SaaStr 的 Jason Lemkin：先跑通人类 SDR 流程，再考虑用 AI SDR

Dear SaaStr: Should I Use an AI SDR Before I Have a Working Human SDR motion?

No.

This is the single biggest mistake I see across all company stages, from $1M ARR startups to multi-billion-dollar public companies, and it will sink your AI SDR deployment.

The Core Problem

If outbound doesn’t work with humans, buying an AI to do it won’t fix that. The AI SDR is not a solution to a broken sales motion—it’s a force multiplier for a motion that already works. If you feed it a playbook that doesn’t convert, it will just execute that bad playbook at infinite scale, faster, and with more consistency. You’ll burn through your budget and damage your brand reputation with spam that the AI sends on your behalf.

Here’s what I’ve learned after deploying 20+ AI agents at SaaStr: the tool doesn’t matter as long as you pick a good one. What matters is how you train it. And you can’t train something well if you don’t know what “working” looks like in the first place.

Why This Matters: The Training Reality

When we first deployed an AI SDR at SaaStr, we failed for the first 30 days. Generic messaging, terrible response rates, prospects marking us as spam. The difference? We were treating the AI like software instead of like a new hire.

What changed everything was this: we stopped trying to let the tool figure it out on its own and instead cloned our best human SDR—the person who was actually converting. That person had a specific cadence, specific messaging angles, specific objection handling. Once we fed all of that into the agent, the results flipped. Now our AI SDR sends 3,000+ emails per month—10x what our human SDRs used to send—with better response rates. We’ve built $2M+ in pipeline from AI outbound alone.

But that only works because we knew exactly what worked first.

"Training an AI SDR to do outbound or inbound really isn't that hard.

You just have to actually do it, and carefully.

Train it on your latest sales collaterial. Your website. Your pricing. Your top 100 questions customers ask.

You just have to actually, really, do this… pic.twitter.com/HGVijlNcKt

— Jason ✨👾SaaStr.Ai✨ Lemkin (@jasonlk) September 21, 2026

The Specific Steps You Need to Take First

Before you even think about deploying an AI SDR, you need to:

Prove the motion with at least one human. This doesn’t mean you need a massive team. Start with founder-led sales or redeploy your best performer to outbound for 4-6 weeks. You need to know which messaging converts, which segments respond, what cadences perform, and what objection handling closes deals.
Document what’s actually working. The prompt isn’t some magical incantation. The prompt is: “Here’s the script I used with a customer I closed.” Write down your best SDR’s playbook. The exact email templates they use, the exact call scripts, the sequences they follow, the timing between touches. This becomes your training data.
Hook it up to real data and results. Once you have proven messaging, you connect it to your CRM (Salesforce, HubSpot, etc.) and your prospect data. The AI ingests all of this and learns what works for your specific business, not some generic “best practice” from the vendor’s demo environment.

The “Pure Cold Testing” Trap

Here’s where founders and sales leaders get seduced into a bad decision: vendors will tell you that you can use their tool for “pure cold testing”—basically, testing new messaging you’ve never tried before at scale. Sure, technically you can do that. But you will almost certainly be disappointed compared to scaling something that already converts.

The ROI and the learning curve are both dramatically different. A well-trained AI SDR cloning your best motion will deliver results in 2-3 weeks. A cold-testing experiment will burn budget for months and teach you very little because you won’t know if poor results are due to the message, the targeting, the cadence, the time of day, or just the fact that you’re testing something nobody validated yet.

What You’re Tempted to Do Instead (And Why It Fails)

I was on a call with a global technology company—one we’ve all heard of, pretty AI-forward—and they said: “We’re thinking about buying our first AI SDR. We’re just going to buy it and hand it to our SDRs and have them figure it out.”

That’s not going to work. And here’s why: you have to train it. You have to train it for a month. You have to iterate every single day. You have to do onboarding. You have to understand what “done” looks like before you can automate it.

The Timeline

Assume this process takes:

2-4 weeks of your best sales brain to nail down ICP and messaging that actually works.
1-2 weeks of technical integration with your CRM and data sources.
30-45 minutes per day minimum for the first 60-90 days of ongoing tuning and iteration.

After that, the AI SDR can handle the segments your human team considers “too small,” the leads they consider “too cold,” and the follow-ups they keep meaning to send. It can scale your warm outbound to volumes no human team can match. We went from 285 emails per rep per month to 3,221 per month with equivalent or better response rates.

What an AI SDR Actually Does (And Doesn’t Do)

Use the AI SDR to cover the motions where speed and volume matter most. Redeploy your human SDRs to the work where judgment and relationships matter most—navigating complex organizations, handling nuanced objections in real-time, building relationships with enterprise buyers.

That’s not a downgrade for your team. That’s a better use of everyone’s time and more pipeline from leads that would have otherwise produced nothing.

But you have to prove the motion first. There’s no way around this step.

10 Things to Know Before You Deploy Your First AI SDR: The Very Latest with SaaStr’s Jason and Amelia