Skip to content

The six pieces of a WhatsApp assistant that truly serves customers

A menu-driven answering machine doesn't serve anyone. Using a bike shop as an example, we explain the six pieces that turn your business's WhatsApp into an assistant that understands, doesn't make up prices, looks at photos, follows up, knows when to hand the conversation over to you, and records everything.

5 min read
Mostrador de un taller de bicis con un móvil, una patilla de cambio y una cubierta de 29

If your shop's WhatsApp answers on its own, you already have an assistant. Or so it seems. Because an assistant that replies “press 1 for sales” doesn't serve anyone: it makes people wait. And the customer who wanted to know whether you carry 29-inch tires goes to the shop across the street.

An assistant that truly serves customers is built piece by piece. Each piece fixes a problem that appears as soon as the previous one works. We'll tell you about them with an example: a bike shop and repair service that serves customers on WhatsApp with ShephERP.

Piece 1. The model: it understands what people write

The problem: the menu-based answering machine forces the customer to talk like a machine. Nobody writes “1”; they write “hey, do you carry 29-inch tires?”.

The piece: the assistant reads the question exactly as it arrives, with typos, abbreviations or in another language, and answers in the same language. No menus. And it always says it's an assistant: on WhatsApp, each of its messages starts with “Virtual assistant (AI)”, because whoever is writing has the right to know they're not talking to a person.

What the customer writes and what the assistant replies, signed as AI. (Sample data.)

Piece 2. The knowledge base: your catalog, not what it imagines

The problem: as soon as it understands, people ask how much something costs. A language model that doesn't know a price tends to make one up, and a made-up price in a chat is a problem with the customer.

The piece: in ShephERP the assistant checks your catalog: the price you've set and the units left in stock. If something isn't there, it doesn't claim otherwise: it says so and, if needed, hands the conversation over to a person.

The shop's catalog: price and stock for each product. It's what the assistant checks before answering. (Sample data.)

Piece 3. Vision: it looks at photos

The problem: many people don't know the name of what they need. They send you a photo of the broken part and a “this thing broke on me”.

The piece: the assistant looks at the photo, describes what it sees (a bent derailleur hanger, for example) and searches your catalog for a match, with its price and stock.

The product page it finds from the photo: price and units in stock.

Piece 4. Follow-up: it comes back when the customer cools off

The problem: you give them the price, they say “I'll think about it”… and they disappear.

The piece: from the customer record you schedule a follow-up for the assistant: when to write and what it needs to know (“I quoted them the price of the hanger and they said they'd think about it”). At that time it messages them in the same chat they used to talk to you, with their history at hand. And after each delivery, if you want, it asks how the order arrived.

Scheduling a follow-up for the assistant from the customer record: when and with what context.

Piece 5. The human handoff: it knows when not to decide

The problem: a club writes in asking for ten bikes at a discount. That's not something a machine should decide.

The piece: the handover to a person. The assistant passes the conversation to your team with the reason (“they're asking for a price on ten bikes for the club”) and the urgency. It goes first to the person responsible for that customer and, if they're unavailable, to whoever is on duty. While a person handles the conversation, the assistant steps aside and doesn't reply over them.

The conversation handed over to a person, with the reason and the urgency.

Piece 6. The CRM: it records everything in the customer record

The problem: if everything lives in the chat, a month later nobody knows who asked what, what was offered, or how it was left.

The piece: if the person writing isn't a customer yet, the assistant creates their record with their WhatsApp phone number. Every conversation is stored in that record inside ShephERP, alongside their orders and history: what they asked for and where things stood. Whoever serves them next, person or assistant, knows where things are coming from.

The customer record with their WhatsApp conversation and their order.

All six at a glance

  • Model: understands what people write, no menus, and says it's an assistant.
  • Knowledge base: answers from your catalog, with real prices and stock; what it doesn't know, it doesn't invent.
  • Vision: looks at photos and finds the part in your catalog.
  • Follow-up: writes again when the time is right, in the same chat.
  • Human handoff: passes on what isn't its call to make, and stops replying.
  • CRM: records everything in the customer record.

If your current assistant doesn't tick one of these boxes, you now know where to start.

In ShephERP it comes pre-assembled

In ShephERP the assistant isn't an add-on: it works on the same catalog, orders and customer records you use to sell and invoice. That's why it can answer with your prices, alert you when it shouldn't be the one deciding, and keep everything recorded.

It's all explained at shepherp.com. And if you want to see it working with your shop's products, write to us from there and we'll show you.

  • #ShephERP
  • #WhatsApp
  • #asistente IA
  • #atención al cliente
  • #CRM
  • #pymes

All news