What it is & the business problem it solves
Contextual Product Recommendation (CPR) is a module of Inspire that answers one question for every product in your catalogue: "Which other products should we show next to this one?"
The recommendations depend on the context (the product a visitor is looking at, or has in the cart), not on who the visitor is. That makes CPR suitable for anonymous visitors and for places where you only know the product, such as a product detail page (PDP) or a cart page.
CPR produces two kinds of lists:
-
Substitutes — alternatives to the product ("You might also like"): help visitors who are not yet convinced find a better match instead of leaving.
-
Complements — products that go with it ("Complete the look", "Frequently bought together"): raise the average order value through cross-sell.
One module instance produces one type. If you want both a substitute carousel and a complement carousel, create two module instances of CPR (a module instance is one configured copy of a module in the cockpit; CM sometimes calls it a "BMI").
How it works
-
Learn which products belong together. CPR does not read customer behaviour itself. It builds on the Base Item Item Recommender, which analyses product views and purchases across all customers and finds products that are viewed or bought together (substitutes) and, if switched on, products from related categories that are bought together (complements).
-
Keep the chosen type. CPR keeps either the substitute or the complement pairs, depending on Recommendation Type.
-
Apply your business rules. It removes excluded products, applies boosts (move products up the list), diversity limits (for example at most 2 products per category) and fixed positions.
-
Top up thin lists with a fallback. When a product has too few learned recommendations, CPR fills the list with a fallback (a backup list, for example the most popular products overall or the most popular products of the same category or brand).
-
Keep only complete lists. Products that still do not reach the required number of recommendations are left out of the output (see Limitations).
-
Deliver. If delivery is switched on with Batch, CPR writes the result into the Product Feed (CM's product catalogue): for each product, one attribute per CPR module instance containing a ranked list of recommended product IDs. Web Personalization reads that list at page-load time, so the carousel on the PDP shows the list for the product being viewed.
The lists themselves are calculated in batch on the module's schedule (a schedule is a cron expression, a standard notation for "run at these times", in Amsterdam time). Page loads only look up the stored list.
Example use cases
-
"You might also like" on the product page
-
Scenario: a fashion retailer shows four alternatives on every PDP so visitors who dislike the colour or fit find something similar.
-
Channel: website (Web Personalization carousel).
-
Required products: CDP or CXP, Webtracking, Product Feed, Web Personalization; Inspire with Base Item Item Recommender and CPR (type Substitute, delivery Batch).
-
Cross-sell in the cart
-
Scenario: a DIY retailer shows matching accessories (screws, drill bits) for the drill in the cart.
-
Channel: website cart page or mini-cart.
-
Required products: as above, CPR type Complement; complements switched on in the Base Item Item Recommender.
-
"More from this brand" carousel
-
Scenario: a beauty webshop shows other products of the same brand on the PDP.
-
Channel: website.
-
Required products: as use case 1, with fallback method Popularity Same Attribute or Catalog Same Attribute on the brand attribute, optionally combined with a business rule that only keeps the same brand.
What you can configure
You can configure the settings below yourself in the cockpit (Inspire's front-end tooling for adjusting the business rules of your predictions). 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 |
|---|---|---|---|---|
|
Recommender Schedule |
When CPR recalculates its lists (cron, Amsterdam time). |
Not scheduled (never runs automatically) |
Always set it; schedule it after the Base Item Item Recommender has finished. |
Standard |
|
Recommendation Type |
|
Substitute |
Choose Complement for cross-sell; create a second instance to offer both. |
Standard |
|
Number of recommendations |
How many products each list contains (1–50). |
1 |
Set to the number of slots in your carousel (for example 4–8). |
Standard |
|
Minimum number of recommendations |
Products with fewer recommendations than this are dropped from the output (0–50; 0 = same as Number of recommendations). |
0 |
Lower it so products in small categories still get a (shorter) list. |
Expert |
|
Item Exclusion Criteria: Enable / Exclusion Criteria |
Removes products matching the criteria from the recommended products (for example out-of-stock items, gift cards). |
Off |
To keep certain products out of carousels. Does not stop a product from receiving recommendations. |
Standard |
|
Enable Numerical Attribute Boost / Numerical Attribute Boost Configuration |
Moves products up based on a numeric attribute (for example margin), strength 0–10. |
Off |
To favour high-margin or high-rated products. |
Standard |
|
Enable Categorical Attribute Boost / Categorical Attribute Boost Configuration |
Moves products matching a condition up (for example own brand), factor 0–10. |
Off |
To promote a brand, collection or new arrivals. |
Standard |
|
Fix position of recommendation with certain characteristic: Enable / Position Fixing Rules |
Pins a product matching a condition to a position (1–15). Rules apply in order; each rule affects one product. |
Off |
To always show, for example, one sale item in position 1. |
Standard |
|
Maximum recommendations per category: Enable / Product Diversity Rules |
At most N (1–15) products with the same value of an attribute (for example category). |
Off |
To avoid carousels full of near-identical products. |
Standard |
|
Maximum recommendations per characteristic: Enable / Product Feature Limit Rules |
At most N (1–15) products matching a condition. |
Off |
To limit, for example, sale items to 2 per list. |
Standard |
|
Fallback Method |
How thin lists are topped up: Popularity, Popularity Attribute, Popularity Same Attribute, Catalog Same Attribute (see below). |
Popularity |
Prefer a "same attribute" method on category or brand so fallback products stay relevant. |
Standard |
|
Number of fallback recommendations |
Size of the fallback pool per product (0–5000). |
100 |
Increase when diversity or position rules need more candidates. |
Expert |
|
Item Attributes / Popularity Columns |
The attribute(s) the "attribute" fallback methods use (for example category, brand) and, for Catalog Same Attribute, how to order the products. |
None |
Required when you choose an attribute-based fallback method. |
Expert |
|
Enable item fallback filter |
Switches on a filter that limits which base products (the products a list is made for) get fallback recommendations. |
Off |
To stop certain products (for example gift cards) getting a list. |
Standard |
|
Filter expression on fallback base items |
The condition for the filter above. |
None |
See above. |
Expert |
|
Relational rules (Enable Numerical/Categorical Relational Boost, Relational Exclusion Rules, Relational Position Rules) |
Rules that compare the recommended product with the base product (for example "only same brand", "cheaper than the base product first"). |
Off |
For precise merchandising rules; see Limitations. |
Expert |
|
Delivery Method |
|
None — delivery is opt-in |
Set to Batch for Web Personalization. |
Expert |
|
Advanced/expert settings (Memory, CPU) |
Computing resources. |
4 GB memory; 0.5 CPU |
Large catalogues. |
Expert |
Fallback methods in plain words
-
Popularity — one list of the most popular products overall, the same for every product. Simple, but can show unrelated products (a tent next to a dress).
-
Popularity Attribute — popular products grouped by the chosen attribute of the base product. Note: this does not guarantee the same brand or category in every slot.
-
Popularity Same Attribute — popular products that share the attribute value with the base product (for example same category). Good default for relevant fallbacks.
-
Catalog Same Attribute — products from the catalogue with the same attribute value, even if they have no sales history yet, ordered by the Popularity Columns you choose. Use this for strict "same brand" carousels and for new products.
Prerequisites & required CM products
|
Requirement |
Why |
|---|---|
|
CDP (CM's customer data platform) or CXP (CM's customer experience platform) |
Inspire reads your customer events from the platform. CPR itself writes its lists to the Product Feed, not to the platform. |
|
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 views and add-to-cart events feed the Base Item Item Recommender. |
|
Purchase data (order lines) |
Purchases are the strongest signal, especially for complements. |
|
Product Feed |
Product IDs, names, images, category and brand; also the destination of the CPR lists. Product IDs in events must be identical to the Product Feed IDs. |
|
Base Item Item Recommender |
Required module; CPR builds on its output. For complements, its Enable Complement Recommendations setting (expert) must be switched on. |
|
Web Personalization |
To show the lists on your website. |
|
Cockpit |
To configure and monitor the module. |
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. The Base Item Item Recommender it depends on is also an Enterprise module.
Data requirements
-
Events: product detail views and purchases (via the Base Item Item Recommender); add-to-cart helps.
-
Product attributes: ID, name, image, category; brand or other attributes if you use them in fallbacks, boosts or diversity rules. Numeric attributes (price, margin) must be numbers for numeric boosts.
-
Volume: products need enough shared views or purchases to get learned recommendations. Products with little history rely on the fallback.
-
Cold start (new products without history): they get no learned recommendations; only Catalog Same Attribute can give them a relevant list straight away.
-
Technical minimum requirements. There is no coded minimum number of products, events or history. The technical conditions that do apply: a product is only published when its list reaches the required length (Number of recommendations, or Minimum number of recommendations if set), and the run stops with an error when more than 50% of the products would be dropped for that reason. That 50% limit is fixed and cannot be changed in the cockpit.
How to check it works
-
Run status: in the cockpit, check that the module instance has a schedule and recent successful runs, after a successful run of the Base Item Item Recommender.
-
Event viewer: open the cockpit event viewer "Contextual Product Recommendation" to see, for a sample of base products, the recommended products with name, category and image, and whether they come from the fallback.
-
Statistics: the module's statistics are Fallback Rate, Fallback Distribution and Fully Fallback Distribution (how many lists consist partly or fully of fallback products), plus distributions of the recommended products over product attributes such as category or brand. Statistics are set up per module instance. A high fallback share means low demand or a tracking gap.
-
Website: open a PDP with the carousel. New lists are not visible instantly after a run: a new run is only served after it has aged a little, and responses are briefly cached. Allow some time before concluding that something is wrong.
Limitations & common pitfalls
-
No CDP or CXP delivery. CPR does not write anything to customer profiles in the CDP or CXP, so it cannot be used directly in email merge tags or segments. For person-based recommendations use Personalized Product Recommendation; for recommendations after a behaviour (abandoned cart, purchase) use Event Based Retargeting.
-
Delivery is off by default. With Delivery Method = None the module calculates but publishes nothing. Use Batch for Web Personalization.
-
Products with too few candidates are dropped and keep their old list. A product that does not reach the required number of recommendations is left out of the new output; its previous list is not cleared and may show outdated (even sold-out) products. Set Minimum number of recommendations (expert) or use a same-attribute fallback so small categories still get a list. If more than half of the products would be dropped, the run fails with an error instead.
-
Item exclusion works on recommended products only. To stop a product from getting a carousel, use the item fallback filter on base products, not the exclusion criteria.
-
Complements need the upstream setting. If complements are off in the Base Item Item Recommender, there are no complement pairs, and cross-category pairs may be labelled as substitutes.
-
Complement lists can be dominated by bestsellers. Store-wide top sellers tend to appear as complements for many products; use diversity rules or exclusions.
-
"Same brand" with Popularity Attribute is not strict. Use Popularity Same Attribute or Catalog Same Attribute for same-brand carousels.
-
Relational rules on empty attributes. A relational exclusion that compares an attribute which is empty for some products can remove all their recommendations. Relational position rules only reorder existing candidates; with a small fallback pool they may never match.
-
Schedules are independent. CPR runs on its own schedule; schedule it after the Base Item Item Recommender so it uses fresh pairs.
-
ID mismatches. If product IDs in web events, orders and the Product Feed differ (for example variant ID vs parent ID), lists stay empty or show the wrong products. With product-variant grouping, one large product group can crowd a list.
-
Empty carousel on the website although everything runs: usually the page does not pass the product ID to Web Personalization correctly.
-
Item Filter is gone. Older documentation mentions a separate "Item Filter"; use Exclusion Criteria instead.