; a customer who then leaves without opening a sofa gets an email with the most popular sofas.

  • Channel: email.

  • Required products: as use case 1, plus category-filter tracking, filter events and the filter mapping in the Data module; trigger abandoned category.

What you can configure

You can configure these settings in the cockpit. Settings marked Expert need expert rights, which CM staff and customer admins can hold; request them via your CM contact person.

The settings are the same in both modules unless stated otherwise. Optional rules (exclusions, boosts, diversity limit, user inclusion, email filtering, fallback filter) each have an "Enable" setting; the rule is only applied when it is switched on.

Note: If a section can be switched on or off, the cockpit shows an on/off switch for it. If a switch you expect is missing, ask your CM contact.

Setting (cockpit name)

What it does

Default

When to change

Who can change

Behavior Trigger

The customer behaviour that starts a retargeting message (see table above). One trigger per module instance.

None — must be set

Always set it; create one instance per trigger.

Standard

Number of recommendations

How many recommended products each customer gets (1–50).

3

Match the number of product slots in your email.

Standard

Enable setting a delay before sending personalization

Switches on a waiting period in days.

Off

For post-purchase flows that should not start immediately.

Standard

Waiting days before sending personalization

Days to wait after the trigger before sending (1–50).

1 day (only when the delay is on)

For example 7–14 days after a purchase (illustrative).

Standard

Trigger Event Lookback Days

How far back trigger events are considered (1–365).

30 days

Shorter for fresher, more relevant reminders.

Standard

Trigger Event History Days

How long sent triggers are remembered so the same trigger is not sent again (1–365).

90 days

Shorter if repeat messages are wanted.

Standard

Trigger Event Type Delay Hours

Hours of customer inactivity after the last trigger event before it fires (0–72).

Batch: 0 hours; Real-time: 2 hours

To avoid messaging customers who are still shopping.

Expert

Max Data Age (Hours)

(Real-time only.) The run fails if the newest event is older than this (1–72).

12 hours

Only if data arrives with a known longer delay.

Expert

Enable Event Trigger Item Exclusion / Event Trigger Item Exclusion Criteria

Trigger products matching the criteria are removed (they neither start a message nor appear in it).

Off

For example ignore gift cards or samples.

Standard

Enable Recommendation Item Exclusion / Recommendation Item Exclusion Criteria

Recommended products matching the criteria are removed.

Off

For example remove out-of-stock items.

Standard

Enable User Inclusion / User Inclusion Expression

Only customers matching the expression get messages.

Off

To target one country or segment.

Expert

Enable Email Filtering

Only customers with a valid email address are included.

Off

For email-only use cases.

Standard

Numerical attribute boost (Enable / Numeric Boost Rules)

Moves products up by a numeric attribute (strength 0–10).

Off

To favour margin or rating.

Standard

Enable Item Categorical Attribute Boost / Feature Boost Rules

Moves products matching a condition up (strength 0–10).

Off

To promote a brand or collection.

Standard

Enable Recommendation Boost / Boosting Arguments

Advanced boosting, including comparisons with the trigger product.

Off

Fine-tuning by experts.

Expert

Limit similar recommendations (Enable / Product Diversity Rules)

At most N (1–15) recommended products with the same value of an attribute (for example category).

Off

To diversify; see Limitations.

Standard

Fallback Method

How missing recommendations are filled: popularity (most popular overall) or popularityAttribute (most popular per attribute, for example category).

popularity

Use popularityAttribute for more relevant fallbacks.

Standard

Number of Fallback Recommendations

Size of the fallback pool (1–1000).

200

Increase if exclusions or diversity rules leave too few products.

Standard

Fallback Column Names

The attribute(s) used by popularityAttribute.

None

Required with popularityAttribute.

Standard

Enable item fallback filter / Filter expression on fallback base items

Limits which products get fallback recommendations.

Off

Rarely needed.

Standard (switch) / Expert (expression)

Group product variants (override) / Product group column

Treats variants (sizes, colours) of one product as one.

Taken from the profile

Keep identical to the Base Item Item Recommender.

Expert

Delivery Method / Batch Method / Event Type ID / CXP Definition ID

Whether and where results are delivered: CDP (event type) or CXP (definition). Set up by CM.

Batch, to CDP

When delivering to CXP or testing without sending.

Expert

Advanced/expert settings (Enable User Information, List of User Features, CPU, Memory)

Customer fields in the event (by default email and webtracker client ID); computing resources.

On; 500m; 4Gi

Large customer bases or extra template fields.

Expert

Price-drop triggers are not available in EBR; use the Behavioral Trigger module for price-drop reminders.

Prerequisites & required CM products

Requirement

Why

CDP or CXP

Inspire reads events from your customer data platform and writes one EBR event per customer back to the CDP or the CXP.

Webtracking (CM's website tracker)

Views, add-to-cart, remove-from-cart, begin-checkout (and category-filter events for abandoned category).

Purchase data (order lines)

Triggers for post-purchase flows; cancels abandonment triggers; strongest signal for recommendations.

Standard event types

Events must use the standard event types (Purchase, Add to cart, …); a custom event type is ignored. On CDP these are set up by the Retail & Ecommerce adapter, which is included in onboarding (nothing extra to buy or configure). On CXP the adapter does not apply; the standard event definitions come from Templates.

Product Feed

Product details and attributes for exclusions, boosts and fallback; IDs must match the events. For abandoned category it also needs the category column.

Data module

For abandoned category: filter events switched on and the filter-to-category mapping.

Base Item Item Recommender

Required module and the source of recommendations. It needs a schedule; for after purchase cross-sell its Enable Complement Recommendations setting (expert) must be on.

MMC Email Campaigns (or another channel)

To send the message, usually via a workflow.

Cockpit

To configure and monitor the module.

Packaging tier

Inspire is part of the MMC packages Pro and Enterprise; it is not included in Basic or Advanced. Event Based Retargeting and Real-time Event Based Retargeting are Enterprise modules (Enterprise only). Near-real-time variants are included in Enterprise (no add-on): an Enterprise customer can choose near-real-time instead of batch. See How near-real-time works above.

Data requirements

Technical minimum requirements (what the module needs to work; these are not quality guarantees):

  • Trigger events: at least one event of the trigger type for your chosen Behavior Trigger (for example add-to-cart for abandoned cart) within Trigger Event Lookback Days (default 30). If there is none, the run fails.

  • Cancel events: for abandoned cart, abandoned browse and abandoned checkout, at least one purchase event (abandoned cart: purchase or begin-checkout) within the lookback period. If there is none, the run fails. This does not apply to abandoned category, after purchase cross-sell and purchase intent.

  • Enough recommendations: every customer must end up with exactly Number of recommendations products. Customers who fall short are left out of the delivery, and the run fails when 1% or more of the customers fall short (abandoned category: more than 20%). The fallback list (default 200 popular products) normally prevents this; the Base Item Item Recommender's own data minimums apply as described on its page.

  • New triggers: if all triggers were already sent within Trigger Event History Days, the run completes and sends nothing.

  • Identified customers: messages can only reach customers with a known email address (or another reachable identifier) in the platform.

  • Product attributes: category, brand and any attribute used in boosts, exclusions or fallback.

  • Cold start: customers whose trigger products have no learned recommendations get fallback products (marked as fallback in the event viewer).

How to check it works

  • Run history: check that the Base Item Item Recommender has a schedule and successful runs, and that EBR runs after each of them.

  • Event viewer: open the cockpit event viewer "Event Based Retargeting" to see, for a sample of customers, the trigger products, the recommended products (name, category, image) and whether fallback was used. Available for both modules.

  • Statistics (batch module only): the batch module has a statistics calculator. Statistics are switched on per module instance by applying the module's default statistics template in the cockpit; until then the module has no statistics. The calculator counts customers and shows the distribution of the recommended products over the product attributes of your profile (for example category and brand). The real-time module has no statistics.

  • CDP or CXP: open a test profile in your customer data platform and check the EBR event (or mapped profile properties) with the trigger and recommended products.

Limitations & common pitfalls

  • A run fails if too many customers get too few recommendations. EBR requires every customer to get exactly Number of recommendations products. Customers who fall short are left out; if 1% or more of the customers fall short the whole run fails (for abandoned category the limit is 20%). Tight diversity limits on substitute-based triggers (for example "max 1 per category" when substitutes come from the same category) can therefore make the whole run fail. Loosen the limit, enlarge the fallback pool or lower the number of recommendations.

  • No own schedule. If the Base Item Item Recommender is not scheduled, EBR does not run.

  • Exclusions describe what to remove. Write the criteria for products you want to drop, not keep. Older documentation mentions "inclusion" settings for trigger and recommended products; these no longer exist.

  • Variant grouping must match. If product-variant grouping differs from the Base Item Item Recommender, customers may get a different size or colour of the product they already have as a "recommendation".

  • Volume is limited to identified customers. Only customers with a known email address can be reached; expect far fewer messages than trigger moments.

  • Complements need the upstream setting. After purchase cross-sell needs complements switched on in the Base Item Item Recommender.

  • Abandoned category ignores several rules. Recommendation exclusions (for example "out of stock"), boosts and the fallback are not applied, so check the category's products yourself before using it in an email.

  • Abandoned category needs filter tracking and mapping. Without filter events and the filter-to-category mapping the module runs green but sends nothing.

  • Standard event types and matching IDs. Custom event types or product IDs that differ from the Product Feed mean no triggers.

  • Workflow filters. Workflows that require profile fields (for example language or country) can silently skip customers who are only known from the website.

  • No price drop. Use the Behavioral Trigger module for price-drop reminders.

  • Near-real-time needs incremental loading switched on in the Data module (see How near-real-time works above).