All postsArtificial intelligence
WhatsApp AI agent for sales, from first message to closed deal
Vinícius Molina
7 min
A WhatsApp AI agent for sales is a digital rep that talks to leads over the official API: it qualifies, handles objections, books meetings, closes when you allow it, and hands off to your team with full context. At Sailer AI, Crew agents share one inbox with your people, write to your CRM and get every conversation measured.
Most AI tools for WhatsApp answer questions. A sales agent needs more than that: it has to know which stage the lead is in, what they already said, what the company can offer, when to push and when to bring in a person. This article explains that difference: what the agent does, how it is built, how long it takes and what it has already delivered in live operations.
What a WhatsApp AI agent for sales actually does
The agent runs the whole conversation, not just the opener. Every step follows your playbook and has a clear point where someone on your team can take over.
| Step | What the agent does | Where your team comes in |
|---|---|---|
| First response | Replies in seconds, 24/7, including transcribed voice notes | Watches the queue in real time |
| Qualification | Asks your criteria questions and logs the answers in the CRM | Reviews borderline cases |
| Objections | Handles price, timing and competitor objections with approved arguments | Takes objections outside the playbook |
| Booking | Checks real availability and books the meeting or visit in the chat | Gets the meeting already on the calendar |
| Closing | Takes the deal to purchase when your operation allows it | Closes deals that need negotiation |
| Handoff | Passes the conversation with a summary, answers and lead source | Picks up exactly where the agent stopped |
| Follow-up | Re-engages leads who went quiet, with a fresh message each time | Tunes the cadence and stop rules |
Follow-up inside and outside the 24-hour window
On the official API, a business can only send free-form messages within 24 hours of the customer's last message. After that, it needs a Meta-approved template. The agent handles this on its own: instead of scheduling a fixed text, it schedules the intent to follow up. At the set time, it checks whether reaching out still makes sense and picks the right format, free text if the window is open or the right template if it is closed. Our guide to the 24-hour window covers the rule in detail.
This matters for results. At Bitcoin Blackpill, 60% of sales came after a follow-up.
How the agent is built and tested
A sales agent is not a single prompt. It is assembled from four parts, all configurable by your operation:
- Playbook. Conversation stages, qualification criteria, follow-up cadence, tone of voice and what the agent must never do, such as offering an unapproved discount.
- Knowledge base. Products, prices, terms, FAQs and commercial rules the agent checks during the conversation.
- Tools. Actions it takes in other systems: checking calendars, reading and writing to the CRM, checking availability, creating deals.
- Autonomy. How far it goes alone and when it calls a person. You can start alongside the team and loosen the rules later, with no re-implementation.
Before any real lead, the agent goes through about two weeks of sandbox simulation. AI-simulated personas talk to the agent across the happy path, objections and edge cases, and your team only reviews the finished conversations. From kickoff to go-live, a typical implementation takes about five weeks. The method is described in how to test an AI agent before going live.
People and AI in the same shared inbox
The agent does not run in a parallel channel. It works in the same shared team inbox as your people, with queues, routing rules and permissions. That changes three things:
- Handoff is a transfer, not a restart. Whoever takes over sees the history, the summary and what has already been qualified.
- Anyone can step in. Supervisors follow live conversations and take over whenever they want.
- One set of rules for everyone. Routing, SLAs and logging work the same whether the conversation is run by AI or by a person.
For teams where each rep sells from their own number, Sailer Fleet connects those numbers to the official API and adds the agent as a copilot: summaries, reply suggestions, follow-up and an updated CRM. See the full platform.
The CRM fills itself from the conversation
Everything the lead says becomes structured data while the conversation happens. The agent reads contact, account and opportunity context before replying, then creates and updates contacts and deals, moves pipeline stages, writes fields and notes, records the loss reason and logs the conversation as an activity.
It works on top of what you already use (Salesforce, HubSpot, Pipedrive, RD Station) or with Sailer's own CRM. At Asaas, the AI sizes each lead's revenue potential, books the meeting while distributing it across reps, and writes fields and notes in HubSpot, end to end.
Every conversation measured, AI and human
An agent can run without errors and still not convert. To know why, you need to look at every conversation, not a sample. Compass, Sailer's measurement layer, scores 100% of conversations against the playbook, groups the most frequent objections, surfaces loss reasons and generates coaching. It measures human conversations, Sailer AI conversations and third-party AI conversations on the same scale.
The full diagnosis is in why an AI agent doesn't convert.
Results from Sailer AI agents in production
| Client | Operation | Result |
|---|---|---|
| Amor Saúde (500+ clinics, Brazil) | Appointment booking and cross-sell of the Cartão de Todos membership | 24% booking conversion, 28% Cartão de Todos conversion, 66% lower cost per lead served, R$14.50 return per R$1 |
| Bitcoin Blackpill (info product, 18k leads/month) | End-to-end sales with no human team, R$400 ticket | Conversion from 0.8% to 9.4%; 60% of sales after follow-up |
| Asaas (200k+ customers) | Qualification, routing and booking with HubSpot | 50% shorter sales cycle |
The three cases are designed differently. At Amor Saúde, the agent checks real calendar availability and spots patients who don't have the membership yet. At Bitcoin Blackpill, it runs the entire sale. At Asaas, it prepares the lead and hands it to the right rep.
How WhatsApp AI agent pricing works
The price of a WhatsApp AI agent has three parts, and knowing each one makes proposals easier to compare:
- Setup. Playbook design, integrations, knowledge base and the simulation phase. Charged once.
- Monthly fee. The platform, connected numbers, team seats in the inbox and support.
- Usage. Proportional to the leads conversed each month.
Meta messaging is billed separately and passed through at cost, with no markup, because Sailer operates as a Tech Provider. What moves the price most is lead volume, the number of integrations and the number of journeys (pre-sales, outbound, reactivation, post-sale). For a proposal based on your scenario, talk to our team.
Frequently asked questions
How much does a WhatsApp AI agent cost?
It depends on lead volume, integrations and how many journeys the agent covers. The model is a setup fee, a monthly fee and usage per lead conversed; Meta messaging is passed through at cost. Published plans range from R$1,449/month (Kayak) to R$4,947/month (Yacht), and enterprise projects start at R$10k/month.
Do I need the official WhatsApp API?
Yes. Sailer is a Meta Tech Provider and runs only on the official WhatsApp Business Platform, with approved templates and the channel's quality rules. That avoids the ban risk of QR-code automations. With coexistence, a number keeps working in the WhatsApp Business app while it is connected to the API.
Does the AI agent replace my sales team?
That is not the design. The agent takes the repetitive volume: first response, qualification, follow-up and booking. People step in where you decide, such as objections outside the playbook, high-value accounts or a lead asking for a human. In most operations, the same team ends up handling many more leads.
How long does it take to go live?
A typical implementation takes about five weeks from kickoff to go-live. Around two of those weeks are sandbox simulation, covering the happy path, objections and edge cases before any real lead sees the agent. What speeds things up most is having the playbook, CRM access and message templates defined early.
Does the agent work with our existing CRM?
Yes. The agent reads contact, account and opportunity context, creates and updates contacts and deals, moves stages and writes fields and notes in Salesforce, HubSpot, Pipedrive or RD Station. Teams without a CRM can use Sailer's own CRM, which fills itself from the conversations.
Will leads know they are talking to an AI?
The agent writes naturally, uses the conversation history and understands voice notes. Even so, we recommend being transparent when a lead asks. What hurts conversion is not knowing it is an AI; it is getting a generic, slow or wrong answer. That is why every playbook is tested in simulation before going live.
