Tuning Elasticsearch relevance with real business logic
Layering a business-tuned boost model on top of endress.com's Elasticsearch-powered onsite search, instead of replacing its relevance engine.
Problem
endress.com's onsite search runs on Elasticsearch, whose relevance scoring is built around text-match quality and has no built-in concept of the business signals that actually determine whether a listing is relevant to a B2B buyer at a given moment: its lifecycle phase, its segmentation, whether it's in stock, or whether it's even available in the customer's market.
Approach
Rather than replacing Elasticsearch's relevance engine, I initiated and led the design of a boost layer on top of it: a calculation model that scores each product on lifecycle phase (new, active, phase-out), segmentation, stock availability, and market availability, and translates that into boosts and penalties applied on top of Elasticsearch's native relevance score. Elasticsearch still does what it's best at, matching and ranking by search intent, while the boost layer adds the business context it has no visibility into by default.
Outcome
Search results and product listings on endress.com now reflect both what a customer is looking for and what is actually relevant to buy right now, without discarding Elasticsearch's underlying relevance model, a boost layer that can keep absorbing new business signals as they become relevant to weight in.
Related knowledge
Working on something similar?
Let's talk about data and AI driven product decisions for your team.
Get in touch