AI customer service on WhatsApp for B2B sales: how to do it

In B2B, the first company to reply often wins the conversation. Yet many leads wait hours for an answer to a simple question: do you work with stainless, do you ship to my region, can I send a drawing here. AI can close that gap, if it is set up to help the sales team instead of pretending to be it. This guide covers how we approach automation and AI for first-response customer service.

What should AI handle, and what should it not?

Good for AI	Keep with people
First reply, any time of day	Custom quotes and negotiations
Opening hours, address, delivery regions	Delivery dates for specific orders
Which services and materials you work with	Technical judgment on a drawing
Collecting name, company, part, quantity, files	Complaints and frustrated customers
Status of an order from your system, if integrated	Anything with legal or safety implications
Routing to the right person or department	Final decisions and commitments

The rule of thumb: AI answers facts you would put on your website, and collects information; people make judgments and commitments.

How do you keep the AI from making things up?

Large language models are fluent, which makes wrong answers sound convincing. Three controls keep them reliable:

1. Approved content only. The assistant answers from a knowledge base you wrote and approved: services, materials, regions, policies, FAQ. Nothing else.
2. Permission to say “I don’t know”. When a question is outside the content, it says so and offers a person, instead of guessing.
3. Logging and review. Every conversation is saved. Someone reviews a sample each week, fixes unclear answers and adds missing content.

When should the conversation go to a person?

Define hand-off triggers before going live:

• The customer asks for a price or delivery date on custom work.
• The question needs technical judgment (tolerance, material choice, feasibility).
• The customer shows frustration or asks for a person.
• The conversation reaches a ready-to-quote state: all details collected.

The hand-off should carry the context, so the salesperson does not ask the same questions again. A short summary (company, need, quantity, files, urgency) at the top of the conversation saves everyone time.

Which tools and channels should you use?

For WhatsApp at business scale, use the official WhatsApp Business Platform (Cloud API), directly or through an official provider. It supports automated replies, templates and integration with your systems, and it requires following WhatsApp’s messaging policies, such as getting opt-in for business-initiated messages. Tools that automate a personal WhatsApp account break the rules and risk losing the number.

The same assistant can serve website chat and email, from the same approved content, so answers are consistent across channels.

What about data protection?

Customer conversations contain personal data: names, phone numbers, company details and sometimes drawings. Privacy laws apply: the GDPR in Europe, the LGPD in Brazil and similar rules elsewhere. In practice:

• Tell people they are talking to an automated assistant and how their data will be used.
• Collect only what you need to answer and quote.
• Protect the data: access control, encryption in transit, providers with clear data processing terms.
• Define retention: how long conversations are kept and when they are deleted.
• Keep drawings confidential: files go to your systems, not into public tools.

International frameworks such as the OECD AI Principles also point the same way: transparency, human oversight and accountability.

How do you roll it out?

1. Map the questions. Export a month of conversations and group them. Usually ten to twenty questions cover most of the volume.
2. Write the answers. Short, approved, in your company’s voice.
3. Define the hand-off rules and who receives each type of conversation.
4. Pilot in parallel with the team for a few weeks, reviewing every conversation.
5. Go live and keep a weekly review routine.

If your bottleneck is not the first reply but the quote itself, automating the calculation behind it may pay off more; see automating machine shop quotes with Python and spreadsheets.

Which numbers show it works?

Metric	Before vs after
Time to first reply	Should drop from hours to seconds
Conversations handled without a person	Share of simple questions resolved
Qualified leads handed to sales	Should rise, with complete details
Hand-offs with missing information	Should fall as content improves
Customer complaints about the assistant	Should stay rare; review each one

What are the most common mistakes?

• Pretending the assistant is a person. Customers notice, and trust drops. Say clearly that an automated assistant answers first and a person follows.
• Letting the AI quote. A wrong price or deadline sent by an assistant becomes a commercial problem. Keep quotes with people.
• Endless menus. “Press 1, press 2” flows frustrate buyers. Let them write naturally and collect details through the conversation.
• No hand-off at night. If the assistant collects everything at 10 p.m., someone must pick it up first thing in the morning; otherwise the speed gain disappears.
• Set and forget. Without weekly review, small wrong answers repeat for months.

A go-live checklist

• Top questions mapped and answers approved.
• Assistant limited to approved content, with “I don’t know” behavior.
• Hand-off triggers defined and tested.
• Official WhatsApp Business Platform set up; opt-in rules followed.
• Privacy notice, data minimization and retention defined.
• Conversation log and weekly review owner assigned.
• Metrics baseline recorded before launch.

Done this way, AI does not replace your sales team. It makes sure no lead waits, and that every conversation reaches a person with the information needed to close.