From DM to booked call: an AI sales workflow

Published 17 September 2026 · Replio

A sales DM is not a lead yet, and a booked call is not a sale yet. Between the two sit five small steps, and each one quietly loses people. This is the workflow, stage by stage: where the drop-off happens, which parts an AI agent should own, and where the time-picking back-and-forth kills deals that were already won.

A booked call is five small yeses

Someone messages "hi, how does this work?" For that to end with a person on a call with your salesperson, five things have to happen in order:

1. They get a reply while they still care.
2. The reply answers the question they actually asked.
3. They are worth a call, and you know it before anyone books a slot.
4. They agree on a time that is really free.
5. They turn up.

The trouble is that these losses multiply. If each step keeps four people in five, only about one in three who messaged ever reaches the call (0.8 × 0.8 × 0.8 × 0.8 × 0.8 ≈ 0.33). Fixing one step helps a little. Fixing all five changes the pipeline. Most teams only work on the first one, faster replies, and never look at steps four and five, which are the ones that lose leads who had already said yes.

Steps one to three: answer, then qualify

The first three steps are covered in detail in our guides on the cost of slow replies and AI lead capture and qualification on WhatsApp, so here is the short version. The AI replies in seconds, at any hour, in the lead's language, from your own knowledge base. Then it trades: it answers their question and asks one of its own. What do they need, how big is it, how soon, and where are they? Four answers are usually enough to tell a real buyer from someone browsing.

What matters for this workflow is what qualification decides. A lead who fits gets offered a call. A lead who does not still gets a complete answer, and maybe a link to buy or a price list, but not a slot in your salesperson's day. Offering everyone a call is how a calendar fills with conversations that were never going to close.

Step four: the time-picking loop

This is the step nobody designs, and it is where warm leads go cold. A qualified lead says "sure, let's talk", and then this happens:

"What time suits you?" · "Maybe Thursday?" · "Thursday we have 11 or 3." · "3 is hard, what about 4?" · (no reply until the next day)

Every round of that exchange is another chance for the lead to get distracted, and when a human is running it, each round waits for the human to be free. The intent was already there. The scheduling lost it.

There are three ways to close the loop, and they are not equally good:

• Keep negotiating by hand. It works when volume is tiny and someone is always holding the phone. It fails at night, at weekends, and whenever two people are chasing the same slot.
• Send a booking link. This is much better. The lead picks a slot, and it cannot clash. But it asks someone who is mid-conversation to leave the chat, open a page and start again, and some of them never tap it.
• Book it inside the chat. The AI asks which call they want and when suits them, checks the host's availability, collects a name and contact, reads the booking back and logs it. The lead never leaves the thread. If they ask a question halfway through ("is the call on video?", "how long is it?"), the AI answers from your knowledge base and carries on.

The strongest setup offers the conversation first and keeps the link as a fallback for the people who would rather choose from a grid.

Make the calendar tell the truth

In-chat booking is only as good as the availability behind it. A few settings do most of the work:

• One event type per kind of call, such as a 15-minute intro or a 45-minute demo, each with its own duration and location (in person, phone or a video link), so the AI books the right length.
• Round-robin team routing when any of several salespeople can take the call, so leads are shared out instead of piling onto whoever set up the calendar.
• Buffers, a minimum-notice window and a booking horizon, so nobody is booked five minutes from now or back-to-back with no time to prepare.

One honest limit: Replio does not yet sync live with Google Calendar or Outlook, so each host sets their weekly availability by hand. If a host's week changes, update it, or the AI will keep offering slots that are no longer free.

Step five: booked is not attended

A booking on the calendar feels like a win, but the sale needs the call to happen. The things that get people to turn up are dull and they work:

• A confirmation read back in the chat, so a wrong date gets caught now instead of at the call.
• A calendar invite (.ics) the guest can add to their own calendar.
• A reminder before the call.
• A self-serve cancel and reschedule link. This one is underrated. Someone who hits a clash and can move the call in two taps stays in your pipeline. Someone who has to message and wait usually just doesn't show up.
• Marking each appointment completed or no-show, so you are measuring attendance and not guessing at it.

For the no-shows, follow up in the same thread. If the lead wrote to you in the last 24 hours, WhatsApp lets you reply freely. After that, Meta requires an approved template, which a WhatsApp broadcast sends to opted-in contacts, with opt-outs handled automatically.

Three ways to run it

What the caller should already know

A call that opens with "so, tell me what you're looking for" wastes the first ten minutes on questions the chat already answered. The salesperson should walk in knowing what the lead asked, what the AI told them, the qualifying answers, and which channel they came from. In Replio that is the same conversation in the shared live inbox, with the answers recorded against the contact by the light CRM. A person can take over the chat at any point: before the call to handle a custom quote, or after it to send the proposal in the thread the lead already uses.

Measure each step separately

Do not track just "leads" and "sales". Count each step, so you know which one is leaking:

Tag each entry point with its own tracked short link or QR code and you can compare sources all the way down to attended calls, not just clicks.

Setting it up in Replio

The pieces all live in one account and share one AI brain:

• Lead capture and AI qualification: every enquiry logged, with the qualifying answers attached.
• Booking inside the chat on WhatsApp, Instagram, Messenger and Telegram, plus the website widget.
• A free booking calendar with event types, round-robin routing, buffers, reminders, a reschedule link and a shareable replio.live/book/ page, free on every plan.
• Comment-to-DM, so a keyword comment on a post opens the same workflow in a private DM.
• A shared live inbox where your team can take over any conversation.

The setup order that works: write the knowledge base, decide the four qualifying questions and your rule for who gets a call, create one event type per kind of call with real availability, connect your channels, then book a test call yourself from your own phone before you point an ad at it.

Frequently asked questions

Related