Recommendation module settings by industry

What is this article about?

Mail recommendation module weights (see Module adjustment and tuning) don't have one universal "correct" setting. The right starting point depends heavily on your business type: how often customers repurchase, how considered the purchase is, and how big your catalog is. This article gives you six starting-point profiles by industry, so you have a validated baseline to configure your e-mail modules from, and then A/B test against your own data.

Prerequisites

An e-mail recommendation module already set up and returning results. See Introduction to Mail Recommendation Strategies.
Familiarity with what each weight does. See Module adjustment and tuning.

The six archetypes at a glance

Parameter	Durable fashion	Beauty / consumables	Food / grocery	High-tech	Travel	Marketplace
History size	50	50	50	30-50	10-20	40-50
Cross-sell weight	+10	+8	+10	+12	+6	+8
Look-alike weight	+7	+5	+3	+4	+12	+8
Buy history weight	-8	+8	+12	-6	-6	-4 to 0
Basket weight	-10	-10	-10	-8	-5 to 0	-10
Visit / search weight	+3	+2	0-2	+6	+6	+4
Popularity / fallback	Low	Medium	Low	Medium	High	Low + shuffle
Trending period	30 days	14-30 days	7-14 days	-	7-14 days + 360 days	7-14 days

🔍 Note: one pattern holds across every archetype: basket weight stays negative everywhere, so items already in the basket are never re-suggested. Buy history weight is the only parameter that flips sign depending on the business, negative for durable purchases and travel, positive for consumables and groceries that get rebought regularly.

1. Durable fashion / ready-to-wear

Parameter	Value	Why
History size	50	Clothing taste is stable, so the underlying preference matters more than the last isolated session.
Cross-sell weight	+10	Completing the outfit (top to bottom, accessory) lifts the basket value, the main lever for average order value.
Look-alike weight	+7	Suggesting similar items in the same style keeps discovery going without drifting from the subscriber's taste.
Buy history weight	-8	Customers don't re-buy the same garment, so it needs active suppression, though not excluded entirely.
Basket weight	-10	Never re-suggest what's already in the basket.
Visit history weight	+3	A recent visit signals useful intent, but weaker than a purchase.
Popularity / fallback	Low	Individual signals are rich, so there's little need to lean on a popularity fallback.
Trending period	30 days	A low-volatility audience, so a long window avoids noise.

2. Beauty / cosmetics / consumables

(fragrance, skincare, hygiene)

Parameter	Value	Why
History size	50	Very regular habits; a long history captures the repurchase ritual.
Cross-sell weight	+8	An associated routine (cleanser, cream, serum) has strong complementarity.
Look-alike weight	+5	Useful to prompt trying a variant, but secondary to replenishment.
Buy history weight	+8	Key inversion versus fashion: the product gets rebought, so re-promote it at the right moment.
Basket weight	-10	Always negative, even for consumables.
Visit history weight	+2	A supporting signal alongside the regularity of purchases.
Popularity / fallback	Medium	A good base for newcomers, but individual data takes priority once available.
Trending period	14-30 days	Matches the pace of launches and beauty trends.

3. Grocery / food / FMCG

(recurring shopping)

Parameter	Value	Why
History size	50	The "typical basket" shows up in the long history, that's where the value is.
Cross-sell weight	+10	Recipe or occasion pairing (pasta, sauce, parmesan) is very strong complementarity.
Look-alike weight	+3	Limited use, the goal is replenishment, not alternatives.
Buy history weight	+12	The most positive of all six archetypes: almost everything here is schedulable replenishment.
Basket weight	-10	Basket anti-redundancy.
Visit / search weight	0-2	Browsing matters little against repeat purchases.
Popularity / fallback	Low	Massive, reliable purchase data makes a popularity fallback less necessary.
Trending period	7-14 days	Short seasonality, weekly products and promotions.

4. Electronics / durable high-tech

(big-ticket, considered purchase)

Parameter	Value	Why
History size	30-50	Few purchases but each is highly meaningful, so keep the entire history.
Cross-sell weight	+12	The accessory (case, warranty, cable) is the biggest post-purchase margin lever.
Look-alike weight	+4	Low, offering a second phone right after a purchase makes little sense.
Buy history weight	-6	Durable goods: customers don't buy another one right away.
Basket weight	-8	Anti-redundancy.
Visit history weight	+6	A long research phase means a visit is a very strong intent signal.
Search history weight	+6	A product search expresses explicit intent ahead of purchase.
Popularity / fallback	Medium	Useful for discovery, but the decision stays individual.

5. Travel / trips

Parameter	Value	Why
History size	10-20	Very few events per person; lean on recent activity rather than a history that mostly doesn't exist.
Cross-sell weight	+6	Becomes more of an upsell (upgraded board basis, extra stars), with limited true complementarity.
Look-alike weight	+12	The core of the engine here: whoever liked the Canary Islands will like the Balearics.
Buy history weight	-6	Customers don't rebook a trip that's already departed.
Basket weight	-5 to 0	Persistent baskets are rare in a travel booking journey.
Visit / search weight	+6	Without a purchase, the destination viewed or searched is the best available signal.
Popularity / fallback	High	Around 92% of customers book only once, so the wisdom of the crowd carries most of the recommendation.
Trending period	7-14 days + 360 days	A short-term window for offer trends (private sales) plus a long window as a stable base.

6. Marketplace / large catalog

(10.000+ SKUs)

Parameter	Value	Why
History size	40-50	Lots of signal available; a long history helps refine within a huge catalog.
Cross-sell weight	+8	Real complementarity exists, but it's variable across such a broad range of categories.
Look-alike weight	+8	Essential for navigating a very broad offering.
Buy history weight	-4 to 0	Depends on the durable/consumable mix, neutral by default, then segmented by category.
Basket weight	-10	Anti-redundancy.
Visit history weight	+4	A good browsing signal in a deep catalog.
Popularity / fallback	Low + shuffle	Otherwise best-sellers crush the long tail; shuffle reintroduces diversity.
Trending period	7-14 days	A volatile catalog with fast turnover.

🔍 Note: these values are validated starting points to configure from and A/B test, not fixed rules. Your own catalog structure, history depth, and customer behaviour will shift the ideal setting over time.