Should AI talk to your inbound leads?

Somebody just filled in your contact form. They spent about eleven minutes on the site first, read two pages properly, and by the time they typed their email address they had already half decided to buy. That's the most valuable thirty seconds you'll get from them all year.

The question is what meets them next. In a lot of B2B businesses built since 2024 the answer is an AI agent, because an AI agent replies instantly and a human replies on Monday, and instant obviously beats Monday.

Instant does beat Monday. That isn't the argument. The argument is that the two things AI is being asked to do here are not the same job, and one of them it does well enough to make you money while the other one costs you the deal.

The short version

AI is allowed to acknowledge, route and book. It replies in under a minute, by name, referencing the specific thing the person asked about, offers two times this week and puts one in the diary. That's admin. Nobody has ever resented a fast, accurate confirmation, because nothing is being sold to them.

AI is not allowed to qualify, handle objections, discuss price, or run the conversation where the buying decision actually gets made. The moment it starts asking about budget and timeline it stops being a receptionist and becomes a gate between a warm buyer and your sales team.

The dividing line isn't the technology. It's the intent of the person on the other end.

Are AI SDRs worth it?

For most founder-led B2B businesses: yes on cold lists and dormant databases, no on live inbound.

The tools genuinely work at the job of contacting people who are not expecting to be contacted, at a volume no human would sustain, for a fraction of a salary. Pointed at two thousand old enquiries, an AI sequencing tool will produce booked calls out of an asset you already own. That is a real return and we turn it on early in most engagements.

Pointed at the person who filled in your form eleven minutes ago, the same tool is spending your most expensive asset on your cheapest task. You paid to get that person to the site. The marginal cost of a human replying to them is a few minutes of somebody's afternoon. Automating that particular few minutes is the worst trade in the stack, and it's the one the category is sold on.

The vendor case rests almost entirely on speed, and the speed argument is correct. It just doesn't require the bot to hold the conversation. An automation that acknowledges in forty seconds and books a slot captures the whole speed benefit without any of the conversational risk.

What people mean by "AI SDR"

The category has got sloppy, and it matters, because these tools carry very different amounts of risk to your pipeline.

Most of the disappointment we see comes from buying row one and expecting the results of row three. They are not the same product and they don't meet the same person.

The evidence, and how it gets mangled

Two bodies of research get quoted at each other in this argument, usually by people selling one of the two answers.

On speed, the anchor is still Oldroyd, McElheran and Elkington in the March 2011 Harvard Business Review. They audited 2,241 US companies by submitting a web form to each and timing the reply. Twenty-three per cent never responded at all. Among those that did respond within thirty days, the average took 42 hours. In a companion dataset of roughly 1.25 million leads, firms that made contact inside the first hour were around seven times more likely to qualify the lead than firms that got to it in the second hour.

That study is fifteen years old, it's mostly American, and the exact multiples have been inflated, rounded and reattributed so many times that the version you see on LinkedIn is usually somebody's guess dressed as a finding. The direction has never been seriously disputed, though, and the direction is the whole point: fast beats slow by a lot, and the drop-off happens in minutes rather than days.

On buyers and AI, the useful recent number is that they're using it. TrustRadius surveyed 1,862 technology buyers in January 2026 and found 63% had used AI somewhere in their purchase process, while 94% of those said they fact-check what it tells them at least some of the time. Gartner's long-running finding that B2B buyers spend only about 17% of their total buying time with all potential suppliers combined points the same way: the research is self-directed, and the moments of actual contact are rare and load-bearing.

Buyers are perfectly happy to use AI as a research tool on their side of the table. That's not the same as being happy to negotiate with yours.

There's a finding that cuts against us and it deserves saying out loud. Gartner reported in March 2026 that 67% of B2B buyers would prefer a rep-free buying experience, up from 61% the year before. Read carelessly, that's a mandate for the bot. Read properly, it's a mandate for better self-serve: published pricing, a real product page, a calendar they can book without asking permission. A buyer who wants no rep does not want a synthetic one. Give them the booking link and get out of the way, or give them a human. The widget is the worst of both.

Be sceptical of the survey genre that reports 85% of customers preferring a human. Those studies are overwhelmingly commissioned by companies that sell human answering services, and the question wording does a lot of work. You don't need them. The argument stands on the mechanics.

The mechanics: why a bot loses a warm lead

Think about the state the person is in. They've known for about three weeks that something is wrong with the pipeline. They've had a bad conversation with their business partner about it. On a Tuesday afternoon between meetings they finally do something about it, land on your site, believe you enough to make contact, and are already braced for the whole thing to be a waste of their time.

That state is fragile in a specific way. They are looking for a reason to conclude this is like the last three suppliers. A bot asking for their budget range hands them one, instantly, at no cost to them, and the tab closes.

The second failure is subtler and costs more. A human on that call would have heard the thing they said in passing about the sales manager leaving in June, and would have changed the entire shape of the proposal because of it. Nothing in your CRM captures that the deal died there, so it never gets fixed. You lose the deal and the lesson at once.