What Is an AI SDR? A Plain-English Guide for Small B2B Teams

What an SDR actually does, before you add “AI” to it

A sales development representative’s job, stripped of title inflation, is three things: work out who to contact at a target company, find a real reason to contact them, and write a first message worth reading. Everything after that (the reply, the call, the deal) is a different job, usually an account executive’s. An AI SDR automates the first three, not the sales conversation itself.

This matters because the three tasks are not equally hard, and they are not equally automatable. Deciding who to contact is largely a filtering problem, and software has been good at filtering for twenty years. Finding a reason is a research problem, which is the part that has only recently become tractable. Writing the message is a composition problem, which language models are genuinely good at and which is also, of the three, the least important, because a well-written email built on nothing still reads as nothing.

Two very different products get called “AI SDR”

The label covers two genuinely different things, and mixing them up is where most disappointment with “AI SDR software” comes from. Someone buys what they thought was an outcome and receives a tool.

Neither shape is wrong. They fit different buyers: the self-serve platform suits a team with the time and appetite to operate a new tool, and a managed service suits a team that wants the output without becoming the operator. The mistake is buying the first while needing the second, which is what happens when both are described using the same two words.

What an AI SDR reliably automates

• Identifying a plausible decision-maker. Matching a target company to the person inside it who owns the problem, which at smaller companies is rarely the person whose title matches.
• Finding and verifying contact details. A business email address and sometimes a phone number, checked against a live source rather than guessed from a pattern. See how verification actually works for why the distinction is expensive.
• Pulling a signal. A hire, a launch, a funding round, a structural change: something public and dated that makes this company worth contacting now rather than in general.
• Drafting the first message from that signal, at a length and in a register a person would actually use.

What almost never should be automated

• Sending without review. Even fully automated tools usually offer a review step, because an unreviewed, wrong, or oddly-worded email sent from a real domain is expensive to undo. The failure is not usually catastrophic; it is a slow accumulation of slightly-wrong emails that quietly teaches a market to ignore your domain.
• The actual sales conversation.Once someone replies, that’s a human conversation. Nothing on the market genuinely closes deals unsupervised, and any claim that it does is worth being sceptical of.
• Guaranteeing outcomes. Replies, meetings, and revenue depend on far more than the first email: your offer, your market, your follow-up, your pricing. Treat any AI SDR claiming guaranteed meetings or guaranteed pipeline as marketing, not a promise.

How to tell if it’s actually researched, or just templated

There is one test, and it takes five seconds. Take the email and replace the company name with a competitor’s. If it still makes sense, it was never about the recipient. The “AI” part was doing formatting, not research.

A genuinely research-based email references something true, current, and specific about that one company. “I saw you’re growing” fails the test, because every company is either growing or would like to appear to be. “I saw you posted for a first RevOps hire last week” passes it, because it could only have been written about one company in one fortnight.

What an AI SDR costs, and what the cost is measured against

Self-serve AI SDR platforms are commonly priced in the $3,000–$5,000 per month range at the level small teams look at, and the seat is only half the cost: someone has to operate it. A full-service outbound agency retainer commonly starts above $10,000 per month, because its cost floor is a person’s salary plus margin. A first SDR hire costs salary plus recruiting plus three to six months of ramp before consistent output.

The comparison people usually get wrong is against zero. The alternative to buying something is not “free”; it is a founder spending evenings on prospecting, or outbound not happening at all. Both have a cost, and the second one is the larger of the two.

Who actually needs one

The clearest fit is a small B2B team with product-market fit and a growth budget but no dedicated SDR function: someone, often the founder, is already doing outbound badly or not at all, and the choice is between hiring, configuring a platform, or handing the research and drafting to a managed service.

The clearest non-fit is a company that has not yet worked out who it sells to. An AI SDR multiplies whatever targeting you give it. If the targeting is a guess, you get a well-researched, beautifully written guess, several hundred times over. Work out the market first. fifty accounts by hand will tell you more than any tool will.