14 September 2026 · Pedro Aldea
The recommendation engine your sales team won't use
Cross-sell lists generate themselves and die in the CRM because they don't explain their reasoning. What changes when every suggestion leads with its why.
If a system tells your reps what to sell and doesn’t tell them why, it has produced a list nobody will work. A recommendation without its reasoning asks the rep to take the machine’s word over their own judgment, and experienced reps don’t do that. What most projects automate is the list, not the justification. The justification is the product.
The scene that repeats
A rep opens the CRM before Tuesday’s route. There’s the list with suggested products, suggested accounts and flagged opportunities. A system generated it from more data than any person could work through by hand. The rep closes it without reading it. That’s not laziness. It’s experience.
Someone who has carried a bag for years knows what sits behind every visit. When the system says “this account might need this” without saying why, it asks them to trust a machine’s opinion over their own read of the account. Every unexplained line is an argument they would have to win with themselves first. So the list goes unread, and the engine becomes one more tab nobody opens.
The project ships as a success. The recommendations generate. The dashboard is green. In the field, nobody sold anything different from last week.
The failure isn’t the model
The failure is the request. The recommendation was treated as a data output instead of a selling tool. A selling tool has to answer the questions a rep faces in front of the buyer, why this product for this account, why now, and what happens when the buyer pushes back. If the answer to all three is “the system calculated it,” the conversation starts lost.
There’s a second problem underneath. Many of these lists run on data nobody cleaned. The same brand spelled five ways, duplicate part numbers, one customer living under two names. On a base like that, the engine can recommend a product the account already buys under a different name. When that happens twice, trust doesn’t come back.
The reason is a data field, not nice copy
In the Zero Friction Method the order matters. First you eliminate what adds nothing. Then you standardize and simplify what’s left. Only at the end do you automate. With recommendations, the order shows just as clearly.
We eliminated recommendations without a reason. If the system can’t explain why it’s suggesting a product, the line doesn’t make the list. It reads like a cut and works like the opposite. Fewer lines, better defended.
We standardized the reason as data. The rule is written before anything gets automated. “This account orders this part every six weeks and the last order was eight weeks ago.” “Three similar accounts started with this product.” Nobody writes that text by hand each cycle. The same data that produces the suggestion produces the sentence, in the same shape every time.
We simplified the output. The recommendation arrives with its argument in front, in the language of selling, not in score columns someone has to translate first. Then we automated the repetitive part, which is reading order history, matching profiles and tracking reorder cadence.
The data underneath needed its own work first. In our catalog categorization case, 327 brands written in more than 1,400 variants ended up with clean names. A recommendation engine that crosses brands and part numbers stands on that base. Without it, it suggests garbage that looks good.
What we built
In the recommendation work for an industrial distributor, the reason moved in front of the suggestion. The receipt closes the story. More than 7,300 B2B customers now receive personalized cross-sell and reorder recommendations, and the 115,000-plus opportunities identified were worked by the sales team, not filed in a report.
The number that matters isn’t the volume. It’s who did the work. The sales team picked the list up because every line carried its why, and that turned an unread tab into a working tool. Adoption wasn’t requested. It arrived when the recommendations stopped asking for faith.
A 15-minute check
Open the last recommendation list your system produced. No system yet? Take your best rep’s notes on what to offer whom. Pick ten lines and mark the ones that answer, on sight, why this product for this account.
If half or fewer pass, the list is going dark, whatever the dashboard says. The fix fits in the same half hour. Whoever generates the list, require a reason on every line, and drop the lines that can’t show one. In two weeks you’ll know whether the rep is opening it again.