Industry study · Retail
Develop shopping assistance around the tradeoffs your merchandising experts understand. Teammately turns their judgments into rubrics and coverage for product recommendations, comparisons, and substitutions.
For shopping-agent owners, merchandising teams, and customer-experience leaders
What changes the decision
Several products can match a request while serving the customer differently. Availability, intended use, attribute priorities, acceptable compromises, and commercial constraints influence what an experienced merchant recommends. A benchmark needs to capture those judgments without reducing every task to one preferred product.
Applied industry study · These are development and assessment methods, not reported customer results.
The context
Two products match the request, but one better fits the intended use while another satisfies an unstated assumption. The agent may need to clarify which attribute matters most.
Judgment to capture
Which evidence supports a recommendation? When is clarification useful, and when are several alternatives defensible?
The context
The preferred item is unavailable. A nearby alternative can look similar while lacking the feature that made the original appropriate.
Judgment to capture
Which attributes are essential to this customer? What may be traded off, and what difference must the recommendation explain?
The context
Commercial priorities can influence recommendations among suitable products, but may favor an option that misses the stated need or implies an unsupported benefit.
Judgment to capture
Where may merchandising preference affect the ranking? Which customer requirements and product facts must remain decisive?
The people behind the standard
Agents prepare product and response comparisons, then adapt follow-up questions as experts explain their choices. Findings become proposed rubrics for fit, substitution, and explanation, with policy boundaries preserving how customer requirements and merchandising preferences relate.
Product fit, meaningful attributes, assortment intent, and the tradeoffs behind a preference.
How customer intent is expressed, when to clarify, and what makes an explanation useful.
Reliable product facts, availability, and the conditions that make an alternative feasible.
What the evidence supports and which comparisons or promises require qualification.
Coverage and improvement
Make the reasons behind a recommendation available to development. Experts establish what fit means; coverage and synthesis expose where that judgment changes; trials test whether the candidate respects those distinctions.
Teammately’s contribution
Rubrics distinguish essential customer requirements, supported product claims, and preferences among suitable options. Policies make clear where merchandising priorities may influence a choice and where customer fit remains decisive.
What this makes possible
A recommendation can be acceptable without matching one fixed product ranking. The criteria explain which customer requirements and tradeoffs justify it.
Teammately’s contribution
Lemon Weave synthesizes requests, catalog context, and alternatives from the selected ontology tuples. Vary one consequential attribute or availability condition so reviewers can establish why the recommendation should change.
What this makes possible
Cases distinguish product similarity from an acceptable substitute and preserve situations with several legitimate answers.
Teammately’s contribution
Trialground compares recommendations and explanations against the rubrics. Investigate whether failures come from missing product context, an assumed preference, or the way the harness selects and explains alternatives.
What this makes possible
A specific basis for testing better clarification, retrieval, or recommendation logic with Lemon Goal’s harness prototypes.
Harness experiments can improve context gathering and recommendation logic. Lemon Weave can also construct training examples around preference boundaries the weights handle inconsistently, using the same expert rubrics and coverage design.
Assessing progress
Inspect recommendation quality by customer intent, attribute combinations, and availability. A gain on straightforward product matches can conceal weak substitutions or an increased tendency to invent benefits.
Keep multiple acceptable options where expert judgment supports them. Conversion or ranking agreement alone cannot establish whether the recommendation respects the intended standard; examine the reasons and tradeoffs in the response.
Returns and recovery use some of the same product and customer distinctions, but add purchase terms, fulfillment history, and remedy authority. A suitable replacement must be both appropriate for the customer and feasible to fulfill.
Develop separate rubrics and coverage for those decisions. Reuse established product knowledge where it applies, then elicit the service-specific conditions behind refunds, replacements, exceptions, and escalation.
Adjacent workflows
Explain consequential differences without inventing attributes or treating every tradeoff as universal.
Preserve product truth and approved claims across generated descriptions.
Judge available remedies against purchase terms, customer needs, and operational constraints.
Start with the customer context, the available products, and the people who can explain the tradeoff. We can identify the criteria and coverage that would make that judgment useful in development.