Cross-sell recommendations
Cross-sell suggestions mined from your own Shopify order history and shown beside the quote your sales team is building — with a merchandising matrix to steer what gets offered with what.
Last updated
The short version
- While a quote is being built, RevLogic shows "Goes with this quote" — products your own customers actually buy alongside what is already in the cart.
- Pairings are mined from your order history with a statistical floor — a rule needs at least three distinct buyers and has to clear a significance test, so a three-sale coincidence never gets promoted.
- Rules rank by the lower bound of their confidence, not raw confidence — a pairing seen five times can't outrank one proven across hundreds of orders.
- The merchandising matrix lets you override the data — pick up to three product types to push whenever a given type is in the cart, and it applies instantly.
The suggestion has to arrive during the call
Cross-sell advice delivered in a monthly report is advice nobody acts on. It has to appear at the one moment it can change the outcome: while a sales rep is on the phone with the buyer, building the order.
That is where RevLogic puts it. On the quote screen, beside the cart, a “Goes with this quote” panel updates as items are added — products your own customers buy alongside what is already there, ready to add with one click at the price the quote is using.
Mined from your orders, not a template
The recommendations come from market-basket analysis of your synced order history. RevLogic looks for products that appear together in the same order far more often than their individual popularity would predict, and stores those pairings as rules for your store.
This is worth being precise about, because the category is full of hand-waving. There is no shared model across merchants, no industry template, and no guessing. If RevLogic suggests an item, it is because your customers have bought it with the anchor item, in your store, in orders you can go and open.
Why naive “frequently bought together” fails on small stores
The obvious implementation — rank by how confident the pairing looks — breaks in exactly the situation most B2B stores are in: a long catalog and a modest number of orders.
Three things go wrong, and RevLogic guards against each:
One enthusiastic customer looks like a trend. A single account that reorders the same two items every month generates twelve co-occurrences that all come from one buyer. RevLogic requires a pairing to be bought by at least three distinct customers before it counts as a rule.
Small numbers produce fake certainty. Two items bought together three times out of three looks like 100% confidence. RevLogic ranks by the lower bound of the confidence interval instead, which is the statistically honest version of “how sure are we, given how little we’ve seen?” A rule proven across two hundred orders beats a perfect-looking three-sale fluke, every time.
Popular items pair with everything. Your best seller co-occurs with half the catalog simply by being in half the orders. A lift test gates this out: the pairing has to beat what you would expect from the two items’ popularity alone.
The effect of all three is a shorter list of suggestions that are worth reading. On the quote screen, a sales rep who gets one absurd recommendation stops looking at the panel — so precision matters far more than coverage.
The merchandising matrix: your rules, on top of the data
Data does not know about your margins, your overstock, your new line, or the supplier deal you signed last week. The Merchandising page is where you tell it.
It is a table of your product types. For each one, you nominate up to three target types to push whenever that type appears in a cart:
| Anchor type | Target 1 | Target 2 | Target 3 |
|---|---|---|---|
| Coffee beans | Filters | Grinders | Mugs |
The first slot carries the strongest boost and is guaranteed a place in the recommendations, so your most important play always shows up. Rows are sorted by sales, so the types that matter to your revenue are at the top of the page.
Two details make it usable in practice:
- It applies instantly. No re-sync, no overnight rebuild. Save the play, and the next quote screen your sales team opens already reflects it.
- It blends rather than overrides. Where a real mined pairing exists for a boosted type, the boost lifts that genuine recommendation. Where none exists yet, RevLogic injects your top seller in that type — so a brand-new strategy has something to offer from day one.
Cross-sell versus gap analysis
These are two different questions and RevLogic answers both, in different places:
- Cross-sell is about the cart, answered during the quote: what goes with what they are ordering right now?
- Product gap analysis is about the account, answered before the call: what whole categories and brands does this customer under-buy compared with similar customers?
A good call uses both — gaps set the agenda, cross-sell fills out the order.
Frequently asked questions
- Where do cross-sell recommendations come from?
- Your own order history. RevLogic mines association rules from the orders it has synced — pairs of products that get bought together more often than chance would explain — and stores them per store. Nothing comes from other merchants' data, an industry template, or a generic catalog. A recommendation is only ever something your customers have actually done.
- How does RevLogic stop rare coincidences from being recommended?
- Three guards. A pairing must be bought together by at least three distinct customers, so one loyal account reordering the same two items monthly counts as one buyer rather than twelve. Its co-occurrence count must clear a statistical significance floor, so chance pairings in a small store are filtered out. And rules are ranked by the lower bound of their confidence rather than confidence itself, which systematically demotes rules with thin evidence.
- How does the recommendation engine actually work?
- It is market-basket analysis — the same family of technique behind "frequently bought together" — trained on your store's own order history. RevLogic mines association rules between products, scores them statistically, and ranks them against the live cart. The method is deliberately explainable rather than a black box: every recommendation traces back to real orders in your store that you can open and check, and nothing is ever suggested that your customers have not actually done.
- Can I control what gets recommended?
- Yes, on the Merchandising page. For any product type in your catalog you can nominate up to three target types to push whenever it appears in a cart. The first slot carries the strongest boost, and your top pick is guaranteed a place in the recommendations. Changes apply the moment you save — there is no rebuild or re-sync to wait for.
- What happens if there is no mined rule for a type I want to push?
- RevLogic falls back to your best seller in that type. That means a brand-new merchandising play has something concrete to offer on day one, and it gets sharper as real order data accumulates behind it.