Why behavioral signals drive conversion today and agent visibility tomorrow
A shopper types “light jacket” into the search bar, the keyword is indexed, the rules fire, results populate, and the shopper leaves. The product catalog had the right products, and the query matched a real category, but the system missed that this shopper had spent 4 minutes browsing summer dresses before searching. “Light jacket” meant an evening layer for a vacation, not a mid-weight outerwear piece for fall, and the difference between those two purchases did not appear in any keyword report.
Ecommerce teams have spent years improving what happens when shoppers type a query, but understanding what shoppers mean when they type it is a harder and more consequential problem. It is also the infrastructure that determines whether an AI shopping agent recommends your products or passes them over.
The keyword gap and why it compounds
Keywords are a proxy for intent, not intent itself. “Running shoes” typed in January carries a different purchase signal than the same query in September. New Year’s fitness buyers are a different shopper type than back-to-school athletes, yet a keyword-based system surfaces the same products to both. “Gift for dad” produces no logical product category from the words alone, leaving the system to default to bestsellers that may or may not fit the occasion, the recipient, or the budget.
Rule-based merchandising built on keywords makes the underlying problem worse. Merchandising teams write boost-and-bury rules for the queries they can predict, but the long tail of actual shopper language is wider than any rules library can cover. Queries carry different meanings depending on the season, session context, and browsing behavior, while a rule written three months ago may conflict with what a shopper needs today.
Across weeks and months, these near-misses compound. A shopper searches, receives results, finds nothing that matches their intent, and leaves. Standard analytics records the session as a bounce and treats the search as resolved. Zero-result pages for queries the product catalog can serve, and high-bounce sessions following a technically successful search, rarely trigger the same urgency as checkout abandonment, which is why revenue loss grows without a response.
What intent-based merchandising means
Intent-based merchandising uses behavioral signals alongside the literal search term to determine which products to surface and in what order. Those signals include browsing history, click patterns within the current session, and contextual factors such as time of year and device type. None of these inputs change what a shopper types, but they substantially change what the query means.
A shopper who spends four minutes browsing summer dresses and then types “cover-up” is most likely looking for a lightweight layer for a warm-weather occasion, while the same query from a shopper browsing knitwear in October signals something different in category, weight, and context. An intent-aware system returns different ranked product results for each shopper, calibrated to the current session, rather than returning the same products to both shoppers.
The distinction from basic personalization matters here. Showing a returning shopper their last-viewed items is a form of personalization, but it does not resolve the intent problem on its own. Intent-based merchandising updates the ranking logic in real time based on what the current session communicates. Personalization identifies who the shopper is, and when that context combines with those live intent signals, the platform produces more accurate results than either input can alone.
How merchandisers stay in control and why revenue follows
For most merchandising teams, the real question about intent-based systems is whether they preserve the ability to run campaigns, enforce promotional priorities, and maintain brand standards while AI dynamically adjusts results. They do, because intent-based systems and manual merchandising control operate at different layers.
Romane Vernet
Michael Stars saw a 951 percent increase in revenue per visit after implementing AI-powered product discovery with Athos Commerce.1 That outcome required intent-aware systems operating at the scale and responsiveness that manual rules cannot match, but it also required merchandisers to make deliberate decisions about promotional priorities and commercial guardrails. The AI executed within those guardrails.
Merchandisers define promotional priorities and commercial guardrails, and the platform applies intent signals within those parameters at a scale no individual rule set can replicate. Early Settler reduced weekly merchandising work from up to 15 hours to three to five hours after implementing automated merchandising rules, while simultaneously achieving a 10 to 15 percent conversion rate improvement on in-stock and ready-to-deliver products.2
Past behavior tells the platform who the shopper is. Session data supplies what they need right now, and a platform that reads both produces sharper results than one reading either in isolation.
Intent data as infrastructure for AI agents
The behavioral data and structured intent signals generated by intent-based merchandising go beyond on-site conversion: they are the same inputs that AI shopping agents use when making purchase decisions on behalf of shoppers.
An AI agent evaluating “trail running shoes under $120” does not browse, filter, or compare the way a human shopper does. It queries structured product data and evaluates which merchants’ products best match the request’s intent and constraints. Merchants who have built richer behavioral and intent signals tend to perform better in those evaluations, while those relying on keyword-only product data are more likely to be passed over.
Investing in intent-based merchandising now pays off beyond on-site conversion. As a retail team builds behavioral data, refines intent signals, and improves its product catalog structure month over month, the AI models serving shoppers onsite and through external agents grow more accurate. Morgan Stanley estimates agentic shoppers could drive up to $385 billion in U.S. ecommerce by 2030, equal to 10 to 20 percent of total ecommerce spend.3 Retailers who begin building the underlying data infrastructure today will compound that advantage over the years it takes for that market to mature. Those who wait will find themselves separated from competitors by years of behavioral data they did not collect, and that data, once absent, cannot be reconstructed.
Where to start: three areas to audit
For most retail teams, the barrier to intent-based merchandising is diagnostic rather than technological. Most teams do not have a clear picture of where keyword-based logic is actively costing them revenue. Three audits tend to surface the largest opportunities.
Zero-result queries: Zero-result pages commonly indicate a mismatch between shopper language and catalog indexing, even when the right products exist. Pulling zero-result query logs and mapping them to products that exist in the catalog but are not appearing in results is the fastest signal that keyword coverage has broken down, and it is the audit that reveals the largest single concentration of missed revenue.
High-traffic, low-add-to-cart searches: Queries that return product results and generate clicks but consistently fail to convert are evidence of intent mismatch at the result level. A shopper searching “home office chair” who clicks through three results without adding anything to the cart likely came with context that the results did not address: ergonomic requirements, a price range, or an aesthetic preference. Identifying these queries through search analytics and reviewing the results makes the intent mismatch measurable.
High-bounce sessions after a successful search: When shoppers search, receive product results, and leave without converting, the session reflects intent; the results are never addressed. Cross-referencing these sessions with browsing history and query sequence reveals the misalignment. The system matched a keyword but missed the meaning, and standard analytics attributed the departure to the destination rather than to the context that preceded it.
Each of these audits requires visibility into search behavior data, not a platform migration. They are sequenced to move from identifying the largest drop-off to measuring it at the result level, and finally to confirming the pattern across sessions. Together, they give merchandising teams a specific picture of where intent-based improvements will deliver the most immediate return.
The shopper who typed “light jacket” left despite the product catalog containing exactly what they needed, because the system read a keyword and missed the four minutes of summer browsing that had given the query its actual meaning.
Retailers who close this mismatch see an immediate lift in conversion, since every near-miss costs revenue today. Over the longer term, this work builds the behavioral data asset that AI agents will query as agentic commerce grows. Every month of intent signals refined and behavioral data collected adds to an advantage that cannot be reconstructed after the fact.
Athos Commerce’s Intelligent Discovery Platform (combining search, merchandising, and personalization) helps brands identify where intent signals are being missed and build a product discovery experience that serves both shoppers and AI agents.
Frequently Asked Questions
What is intent-based merchandising?
Intent-based merchandising uses behavioral signals alongside a search query to determine which products to surface and in what order. Those signals include browsing history, click patterns within the current session, and contextual factors such as time of year and device type. Unlike keyword-based systems, which return the same products to every shopper typing the same query, intent-based systems calibrate results to what the current session communicates and return different ranked products for shoppers with different browsing contexts.
How does intent-based merchandising differ from keyword-based search?
Keyword-based search returns results by matching words in a query against words in a product catalog. Intent-based merchandising reads behavioral signals alongside the query (browsing history, click patterns, session context) to interpret what the query means for this particular shopper. A shopper browsing summer dresses who types “cover-up” receives different ranked results than a shopper browsing knitwear in October typing the same query, because the intent behind identical queries differs by session context.
How is intent-based merchandising different from personalization?
Personalization identifies who a shopper is by drawing on past behavior such as purchase history or last-viewed items. Intent-based merchandising updates ranking logic in real time based on what the current session is communicating, independent of long-term shopper identity. The two inputs are complementary: personalization supplies the shopper profile, session intent signals supply what they need right now, and a platform that reads both produces more accurate results than either input alone.
Why does intent-based merchandising matter for AI shopping agent recommendations?
AI shopping agents evaluate product catalogs to make purchase recommendations on behalf of shoppers. They query structured product data and assess which merchants’ products best match the intent and constraints of the request; they do not browse or filter the way a human shopper does. Merchants who have built richer behavioral and intent signals perform better in those evaluations. Those relying on keyword-only product data are more likely to be passed over, because sparse data signals poor catalog quality to an evaluating agent. Athos Commerce’s Merchandising and Search products are built to surface and act on these signals at scale.
Where should ecommerce teams start to identify intent merchandising gaps?
Three audits surface the largest opportunities. First, pull zero-result query logs and map them to products that exist in the catalog — this identifies where keyword coverage has broken down. Second, find high-traffic searches with low add-to-cart rates, where products appear but intent is mismatched at the result level. Third, cross-reference high-bounce sessions following a successful search with browsing history and query sequence to confirm where the system matched a keyword but missed the meaning behind it.
Sources & Further Reading
- Athos Commerce. “Michael Stars Case Study.” Athos Commerce, 2025. https://athoscommerce.com/case-studies/michael-stars/
- Athos Commerce. “Early Settler Case Study.” Athos Commerce, 2025. https://athoscommerce.com/case-studies/early-settler/
- Morgan Stanley. “Agentic Commerce Market Impact Outlook.” Morgan Stanley Insights, December 2025. https://www.morganstanley.com/insights/articles/agentic-commerce-market-impact-outlook
