Should you replace your SDRs with AI?

Probably not, and the companies furthest ahead on AI aren't either. If you've got SDRs who book meetings that turn into pipeline, the better move is to give them AI for the research and admin, then point their time at the parts of outbound a buyer can tell a human did.

That's my take. It also lines up with something published last week that's worth a read if you run a sales team.

What the research found

Madrona, the Seattle venture firm, published a conversation with Sam Nelson, founder of SDRLeader.com, on what's happening to sales development teams. His view is that the companies closest to the frontier of AI (the model labs and the infrastructure companies around them) are adding SDRs faster than any other go-to-market role. He puts the headlines about SDR layoffs down to companies that are struggling to win new business, rather than companies that have automated the job away.

He also makes a historical point. Sales engagement platforms automated most of an SDR's manual tasks years ago, and SDR teams grew a lot afterwards. Automation made each SDR more productive, so companies hired more of them.

The same week, Salesloft announced a set of agentic features, including a meeting agent that writes pre-call briefs and live battlecards inside Zoom. Tucked into the release is a figure from its own survey of 500 US sales leaders: every one of them uses AI somewhere in the revenue process, but only 20.6% have it in production with outcomes they can measure. So almost everyone is trying AI in sales, and about one in five can show you what it's done.

Why outbound still needs a person

The argument for replacing SDRs goes like this: an AI agent can research an account, write a personalized email and send thousands of them a day. All true. The problem is that every one of your competitors can do the same thing, and your buyer's inbox knows it.

Nelson's point is that AI has made the personalized cold email ordinary. A line about the prospect's latest LinkedIn post used to show effort. Now it shows a prompt. What still works is proof that a person is behind the message: a reference to a conversation you've had, a named colleague who suggested the introduction, a phone call. As he puts it, "It's very easy to ignore an email." It's harder to hang up on someone who's been polite and done their homework.

This is the same reason I think AI should take the busywork so sellers can spend more time being human with buyers. Used to send more volume, it gets you more of the thing buyers already delete.

There's a practical risk too. The Madrona piece warns that many autonomous outbound tools send through your own email domain. Push enough low-quality volume through it and your messages start landing in spam, including the ones your account executives send to live deals. Teams then blame the copy or the targeting when the domain is the problem. That's an expensive way to learn about email deliverability.

Where AI should go in your sales development team

I built my own AI outbound engine on LLM APIs and MCP, so I'm biased towards using AI here. I just put it in different places than the "AI SDR" pitch does. In order, this is what I'd hand over first:

1. Inbound speed to lead. Nelson expects AI agents to own this, and I agree. When someone fills in your demo form at 9pm, an agent that answers in seconds, qualifies and books a time beats a human who sees it tomorrow.
2. Account research. Pulling together what the company does, who's been hired, what they've said publicly and what they use today. This is the bit that used to eat an SDR's morning.
3. First drafts, edited by a person. The SDR decides who to contact and why, and the draft gets them most of the way. They add the bit only they can know.
4. CRM updates and call notes. Logged automatically, so your pipeline data is complete without anyone nagging. I wrote about why that frees reps up in Sales admin is being automated. What are reps for now?
5. Meeting prep. A one-page brief before every first call. Salesloft is building this into its product.

What stays with the person: choosing the accounts, the phone, the follow-up after a conversation, and the judgment about whether a meeting is worth an account executive's time.

Measure the meeting, not the activity

If you're going to change how your SDRs work, change how you measure them first. Nelson suggests one goal per SDR: a meeting between an account executive and the right buyer. Review it weekly or monthly, and if those meetings don't convert, tighten the definition of "right buyer". That's also a useful signal that your ideal customer profile needs work.

His other numbers give you a rough benchmark. In SDRLeader.com's research, about 80% of booked meetings actually happen, and about 80% of those get accepted by account executives as genuine opportunities. He also notes that a held rate close to 100% can mean reps are only working the easy leads.

Counting emails sent was never a great measure, and AI makes it useless. An agent can send 10,000 emails before lunch. I made the longer case for this in Why more activity won't fix your pipeline.

One more warning from the piece: connect your data before you use AI to judge rep productivity. If email isn't syncing to the CRM, a hardworking rep can look idle on a dashboard.

What I'd do this quarter

If you're a founder or sales leader deciding whether to cut SDR headcount for an AI tool, run through this first:

• Check your meeting-to-opportunity rate. If it's healthy, your SDRs are doing their job, and the question is how to give them more hours for it.
• Move inbound response to an agent. It's the clearest win and the lowest risk to your brand.
• Give every SDR AI research and drafting, and keep a human choosing and sending.
• Protect your main domain. If you do test an autonomous outbound tool, run it on a separate sending domain with tight volume limits, and watch bounce and spam rates weekly.
• Measure one thing: qualified meetings held with the right buyer.

If you're early and the founder is still doing the selling, you may not need an SDR yet. AI research and drafting can cover a lot of what a first SDR would do, and buyers often want to hear from the founder anyway. When you do hire, hire someone who's curious about people and comfortable on the phone. That's harder to automate than the email.

We help teams sort out which parts of the sales process to hand to AI and which to keep human. That's most of what our AI for revenue teams work covers.

And if an AI vendor tells you it'll replace your whole SDR team by spring, ask them how many SDRs they've hired this year.