HandiRamp Search Token Scoring

A five-slide explanation of why keyword scoring surfaced the Ford E-Transit article for “pipe ramp,” and why the search fix belongs in product-family routing before rank math.

Baseline search How Relevanssi Handles Search It improves WordPress search by indexing words, finding term matches, and ranking pages mathematically. Site content Products Pages Posts, categories, tags Relevanssi index Searchable word map across selected fields Title, body, excerpt, fields User query pipe ramp Search phrase entered onsite Terms scored pipe ramp Matches get relevance weight What “scoring” means Each result earns points based on where terms appear and how strong the term match looks. Common inputs Title match Body/excerpt match Custom fields, taxonomy, SKU Simple mental model score = term overlap + field weight + term rarity Takeaway: Relevanssi is keyword/token relevance search. It is stronger than default WordPress search, but it does not inherently understand product-family intent. Slide 1 of 5 HandiRamp storefront search review
Agent-built prototype How Token Scoring Connected “Pipe Ramp” to Ford E-Transit The prototype made a mathematical language connection across indexed text, not a product-category decision. 1. Query entered pipe ramp Customer meant protector ramps 2. Tokens created pipe ramp The scorer evaluates each token 3. Indexed article text “zero tailpipe emissions” “Van Ramp” 4. Score awarded tailpipe = pipe hit ramp = ramp hit Both query tokens appeared Token math view pipe match ramp match guide weight Calculated relevance pipe + ramp + page weight Enough score to rank in All results Selected result Ford E-Transit Electric Van Ramp A guide/article page surfaced Technical summary: the agent-built search prototype ranked the Ford article because token scoring found “pipe” inside “tailpipe” and “ramp” in the article context. Slide 2 of 5 HandiRamp storefront search review
Broader search concept Search Engines Rank by Matches Plus Weights The Ford result was a small-scale version of the same ranking problem every search engine solves. Search query pipe ramp Text signals Token matches Phrase matches Field weights Title > heading Body > footer text Page weights Product, guide blog, locator Rank #1 selected Generic ranking formula relevance score = text match strength + field importance What page weight means A match in a title, product page, or guide may receive more value than the same word in plain copy. HandiRamp example The prototype gave points for: pipe + ramp + guide/article weighting Takeaway: ranking quality depends on the weighting model. If token overlap outweighs product-family meaning, the math can select a technically related but commercially wrong page. Slide 3 of 5 HandiRamp storefront search review
Moving forward From Token Matching to Product Meaning The next search layer routes each query through HandiRamp product families before ranking pages. Customer query pipe ramp Normalize intent Aliases first hose ramp, cable protector, hose bridge, crossover Map to family MATERIAL_HANDLING Pipe/Hose/Cable Ramps Hose/Cable Ramp Correct result pool Public products shop hubs relevant articles Rank inside family What changes now Token matches no longer choose the result pool by themselves. They support a family decision. Business rules first SKU and size intent stay precise. Category queries get hubs/lists. Word boundaries block tailpipe. AI's role Optional second pass: map messy language to a known family, then explain/rerank. Takeaway: the system should first understand what product family the customer means, then rank results inside that family. Slide 4 of 5 HandiRamp storefront search review
Learning loop How Search Gets Better Over Time We log searches and outcomes, then use the evidence to improve aliases, family routing, and ranking weights. 1. Search logged Query, timestamp, page context 2. Results shown Rank order, tabs, zero-result cases 3. Behavior observed Clicks, refinements, time on page 4. Outcomes measured Quote, cart, phone, form, or quick return to search Strong learning signals Quote request or add to cart Configurator use Phone-click or contact form Weak but useful signals Time on page Bounce or quick return Search refinement patterns What improves Alias map Family routing Hub and result weighting Takeaway: the system does not guess blindly. Search behavior becomes a feedback loop for improving the ontology and ranking model. Slide 5 of 5 HandiRamp storefront search review