Request a demo
Pro plan: €199/mo with Web and WhatsApp, onboarding included and no setup fee. We build it and leave it running, with 3 maintenance sessions in the first year. Request your demo →

Article Ecommerce

AI agent for WhatsApp with a product catalog: when to use exact search and when to use RAG

How to connect an AI agent to products, compatible references and stock without making every answer slow or unreliable.

Practical guide

Built to help you decide, prepare and review a concrete part of your agent before customers use it.

Category Ecommerce
Published
Author David Cedillo

When a business sells many products, an agent cannot rely only on a FAQ page. It needs to check catalog data, references, compatibility, stock and conditions before replying.

The key is not putting everything into a huge document. The key is deciding what needs exact search and what should be handled with semantic knowledge.

Not everything should go into RAG

RAG works well for policies, explanations, repeated questions and content that can be phrased in many ways. A catalog with references, compatible vehicles, sizes, variants or SKUs needs structured search.

If a customer asks for an exact reference, the agent should check exact data. If the customer asks “do you have brake pads compatible with my model?”, it needs to combine compatibility search with a clear answer.

What to separate

Data typeBetter approachWhy
SKU, reference, EANExact searchThere is no room for approximation
Product nameText search with typo toleranceCustomers may write it incorrectly
CompatibilityStructured queryIt depends on brand, model, year or variant
FAQsRAGThe wording changes a lot
Policies and shippingRAG + rulesConditions need explanation

This separation makes the agent faster, cheaper and more reliable.

What the customer receives

The customer should not notice the architecture. They should see a useful answer:

  • Product found or a close alternative.
  • Compatibility explained carefully.
  • Price or range if available.
  • Stock state if the business publishes it.
  • Next step: buy, reserve, leave contact details or talk to the team.

If there is no certainty, the agent should say so. In parts, health, technical products or regulated categories, invented compatibility can be expensive.

Sync: simple and frequent is usually enough

For many businesses, syncing every few hours is enough. If stock changes very quickly, the agent can query live data. The important part is deciding which data needs real time access and which data can live in a prepared copy for the agent.

A good first version is usually:

  1. Sync products and compatibility data.
  2. Create exact search for references.
  3. Add typo-tolerant search for product names.
  4. Use RAG for questions, policies and commercial explanations.
  5. Store conversations to find catalog gaps.

That gives the agent a stronger first version and makes it improve with real usage.

captia

Want to see this applied to your business?

We can show how Captia would work with your real information, channels and rules.

Request a demo
Back to blog

03 Request a demo

Request a demo

Tell us a bit about your business and we'll show you your agent in action. No commitment.