10 Lessons From the First Wave of AI SDRs — and Why Most Fail

SaaStr has now deployed over 20 AI agents, sent more than 60,000 AI-generated sales emails, and generates over half their pipeline from AI sales agents. Jason Lemkin published what he'd tell himself before starting, and the first line is the one most teams won't want to hear:

The tool is maybe 20% of the outcome. The other 80% is how you train it, who owns it, how you segment, and whether you're willing to put in the ramp time.

That ratio matters. Because most teams shopping for AI SDRs are spending 90% of their time evaluating tools and 10% planning how they'll actually use one. The ratio should be inverted.

Here are 10 lessons from SaaStr's experience, and what they mean for a team with 5-10 reps instead of a media company with 163K YouTube subscribers.

---

1. Clone your best rep's playbook before you touch AI

SaaStr cloned their top SDR's actual sequences — the ones producing 12%+ positive response rates on warm audiences. Not the playbook document. Not the manager's version of what reps should do. What the best rep actually does when hitting quota.

The distinction matters for smaller teams too. Your founder who closes 3 deals a month probably has habits they've never written down. The way they research a company. The subject lines they default to. The follow-up timing that works. Get those on paper first.

If humans haven't proven something works, AI won't discover it. AI scales what exists. It doesn't invent what should exist.

2. Start with the work nobody wants to do

For six years, SaaStr tried to get human SDRs to re-engage past event attendees about tickets. Incentives, processes, monitoring — nothing worked. Humans thought it was beneath them.

They gave that segment to an AI agent. It now generates 15% of their event ticket revenue. Six years of leaving money on the table.

Every sales team has a version of this. Churned accounts nobody follows up with. Inbound leads that arrive at 11 PM. Small accounts that aren't worth a rep's time individually but add up. None of it is glamorous. All of it drives revenue when done consistently.

The question for your team: what are the things you know you should be doing but consistently don't? That's your first deployment.

3. Keep campaigns to 800-1,000 contacts

One of SaaStr's clearest findings: small, tight campaigns outperform large blasts. They cap campaigns at 800-1,000 contacts. Their best response rates — 11-12% — came from targeting recent event attendees with specific context. When they broadened targeting, response rates dropped by half.

Three reasons this matters even more for small teams:

• You can't handle the replies. If you send 10,000 emails and 400 people respond, your 5-person team drowns. Response quality drops. Leads go cold while waiting.
• Personalization breaks at scale. At 500 contacts you can be specific about why you're reaching out. At 5,000 you're sending generic messages with a name merged in.
• Smaller segments teach you faster. When a 500-person campaign gets 12% response rate and another gets 3%, you learn something actionable. When 10,000 contacts return a blended 6%, you learn nothing.

4. "Pretty good" at scale beats brilliant and inconsistent

SaaStr's AI SDR sends 3,221 emails per month from a single platform. Their human SDRs used to send 75-285 each. That's an 11-40x increase in volume at consistent quality.

The quality isn't exceptional. It's not the single best email your top rep ever wrote. It's reliably solid — every time, at scale.

This is the reframe most teams need. Stop trying to make AI produce your best rep's best day. Train it to produce your average rep's average day. Then scale that.

For a 5-person team, the math is straightforward. If each rep sends 150 emails a month, that's 750 total. An AI agent doing 3,000 at the same quality level is 4x the team's output from a single tool. Not because each email is better. Because every email actually gets sent.

5. You need two humans, not zero

The "set it and forget it" pitch is wrong. SaaStr says you need two specific people:

Person 1: A vendor-side implementation specialist. Not a sales rep. Not a customer success manager who checks in monthly. Someone inside the platform with you, troubleshooting training issues and pulling data. The best vendors have these people. If yours doesn't, that's a signal.

Person 2: An internal owner who's obsessed with the output. Someone who reads every message in the first 30 days, watches response rates daily, and treats the AI agent like a new hire that needs coaching.

For a small team, person 2 is probably the founder or sales lead. The time commitment is real — maybe 30-45 minutes a day for the first month. But without that investment, you get mediocre output, blame the tool, and conclude AI SDRs don't work.

They work. They just can't run in a vacuum.

6. Read every message for the first 30 days

SaaStr skipped this early on. It cost them.

In the first month, the AI got facts wrong, used awkward phrasing, misread company context, and replied to objections in ways that would make you cringe. They didn't know because they weren't reading everything.

Their rule now: for the first 30 days of any new agent, read every outbound message and every inbound reply. Every one. After 30 days — once you've confirmed the agent handles edge cases well and responses are on-brand — move to spot-checking.

If you're not reading the output, you're not running an AI SDR. You're sending unsupervised emails to your best prospects.

7. Budget 30 days of ramp, not 30 minutes

Every vendor will tell you setup takes a day. Technically true. Being good takes 30 days minimum per agent.

Those 30 days go toward:

• Training on your ICP with real examples, not generic descriptions
• Testing messaging and iterating based on actual replies
• Building escalation rules (what goes to a human, what the AI handles)
• Calibrating tone, length, and follow-up cadence
• Finding which segments actually respond versus which ones ignore you

The companies that conclude "AI SDRs don't work" are almost always the ones who judged the tool in week one. You wouldn't fire a new SDR after seven days. Don't do it to your AI agent either.

8. Start with chat. Voice can wait.

SaaStr's data: about 85% of prospects prefer chat-based outreach. 15% prefer voice. Chat is easier to implement, train, monitor, and fix when something goes wrong.

Voice is the natural second step once chat is working. Video AI is "two orders of magnitude more work," according to Lemkin, and not worth it unless you have a very high average deal size.

For a small team, this simplifies the decision considerably. Pick email and chat. Make those work. Don't let a vendor upsell you into a multichannel AI stack before you've proven value in a single channel.

9. Don't build around someone who might leave

GTM turnover is high. If your AI SDR strategy is built around one person's voice clone, messaging style, and relationship knowledge, and that person leaves, you lose everything.

SaaStr watched founders build entire AI SDR implementations centered on a CRO or VP Sales who was gone three months later. All the training, voice cloning, and sequences — orphaned.

For small teams, this is especially real. If your best closer leaves, the institutional knowledge often walks out with them. Build your AI agents on documented processes that multiple people understand, not on one person's style.

10. AI scales what exists — including your problems

The hardest lesson, and the one Lemkin puts last.

If your outbound doesn't work with humans, AI won't fix it. Bad messaging gets scaled. A wrong ICP gets targeted more efficiently. An uncompelling offer gets delivered to 40x more people. The tool amplifies whatever is already there.

The fix-the-fundamentals checklist:

If all four are yes, AI will scale them and the results will be real. If any are no, AI will scale the gap.

---

What this means for a 5-10 person team

SaaStr is not a typical sales org. They're a media company with a massive audience, warm lists from events, and a Chief AI Officer dedicated to running agents full-time. Most of what they learned translates, but the scale doesn't.

The adjusted version for smaller teams:

Start with one agent, one segment, one channel. Don't try to replicate SaaStr's 20-agent setup. Pick the one workflow your team consistently neglects and deploy there. Re-engaging old leads is a good first candidate. So is working small accounts nobody wants.

Your founder is probably person 2 — the internal owner who reads every message and coaches the AI. Budget 30-45 minutes a day for the first month. After that, 15 minutes.

You don't need 800 contacts. A campaign of 300-500 highly targeted contacts will teach you more than 5,000 generic ones. And you can actually handle the replies.

Most importantly: fix the playbook first. If your reps can't articulate why a prospect should respond to them, the AI can't either. Get the human version working. Then hand it to the machine.