Ecommerce ops · Production
From Filters to Phrases: AI-powered natural language car search at coches.net
The problem
Users of coches.net had to navigate through multiple filters to find specific cars, as the system required knowledge of which filters to apply rather than supporting natural language queries.
Workflow diagram · grounded in source
1
User submits natural language query
trigger
“enable searches like "family-friendly SUV for mountain trips" or "sporty convertible under €30,000 in Madrid," making the search process more intuitive”
2
Content moderation check
validation
“we send the user's question to the AI model and ask it to return the following codes in its response: 0: if the question is valid, 1: if the question is unethical, 2: if the question is invalid”
3
AI translates query to filters
ai_action
“We employed the Few-shot prompting technique, supplying the AI with context about our filters and examples of searches. This approach helps the AI generate accurate and relevant responses.”
4
Verify and apply filters
output
“After verifying the AI's response, we send the filters to the Frontend, where the website applies them to perform the search”
Reported outcome
The AI search was deployed to production at coches.net with around 20,000 model invocations per day at a cost of about €19 per day.
Users leveraging AI search generated more value than those using traditional filters, though adoption remained limited.
Reported metrics
Daily model invocationsAround 20,000
Daily input tokens55 million
Daily output tokens4 million
daily AI search costabout €19 per day
Show all 8 reported metrics
daily model invocationsAround 20,000
daily input tokens55 million
daily output tokens4 million
daily AI search costabout €19 per day
input tokens per queryapproximately 6,000
output tokens per query500
user value from AI search vs traditionalgenerated more value compared to traditional methods
AI search adoptionadoption is still limited
Reported stack
Amazon BedrockClaudeHaikuFew-shot promptingAWS
Source
https://medium.com/adevinta-tech-blog/from-filters-to-phrases-our-ai-revolution-in-car-search-2d7c73ca4886
Read source ↗Frequently asked questions
What did this team achieve with this AI workflow?
The AI search was deployed to production at coches.net with around 20,000 model invocations per day at a cost of about €19 per day.
What tools did this team use?
Amazon Bedrock, Claude, Haiku, Few-shot prompting, AWS.
What results were reported?
Daily model invocations: Around 20,000; Daily input tokens: 55 million; Daily output tokens: 4 million; daily AI search cost: about €19 per day (source-reported, not independently verified).
How is this ecommerce ops AI workflow structured?
User submits natural language query → Content moderation check → AI translates query to filters → Verify and apply filters.