Skip to content
Start free trial
WhatsApp and customer experience

AI stylists and product recommenders on WhatsApp: how they work

How a WhatsApp assistant learns a shopper’s taste in three questions, ranks products from your catalogue, explains why they fit and hands over to a person.

By Orvexa TeamUpdated 3 min read
A lit-up mobile phone shop at night in Dabri, Delhi
Photo: VartamanVarta, CC BY 4.0 · All photo credits
Key takeaways
  • A good recommender asks a few easy questions, then explains why each product fits
  • It recommends only from your catalogue, which keeps it accurate and within WhatsApp's rules
  • Quality depends on well-tagged products and a smooth hand-off to a person

A shopper messages your store at 11 p.m. asking what would suit a cousin's wedding. If nobody answers until morning, you may have lost the sale. A recommendation assistant answers at once, and it can do so in a way that feels like a good shop assistant, not a search box.

The basic flow

  1. A shopper starts the chat, from an ad, a QR code or your website.
  2. The assistant asks a few quick questions, usually with tap-to-reply buttons. For a fashion brand: the occasion, the style they like, their budget.
  3. It builds a taste profile from the answers, such as "earthy, handmade, festive".
  4. It ranks your products against that profile and shows the best few, with price and a one-line reason ("great for weddings, within your budget").
  5. It offers next steps: order, see more, or talk to a person.
  6. It remembers the shopper, so next time it can open with new arrivals in their style.

You can see this flow in the demo on our site. It is a simulation with a sample catalogue, and the assistant itself is coming soon.

What makes the recommendations good

  • Product tags. Each product needs structured information: occasions, styles, materials, price band. The assistant can only match what you have described.
  • A clear scoring rule. A transparent approach, such as points for matching the occasion, the style and the budget, is easy to test and explain. It also avoids made-up answers.
  • Reasons shown. "Why this?" builds trust and teaches you what customers value.
  • A minimum quality bar. If nothing fits well, say so and offer a person instead of showing weak matches.

Why stay within your catalogue

A store assistant that talks only about your products is accurate, and it fits Meta's rules, which restrict general-purpose AI assistants on the WhatsApp Business Platform while allowing business assistants for support and sales (the rules). It also avoids the risk of inventing products or promising things you cannot deliver.

Human hand-off matters

Give people a clear way to reach a person at any point, and pass along what the shopper has already said, so they do not repeat themselves. Plan for the hours when nobody is online: tell the shopper when to expect a reply.

Metrics to watch

MetricWhy
Completion rate of the questionsAre they too long?
Click-through on recommendationsAre matches relevant?
Orders from chatsThe business result
Hand-off requestsWhere the bot falls short
Opt-outsWhether it is being used well

Respect privacy

Tell people what you store and why, and let them delete their profile. See taste profiles and personalisation.

What to do next

Pick ten best-selling products and tag them with occasion, style and price band. Even before you build a bot, that exercise shows how well your catalogue is described.

Frequently asked questions

Do I need AI to make recommendations?

Not necessarily. A rules-based ranking on well-tagged products can work well. AI can help with language and tagging, but good data matters more.

Will it replace my team?

No. It handles the first questions at any hour, and hands over to your team for anything it cannot do.

Can it work in Hindi and other languages?

That depends on the setup. Consider supporting the languages your customers use, with simple button choices that translate well.

Topics:ai stylistrecommendationswhatsapppersonalisationchatbot

Keep reading