WhatsApp + lead qualification

WhatsApp AI lead qualification: every chat answered, scored and sent to your CRM

WhatsApp AI lead qualification is an agent that answers every enquiry on your business number within seconds, hundreds at the same time if needed, asks the qualifying questions your team defines, scores the lead and sends it with the full transcript into your CRM. It is built for businesses that get more WhatsApp enquiries than their team can answer by hand: academies with enrolment questions, clinics, service companies, B2B sellers. The agent does not guess. It answers from your own course list, prices and policies, follows rules you write (course, start date, budget, location), and hands ready buyers to a person with a summary. I build these as an independent AI systems engineer (Sr DevOps, ten years of production infrastructure) on Meta's official WhatsApp Cloud API, and I run one in production today. You own the code and every account. Three-week sprint, a single fixed price agreed up front, and a 60-day operational guarantee.

Can an AI handle hundreds of WhatsApp conversations at the same time?

Yes, and technically it is the easy part. Each incoming message arrives at a webhook as an independent event, and the agent processes every conversation separately, with its own memory tied to that phone number. Fifty people writing at 9 p.m. after an ad goes live are fifty parallel requests to a model API, not fifty chats waiting in line for one person. Every lead gets a first answer in seconds, including nights and weekends, when most enquiries to small businesses sit unanswered until the next working day.

The real limits are elsewhere. One is the messaging limit of your WhatsApp account tier and the rate limit of the model API, both configured and monitored with alerts. The other is your team's capacity to take the leads the agent passes on, and that is why qualification matters. An agent that answers everyone instantly but forwards every chat to a person has only moved the bottleneck. The goal is that your team receives only the conversations worth their time, already summarized.

Each conversation processed independently, with its own memory
First reply in seconds, at any hour
No queue: spikes after an ad or a launch are absorbed
Channel and model rate limits monitored with alerts
Your team receives only the leads that pass your rules

What does qualifying a WhatsApp lead with AI actually mean?

Qualification is a short conversation that collects the facts your sales team needs to decide whether to call, plus a rule that turns those facts into a decision. You write both. For an academy the questions might be which course, the preferred start date, the schedule (day, evening, online), budget or financing, and location. For a service company it could be the type of job, the address, the timeline and whether the person decides. The agent asks naturally, in whatever order the conversation allows, and never asks again for something the lead already said in the first message.

The scoring is plain rules you can read, not a black box. A lead who wants a course starting next month, in a city you serve, with a budget inside your range is hot and goes to an advisor now. A lead asking for something you do not offer gets a polite answer and an alternative. Someone who is just browsing gets the information and a follow-up later. The model is good at extracting answers from free text; the decision about what counts as qualified stays in rules your team owns and can change.

Questions and order written by your team, in your tone
Answers extracted from free text, typos and transcribed voice notes
Scoring rules you can read and edit: hot, warm, not a fit
Hot leads handed to a person immediately, with a summary
Not-a-fit leads answered politely, never ignored

Can the agent send qualified leads into my CRM and set up follow-ups?

Yes. When a lead qualifies, or at any stage you choose, the agent creates or updates the contact in your CRM through its API: HubSpot, Pipedrive or a similar CRM, with the answers mapped to your fields, the score, the source channel, and the transcript attached as a note. If you do not have a CRM, the build can include a simple pipeline in your own database, with stages your team moves leads through. Either way, nobody retypes a conversation into a form afterwards.

Follow-ups follow WhatsApp's rules. While the customer service window is open (24 hours after the lead's last message, at the time of writing) the agent can reply freely. After that it can only write with a pre-approved message template, and only to people who opted in. So the follow-up sequence is a set of templates you approve, such as a reminder before the start date or a nudge after a few days of silence, scheduled from the pipeline stage. When someone writes "stop" or asks not to be contacted, the sequence ends and the contact is marked, automatically.

Contact created or updated in your CRM via API, fields mapped
Transcript and summary attached to the lead record
Built-in pipeline in your own database if you have no CRM
Follow-ups as approved templates, timed from the pipeline stage
Opt-out stops every sequence and is recorded

What does this look like for an academy with hundreds of enrolment enquiries?

A training center or academy is the clearest case. Enquiries arrive in waves: after an ad, before an intake date, on Sunday night. Most ask the same things: price, schedule, start date, requirements, whether there is an online option, whether they can pay in instalments. Staff answer the same paragraph dozens of times a day, and serious applicants wait in the same line as the curious.

With an agent, the course catalog, fees, calendar and requirements live in a knowledge base it answers from. It asks the three or four questions that separate an applicant from a browser, records the answers in your admissions CRM or sheet, and sends the person who wants to enrol in the next intake straight to an advisor with the conversation summarized. Everyone else gets accurate information and a follow-up template before the intake closes. Advisors spend their day on people about to enrol, not on repeating the fee schedule.

Answers on fees, dates, schedules and requirements from your catalog
Qualifies by course, intake, modality and financing
Ready applicants routed to an advisor with a summary
Reminders before the intake closes, as approved templates

Can one AI inbox handle WhatsApp, Instagram and Messenger together?

Yes. Instagram Direct and Messenger are reached through Meta's APIs too, so the same agent, knowledge base and qualification rules can serve all three channels. Each message is tagged with its source, the contact is matched or created once, and your team sees every conversation in one place instead of three apps on one phone. A lead who first asked on Instagram and later wrote on WhatsApp stays one contact with one history, as long as there is something to match on, like a phone number or email they shared.

This is how I run my own operation. Inbound messages from WhatsApp, Instagram and Messenger go to an agent that drafts the reply from the knowledge base, a person approves or edits it, and then it sends. Contacts, pipeline stages and appointments live in one place. Whether replies go out automatically or through approval is a decision per channel and per topic. Most businesses start with approval and loosen it as they come to trust the answers.

Should I buy a WhatsApp CRM with built-in AI or build a custom agent on my own CRM?

There are WhatsApp CRM and shared-inbox products that include an AI assistant and a pipeline out of the box. If your process fits their model, your team is happy to work inside that tool, and you do not already depend on another CRM, a subscription is often the faster and cheaper choice. You configure it yourself and the vendor maintains it. I will say so on the first call; I do not resell tools and earn nothing from recommending one.

A custom agent makes sense when the CRM you already use is where sales actually happen and you do not want a second system with its own copy of your contacts, when qualification depends on data in your own systems (intake capacity, territories, pricing exceptions), or when you need control over the model, the prompts, the logs and where the data is stored. The trade-off is real. Custom gives you exact fit and ownership, and you take on a build and someone to maintain it. Off the shelf gives you speed and a vendor's roadmap, and you accept its limits.

Off the shelf: fast, configured by you, bound by the vendor's limits
Custom: fits your CRM and rules, you own code and data, needs a build
Hybrid: keep the inbox you pay for and add the integration it lacks

What does the build look like, from WhatsApp message to CRM?

A message arrives at a webhook, an HTTPS endpoint Meta calls when someone writes to your number, Instagram account or Page. I build on Meta's official Cloud API, and I run an agent on it in production for a small business in Chile in coexistence mode: the business kept its existing number and still uses the WhatsApp Business app on the phone while the agent works through the API. The webhook verifies the signature, deduplicates and hands the message to the runtime, an n8n workflow or a small Python service. The runtime loads that contact's conversation, retrieves the relevant parts of your catalog and policies from Postgres with pgvector, and calls the model (Claude or OpenAI) with your qualification script and rules.

The model returns a reply, extracted answers or an escalation. Extracted answers update the lead record and the score. Routine replies go out directly; replies that commit you to something (a price exception, a scholarship, a reserved seat) wait in an approval queue where a person approves or edits them in one tap. Hot leads trigger a notification to an advisor in Slack or WhatsApp. Everything is written to an audit log in your database. Handoff includes documentation, the code in your repository and Loom walkthroughs.

Trigger: an inbound message on WhatsApp, Instagram or Messenger
Data: your catalog and policies in Postgres with pgvector, plus your CRM
Model: Claude or OpenAI, prompts and rules versioned in your repo
Guardrails: approval queue, escalation, opt-out handling, audit log
Runs in: your Meta account and your hosting (Railway or your cloud)
You own: code, workflows, prompts, data and every account

How much does WhatsApp AI lead qualification cost, and when do you not need me?

The reference build is the AI Intake Agent with knowledge base, a productized scope quoted at a fixed price, with an optional monthly operation for iteration after the guarantee period. That covers the scope on this page: the channel set up, the agent answering from your documents, qualification with your rules, the CRM integration and follow-up templates, escalation and the audit log. Larger scopes, for example several channels with approval queues and a pipeline built from scratch, are quoted as a single fixed price after a discovery call, with the AI Operating System as the upper reference. Meta messaging fees, model API usage and hosting are billed to you directly, with a cap declared in the proposal.

You do not need a custom agent if you get a handful of enquiries a day: quick replies in the WhatsApp Business app are free and enough. You do not need one if a WhatsApp CRM with built-in AI fits your process. And I do not build cold outreach on WhatsApp to bought lists, which breaks Meta's policies and gets numbers banned. If your team spends hours answering the same questions and good leads wait behind curious ones, the 15-minute call is where we check it with your real numbers.

Frequently asked questions

How many WhatsApp conversations can an AI agent handle at once?

Hundreds, because each conversation is processed independently and in parallel, not in a queue. The practical limits are the messaging tier of your WhatsApp account, the rate limit of the model API and your hosting, all of which are sized and monitored before launch. What usually runs out first is your team's capacity to call the qualified leads, which is why the agent only passes on the ones that meet your rules.

Which CRMs can the agent send leads to?

Any CRM with an API: HubSpot, Pipedrive and similar tools, or a Google Sheet if that is what your admissions or sales team uses today. The agent maps the answers to your fields, attaches the transcript and summary, and sets the stage. If you have no CRM, the build can include a simple pipeline in your own database.

Can I keep my current WhatsApp number and the phone app?

Usually yes. Meta's coexistence mode lets a number registered in the WhatsApp Business app also work through the Cloud API, so your team keeps the app on the phone while the agent answers through the API. I run an agent this way in production. Availability and steps depend on Meta's current rules, which I check with you in the first week.

Who decides whether a lead is qualified, the AI or my team?

Your team. The model extracts the answers from free text, but the rules that turn those answers into hot, warm or not a fit are written by you and readable by anyone. Changing a threshold, adding a question or opening a new location is an edit to those rules, not a retraining.

Can the agent follow up with leads who stop replying?

Yes, within WhatsApp's rules. Inside the 24-hour customer service window it replies freely. After that it sends only pre-approved templates, and only to people who opted in. You approve the sequence and its timing. Any opt-out ends every follow-up for that contact and is recorded, so nobody is messaged again by mistake.

What are the limits of WhatsApp AI lead qualification?

It does not close the sale; it gets the right people to your advisors faster. It can misread an ambiguous answer, which is why scores are rules you can audit and doubtful cases go to a person. It does not work on a personal WhatsApp account. And I do not build cold messaging to bought lists, or replace a WhatsApp CRM that already fits your process.

Fifteen minutes on the leads waiting in your WhatsApp.

A free call. I'll tell you straight which processes today's AI can solve and what your infrastructure needs for them to actually run. If it fits, we move forward. If not, I point you the right way, free.