What is an AI SDR? What it does, what it doesn't, and where your reps still win

An AI SDR is an artificial intelligence agent that does pre-sales work on messaging channels: it replies to every inbound lead immediately, at any hour, asks your qualifying questions, logs the answers in your CRM and books a meeting with a human seller, who receives the conversation with full context. It does not close deals and it does not replace your reps. Its job is to make sure no lead waits, every conversation arrives qualified, and sellers spend their time where only they can make a difference.

The rest of this piece covers how that works in practice, especially on WhatsApp, and what you need to decide before you switch one on.

What an AI SDR is — and what it isn’t

The name helps and hurts. It helps because it describes the function: an SDR (sales development representative) is the person who picks up the lead, works out whether there is a fit and books the conversation with the person who sells. It hurts because it suggests an “AI employee” with the same responsibilities as a human. There isn’t one.

An AI SDR is:

• An agent that speaks on behalf of the company, in text, on the channels you connect
• Something that always answers from what the company taught it: products, pricing, policies, common objections
• A qualification process that runs the same way in every conversation, at 2 p.m. on a Tuesday or 11 p.m. on a Sunday

An AI SDR is not:

• A seller. Negotiating, tailoring a proposal and closing are done by a person
• A cold outreach blaster. The starting point is a lead who came to you, not a purchased list
• A black box. Every conversation is recorded and can be read, corrected and taken over by someone on the team

This distinction isn’t just vocabulary. It defines where the AI helps and where it starts to cost you.

What it does on WhatsApp and Instagram

In Brazil, where Contorna is built, B2B leads rarely fill out a form and wait for an email. They message. They tap an ad, land in WhatsApp, type “hi, how much is it?” and get on with their day. If nobody answers soon, they send the same question to two other vendors. The same pattern is spreading wherever messaging apps have become the default way to reach a business.

That is where an AI SDR works. A typical conversation has four moves:

1. Instant response. The buyer gets a useful answer in seconds, not “we’ll get back to you.”
2. Qualification. The agent asks the questions your company considers decisive — team size, timing, who decides, what they’ve already tried — without turning the chat into an interrogation.
3. Logging. Answers become data on the CRM contact, not a screenshot buried in someone’s phone.
4. Booking and handoff. When the lead is ready, the agent books the meeting on the seller’s calendar and hands over the history: what was said, what was asked, what is still open.

There’s a fifth move most teams forget: reviving the conversation that went quiet. A lead who stopped replying mid-qualification isn’t lost. They need a good second touch at the right moment.

Here is roughly how it sounds:

Buyer: Hi, saw your ad. Do you work with small companies?

Agent: We do. So I can point you the right way: how many people on your team talk to customers today?

Buyer: About 6.

Agent: Got it. Is the bigger bottleneck lead volume, or converting the people who already reach out?

Buyer: Converting. Lots of people ask for the price and disappear.

Agent: That’s exactly the kind of case our team looks at in a 20-minute call. I have Thursday at 10 or Friday at 3. Which works better?

Nothing clever. Just fast, polite and pointed in the right direction — which, at 10 p.m. on a Thursday, is already more than most companies deliver.

AI SDR vs. chatbot vs. human SDR: what’s the difference?

The most common confusion is between an AI SDR and a chatbot. The difference shows up the first time a buyer leaves the script.

	Menu chatbot	AI SDR	Human SDR
How it talks	Buttons and a fixed tree	Natural language, grounded in company knowledge	Natural language, with personal judgment
When a question is off-script	Gets stuck or repeats the menu	Answers if it’s in the knowledge base; if not, doesn’t make it up and calls a person	Improvises (well or badly)
Hours	24/7	24/7	Business hours, give or take
Qualification	Collects fields	Asks questions and interprets answers	Asks and interprets, varies by person
CRM logging	Depends on integration	Automatic, on the contact	Depends on discipline
Scale	High	High	Limited by headcount
Negotiation and close	No	No	Hands off to a seller

The last row is the one that matters. None of the three closes a complex deal alone. The question isn’t “AI or people,” it’s who does which part.

Where human sellers stay essential

An AI SDR takes the repetitive, time-sensitive work off your reps. It doesn’t take the hard work. These stay with people:

• The meeting. Real discovery, reading what wasn’t said, the right question at the right moment
• The negotiation. Commercial terms, exceptions, discounts, timelines — anything that needs authority and judgment
• The decision-maker who wasn’t there. Winning over the partner who skipped the call is human work
• Sensitive cases. Complaints, strategic accounts, anything off-pattern

One operating rule prevents most problems: when someone on the team replies in the conversation, the AI stops. Handing the conversation back to the agent is an explicit action, never something that happens on its own. That way there are never two voices speaking for the company at once.

How to set up an AI SDR without losing control

The agent is only as good as what you teach it. Four decisions do most of the work.

1. Knowledge

Products, what each one solves, who it’s not for, which prices can be quoted, policies, the objections that come up every week and how your team answers them. If your best rep answers something from memory, it needs to be written down. What isn’t in the knowledge base, the agent shouldn’t claim.

2. Qualifying questions

Use the criteria your team already trusts. Public frameworks like BANT (budget, authority, need, timeline) or GPCT are good starting points, but the final list is yours. Three to five questions is enough; more than that is a form in disguise. For more on choosing criteria, see our piece on lead qualification.

3. Booking rules

Who gets which lead, during which hours, with how much notice, and what happens to leads that don’t qualify. A poor-fit lead deserves a polite exit, not a meeting nobody wants.

4. Automation mode

This is the most important control and the most overlooked. There are three levels:

• Notify only: the AI reads the conversation and alerts the team, but writes nothing
• Draft for approval: the AI writes, a person approves or edits before it’s sent
• Send on its own: the AI replies and leads the conversation

Start with drafts. Read the replies for a week or two, fix the knowledge base wherever it got things wrong, and only then turn on automatic sending — ideally after hours first, where the upside is biggest and the risk of colliding with your team is smallest.

Risks and guardrails to define up front

A poorly configured AI SDR makes mistakes quickly and at scale. The risks are predictable, so you can fence each one in:

• Promising what doesn’t exist. Define what the agent may never state: discounts, implementation timelines, integrations you don’t have. No source, no claim.
• Pushing too hard. Cap how many follow-ups a conversation gets and how far apart. An annoyed lead becomes a complaint.
• Conversations that should be human. Words like “cancel,” “complaint” or a strategic account’s name should call a person in.
• Personal data. Conversations are your customers’ data. Check how the vendor handles privacy law (in Brazil, LGPD) and whether conversations are used to train public models.
• Nobody watching. The agent doesn’t correct itself. Someone has to read samples every week.

How sales leaders should measure it

Before you look at numbers, look at conversations. In the first phase, the right questions are qualitative:

• Are after-hours leads getting answered, and answered well?
• Do meetings arrive with enough context that the seller doesn’t repeat questions?
• Do reps trust what’s in the CRM, or do they still double-check everything in WhatsApp?
• Are quiet conversations being revived, and is the tone of that follow-up right?
• How often did someone have to step in this week, and why?

Then compare against your own history: time to first response, share of leads that become meetings, and how much of a seller’s day still goes to triage. Response time alone explains a lot — our piece on speed to lead shows how to measure yours.

Where Contorna fits

Live Coach is Contorna’s AI sales copilot. It follows sales conversations and suggests responses to help sellers handle objections. Roleplay lets your team practice with a simulated buyer before the next conversation.