Base Item Item Recommender

What it is & the business problem it solves

The User Item Recommender is the classic engine (based on collaborative filtering, see below) behind Personalized Product Recommendation in Inspire. It works out, for each customer, which products they are most likely to be interested in, based on what they and similar shoppers have viewed and bought.

Since the summer of 2026 Inspire's recommender is powered by a transformer model. This page describes the collaborative-filtering engine that sits next to it in the cockpit. The cockpit is Inspire's front-end tooling, where you adjust the business rules of your predictions.

New customers start with the transformer-powered recommender by default.

It is a base (core) module. It does not send anything to your channels on its own. Personalized Product Recommendation reads its lists, applies your business rules (exclusions, boosts, fallback) and delivers the final result. Event Based Retargeting and Contextual Product Recommendation do not use this engine. They use the Base Item Item Recommender.

This engine has its own page because it has its own settings and its own schedule, even though customers meet its output only through Personalized Product Recommendation.

How it works

The engine uses collaborative filtering: "customers who engaged with product X also engaged with product Y".

  1. Score interactions. Every interaction counts with a weight. By default a purchase counts 1 and a product view counts 0.25. Add-to-cart and remove-from-cart events are ignored unless someone with expert rights adds a weight for them.

  2. Optionally remove outliers. Customers with extremely high volumes, such as B2B accounts or bots, can be left out so that they do not distort the patterns.

  3. Build a product-to-product table. From all customers, Inspire calculates how strongly each pair of products is linked.

  4. Build personal lists. For each customer, Inspire takes the products they engaged with, looks up the most strongly linked products and adds up the scores. The result is a personal list of up to 100 products, before any business rules.

  5. Keep it efficient. Customers with no activity in the last 31 days are skipped. A customer's list is refreshed at most every 7 days. Stored lists are kept for up to 31 days, and delisted products are removed.

There are two scoring methods:

  • Count adds up weighted interactions.

  • Seasonality also weighs recent interactions more heavily (a time decay factor) and can take seasonal patterns into account (a seasonality factor). Time decay is part of Seasonality, not a separate method.

Example use cases

The engine is always used through Personalized Product Recommendation:

Use case

Scenario

Channel

Required products

Personal newsletter block

A fashion retailer shows 4 "picked for you" products (illustrative) in its weekly newsletter

Email

CDP or CXP, Webtracking, Product Feed, MMC Email Campaigns + Personalized Product Recommendation

Homepage "For you"

A DIY retailer shows a personal carousel to logged-in or recognised visitors

Website

CDP or CXP, Webtracking, Product Feed, Web Personalization (Recommendations API) + Personalized Product Recommendation

B2C and B2B separately

A wholesaler-retailer calculates recommendations separately per customer segment (for example consumer vs. trade)

Email / website

As above, plus a segment attribute on the customer profile

What you can configure

You can configure the settings below yourself in the cockpit. Settings marked Expert need expert rights. CM staff and customer admins can change expert settings; to get expert rights, ask your CM contact person.

Setting (cockpit name)

What it does

Default

When to change

Who can change

Schedule

When the engine runs (a cron schedule, which is a standard notation for "run at these times", in Amsterdam time)

Not scheduled: the module stays idle until set

Always set at go-live. Run it before Personalized Product Recommendation (for example nightly, illustrative).

Expert

Number of Recommendations

How many products are stored per customer before business rules

100

Raise it if strict business rules leave too few products

Standard

Recommendation Expiration Days

How long a stored list is kept

31 days

Rarely

Standard

User-Item Recommender User Scoper (Enable Scoping of Users, Inactive Days, Outdated Recommendation Days)

Skips inactive customers and lists that are still fresh

On; 31 days inactive; 7 days outdated

Lengthen Inactive Days if your customers buy less often than monthly (for example furniture)

Standard

Product variants (Group product variants (override), Product group column, Which item to recommend (override))

Treats sizes and colours of one product as one product when learning, then recommends either the main product or all variants

Follows the profile-level product variants setting; recommend all variants

When variants crowd out other products, or when events and feed use different ID levels

Expert

Run Per User Segment (Toggle Run Per Segment Option, User Segmentation Column)

Calculates recommendations separately per customer segment

Off

For clearly separate audiences (for example B2B vs. B2C, or countries)

Standard (toggle) / Expert (column)

Item-Item Matrix Outlier Exclusion

Leaves out very high-volume customers (for example B2B accounts or bots)

Off

When a few customers dominate purchases

Standard

Engagement scorers (Engagement Scorer Method, Scorer Method, Identifier, Engagement Type Weights)

Count or Seasonality; per customer or per session; weights per interaction type

Count; per customer; purchase 1, view 0.25

Add add-to-cart, or switch to Seasonality for seasonal ranges

Standard (weights: Expert)

Item Filter (Inclusion Criteria)

Limits which products may be recommended

Off

To keep only sellable product groups

Standard

Advanced/expert settings (Engagement Scopers, Similarity Method, Pre-Filter Item Item Matrix, Boost options, Recommendation Filter, System Resources)

Fine-tuning of the data window, similarity calculation, boosts, filters and computing resources

Off / tuned defaults

Only with CM guidance

Standard or Expert

Prerequisites & required CM products

Requirement

Why

CDP (CM's customer data platform) or CXP (CM's customer experience platform)

The platform that holds your customer data. Lists are delivered to either platform (through Personalized Product Recommendation).

Standard event types on CDP: the Retail & Ecommerce adapter

Sets up the standard event types (Purchase, Add to cart and so on). It is included in onboarding: there is nothing extra to buy or configure. What matters is that events use these standard event types; custom event types are ignored.

Standard event types on CXP: Templates

The Retail & Ecommerce adapter does not apply to CXP. On CXP you use Templates to set up the equivalent events.

Webtracking (CM's website tracker)

Product detail views

Purchase lines (order lines from the webshop or back office)

Purchases carry the highest weight

Product Feed

Product catalogue. Product IDs in events must match the feed's item ID.

Data module

Loads events and the product feed

Personalized Product Recommendation

Needed to apply business rules and deliver the lists

This module works with both CDP and CXP data.

Packaging tier

Available in the MMC Enterprise package only (not in Basic, Advanced or Pro). Inspire as a whole is available from the Pro package upward; this module is one of the Enterprise-level modules.

Data requirements

  • Events: product detail views and purchase lines (used by default). Add-to-cart and remove-from-cart are only used if someone with expert rights adds a weight for them.

  • Purchases matter. If no purchase events reach Inspire, the purchase weight matches nothing and recommendations are based on views only.

  • Cold start: customers with no interactions get no personal list. Personalized Product Recommendation can give them a fallback list instead (see that page). New products only appear once customers have engaged with them together with other products.

  • Technical minimum requirements. The engine has no coded minimum number of customers, orders or history: it runs on any amount of data, but quality grows with the number of customers and interactions. The technical conditions that do apply by default: a customer only gets a list when they were active in the last 31 days (Inactive Days), and a stored list is only refreshed when it is older than 7 days (Outdated Recommendation Days).

How to check it works

  • Event viewer: open the User Item Recommendation events in the cockpit to see a sample of customers and their raw lists, with product name, category and image.

  • Statistics: the cockpit shows how the recommended products are distributed over product attributes such as category or brand. No other module-specific statistics are predefined for this engine. Statistics are set up per module instance.

  • Run history: check that the module runs on its schedule, then check Personalized Product Recommendation for the delivered result.

Limitations & common pitfalls

  • Idle after setup. The schedule starts as "not scheduled", so nothing is calculated until it is set.

  • Cart events are not used by default. Only purchases and views count unless weights are added.

  • ID mismatch. If events use variant IDs and the feed uses parent IDs (or the other way round), lists stay empty or thin.

  • Variant crowding. Without variant grouping, one product's sizes or colours can fill several slots.

  • Purchase cycle vs. 31-day window. For products bought every few months, many customers count as inactive and keep older lists. Adjust Inactive Days.

  • B2B or bot traffic can dominate the patterns. Use outlier exclusion.

  • Lists are kept up to 31 days. A customer's list does not change daily; it is refreshed at most every 7 days.