What is an AI SDR?

An AI SDR is software that does the first part of a sales development rep's job. It writes to prospects, follows up, answers replies and books meetings inside channels such as LinkedIn and email. A person sets the offer, takes the call and closes the deal. Results depend more on audience and offer than on the software.

What does an AI SDR actually do?

SDR stands for sales development representative: the person who contacts prospects, gets a first conversation and passes the interested ones to a salesperson. An AI SDR does the same job in software. In practice it covers five tasks:

1. Takes a list or a signal. The people to contact come from a search, an imported list, or an intent signal such as someone engaging with a post about your topic.
2. Writes the first message for each person, usually using a detail from their profile or company.
3. Follows up on a schedule, with a different angle each time, and stops when the person replies or asks to stop.
4. Reads and answers replies, often within minutes, including at night and at weekends. It answers questions, handles objections and asks a qualifying question.
5. Books the meeting by sending a booking link or proposing a time, then hands the conversation to a person.

At Sellinger, LinkedIn and email carry most of the volume. WhatsApp runs through your connected WhatsApp account, not the WhatsApp Business API. First messages go out at a small daily volume, and the agent also answers people who write first. Instagram and live chat are used mainly to answer people who write to you first.

What does an AI SDR not do?

It does not set your offer, negotiate a price, or run the sales call. It does not know facts you have not given it. In our studies the agent's job ended at a booked call, a quote or a qualified question, and a person did the rest:

• At Hotel Pirin, the agent answered guests in a median of 41 seconds and quoted a price, but the owner and his reply desk negotiated every group deal (Hotel Pirin case study, a related party: our own hotel).
• In the installer survey, the agent answered installers in about a minute and a Shelly person took over pricing, support and big projects (Shelly case study).
• For the foundry, the agent handed requests for quotes, drawings and NDAs to the company's president by email (Lamothermic case study).

How is an AI SDR different from a sequencing tool or a human SDR?

The last row matters most. All three need the right people to write to. For a longer side by side, see AI SDR vs human SDR.

Where do AI SDRs fail?

Five patterns come up in our own data and in the studies we have published. Each one is a setup problem more than a software problem.

• The wrong audience. On our own LinkedIn outbound, people who engaged with posts from AI SDR vendors and automation creators booked 2.6% to 3.7% meetings per person messaged. People who engaged with enterprise sales voices booked under 0.4%. The agent was the same (Sellinger case study).
• The wrong offer. An agency added a Connected TV option to its pitch. Replies kept coming, but the 181 conversations where it came up produced no meetings (Uberwood case study).
• A hand off nobody picks up. The agent is fast. A thread that needs a person is only as fast as that person. In the hotel study, staff answered group requests in a median of almost 4 hours.
• Facts the agent does not have. An agent that is allowed to say anything can say something the sender did not approve. Give it exact limits on day one. The foundry in our Lamothermic study told its agent what it casts (steel, stainless and bronze) and what it does not, and the agent could then answer capability questions precisely.
• Counting the wrong thing. A calendar marks a meeting completed when its time slot ends, whether or not anyone joined. Count held meetings, not booked ones. We explain how in Count meetings honestly.

How does one LinkedIn conversation run from first message to meeting?

Here is the path on our own account, from the Sellinger case study. It shows where the agent works and where a person starts.

1. A signal. A person likes or comments on a post about sales automation, AI SDRs or go to market. They are filtered to decision makers in growth and revenue roles.
2. A connection request. On our own account, 41% of 68,818 requests were accepted.
3. An opener on acceptance. One observation about their work. No pitch, no link, no vendor name.
4. Follow ups. About every five days, up to three, with a new angle each time, then a clean exit line.
5. A reply. The median agent reply since May 2026 was 2.3 minutes, and 90% came within an hour. The median wait from first message to first prospect reply was 10 hours.
6. "What is this?" One line that names Sellinger and ties it to the person's own words, then an offer of a short showcase.
7. A booking, and a nudge. The booking link typically went out after three prospect messages. More than half of the bookings confirmed in the chat needed one more nudge after the link.
8. A person takes over. A Sellinger team member runs the 15 minute showcase.

In the study, 80% of conversions came from a reply to the first message. The agent covers steps 3 to 7 of this path, the signal and the list come from the platform, and a person does step 8.

Why are replies not the number to watch?

Replies held at about 40% across the year on our own account, while meetings per person messaged peaked at 2.4% in November 2025 and drifted below 1% from June 2026. Replies did not predict meetings. That is the reason to be careful with any AI SDR demo that leads with a reply rate. Replies are the easy part. We cover why in Replies are easy, conversion is audience and offer.

How should you evaluate an AI SDR?

Ask these six questions before you sign.

• Can I see results with the denominator (people messaged, not just "replies")?
• How does it count a meeting: booked, held, or both?
• Who takes over when someone is interested, and how fast do they know?
• What can I stop it from saying, and how do I check what it said?
• Which channels does it really run, and at what daily volume? Ask what is inbound only.
• What does it do when the person says no, or says nothing?

What does timing look like in practice?

Bookings take time. In the ShortVibe study, a solo founder's median from first touch to booking was 22 days, and 11 of 19 bookings came more than two weeks after the first touch (ShortVibe case study). On our own account the median wait from first message to the prospect's first reply was 10 hours, so a quiet first day says little. Judge an AI SDR over weeks, on a defined audience and offer, not over an afternoon.