Contextual spice and recipe suggestions
The Spice House lets shoppers search the way they cook.
Semantic search turns cooking questions into relevant spices and recipes, even when the shopper does not know the exact product name.
The impact
38%
fewer searches without a useful result
Contextual spice and recipe suggestions
Case study facts
- Industry
- Food & specialty retail
- Team
- Marketing
- Systems connected
- 2
- Time to live
- 1 week
- Engagement
- AI implementation
The impact
Results for the team
- fewer searches without a useful result
- 38%
- fewer searches without a useful result
- More cooking queries return relevant spices or recipes, reducing searches that leave shoppers without a useful match.
- higher search-to-product engagement
- 24%
- higher search-to-product engagement
- Shoppers engage more with catalog products surfaced in response to their cooking questions.
- more recipe-to-product visits
- 29%
- more recipe-to-product visits
- More visits move from recipe content to the related spice products.
Situation
Shoppers knew the dish, not the product.
Keyword search works best when a shopper already knows the catalog's language. A request about a dish, flavor or ingredient substitution may not share a single word with the right product.
Challenge
Keywords missed the cooking intent.
A natural-language request needed to find products and recipes even when the wording did not overlap. At the same time, exact-name search, ingredient restrictions and requests with no suitable match still needed reliable handling.
What we built
Search that connects recipes and spices.
The search agent interprets the cooking intent, retrieves relevant recipes and products, and reranks them against the request. Exact product-name search remains available alongside semantic retrieval.
Answers cite the recipe or product facts behind each suggestion, and unclear constraints trigger a question instead of a confident but unsuitable answer.
- 01
Interpret the question
- Intent recognition
Identify the dish, desired flavor and any explicit constraints.
- 02
Retrieve by meaning
- Hybrid search
Search recipes and products semantically as well as by keyword.
- 03
Apply constraints
Respect ingredient exclusions and what is actually in stock.
- 04
Rerank for intent
- Reranking
Prioritize results that answer the actual cooking request.
- 05
Show the connection
- RAG
Explain which recipe or product attribute makes each result useful.
- 06
Refine together
Shoppers can clarify heat, ingredients or cooking style in a follow-up.
Evals and guardrails
Find relevant recipes. Respect restrictions.
Search evals
Measure whether suitable recipes and products are found and ranked first. Test misspellings, exact names and requests with no suitable result.
Evidence evals
Check answers against the recipes and products found. Test source references and suggestions for ingredient swaps without supporting evidence.
Guardrails
Apply ingredient restrictions before showing results. Clarify unclear requirements and return no match when there is insufficient evidence.
Results
Cooking questions lead to useful products.
More cooking queries return relevant spices or recipes, reducing searches that leave shoppers without a useful match. Shoppers engage more with catalog products surfaced in response to their cooking questions.
More visits move from recipe content to the related spice products.
- fewer searches without a useful result
- 38%fewer searches without a useful result
- higher search-to-product engagement
- 24%higher search-to-product engagement
- more recipe-to-product visits
- 29%more recipe-to-product visits
From The Spice House
“An invaluable resource to have on your side. They work quickly, find solutions tailored to how the business actually runs, and consistently deliver high-quality work. Beyond the results, they are a pleasure to work with.”
Build Capacity To Grow. Own Your Intelligence.
Bring the workflow slowing your sales, orders or delivery. We'll map the bottleneck and show you what the AI system could look like.