The short answer
Fix search before recommendations. A shopper using search has already told you exactly what they want — failing them there is the most expensive miss in the funnel. Semantic search, which matches meaning rather than keywords, is usually the highest-return AI investment an online store can make.
Why keyword search loses sales
- A shopper searching "warm jacket for hiking" gets nothing if your catalogue says "insulated outdoor shell".
- Typos and plurals produce zero results in exact-match systems, and a zero-result page is an exit.
- Product descriptions are written by merchandisers; searches are written by customers. They rarely use the same words.
Semantic search closes that vocabulary gap because it matches on meaning.
Where recommendations do pay
Complementary items at cart
- Genuinely useful, and the intent signal is strong.
Re-engagement
- Based on real behaviour, not a generic "customers also viewed" carousel that everyone ignores.
The unglamorous prerequisite
None of this works on bad catalogue data. Inconsistent attributes, missing descriptions, and duplicate SKUs defeat every model you put on top. The AI project is usually a data-quality project wearing a better title.