SaaStr实战：用AI Agent接管联系表单，月贡献71%成交

If you’re trying to figure out where to deploy your first real AI agent in go-to-market, start with the most boring page on your website. The Contact Me form.

No one should wait days to hear back from a 21-year-old human whose job is to qualify them. And yet that’s still how most B2B companies run inbound in 2026, including a surprising number of AI startups we meet. They’ve raised on an agent story and their own website runs a form, a queue, and an SDR who gets to it tomorrow.

We ran it that way for years at SaaStr too. Here’s what changed when we put an agent on it, and what we learned running it across 2M+ sessions.

"One of the simplest, highest ROI thing to put an agent on?

Your Contact Me form.

No one should ways days to hear back from a 21 year old human to qualify them." pic.twitter.com/9Ftfu34hdw

— Jason ✨👾SaaStr.Ai✨ Lemkin (@jasonlk) October 15, 2026

Our Old Flow Took 2 to 3 Days, and About 1,000 Sponsor Leads Never Got a Reply at All

The old process on our Squarespace site (yes, Squarespace!) was the standard one. A prospect filled out the form. It went into a queue. A human round-robined it to an AE. The AE followed up when they got to it. Then came the back-and-forth on calendar times, and finally a meeting where the first 10 minutes were spent on basic discovery we should have already known.

End to end, two to three days on a good week.

That was the version that worked. After one SaaStr Annual, I audited Salesforce and found roughly 1,000 people who had filled out our “I’m interested in sponsoring” form and gotten zero human follow-up. Ever. These were people who raised their hand and asked to give us money.

That’s the real cost of a form. Every handoff is a place a warm lead cools, and some of them just fall on the floor and nobody notices.

What the Agent Did Instead: 614 Meetings at an ~$85K Average Deal Size

Our inbound agent is called Amelia AI. Under the hood she’s Qualified’s Piper, built around a custom avatar of our Chief AI Officer, Amelia Lerutte. She launched to fix inbound, and the old flow is gone.

The numbers for a single SaaStr AI Annual cycle:

• 614 qualified meetings booked
• ~$85K average ticket size on those deals
• ~2.25 million sessions on the Annual site
• 400K+ interactions handled

Not every one of those meetings closed. If they had, we’d have $60M in sponsors instead of $10M. But the efficiency is real, and the cumulative result is $1M+ in closed-won revenue sourced directly by the agent.

The stat that mattered most to me: in one month, 71% of our closed-won sponsorship deals came from AI-qualified inbound leads. Our historic share of closed-won from inbound had been 29% to 34%.

We could never staff that volume with humans. It would take three BDRs, and they’d quit every three months.

A well trained agent is better in almost every single way than a junior human SDR in qualifying inbound https://t.co/eiMOQIYba0

— Jason ✨👾SaaStr.Ai✨ Lemkin (@jasonlk) October 26, 2026

Why This Is the Easiest Agent to Get Right

Most agent deployments are hard because the job is fuzzy. Inbound qualification is the opposite.

The buyer has already raised their hand. The job is bounded: answer questions, qualify, route, book. The success metric is obvious: meetings booked and revenue closed. And the downside is low, because the alternative you’re replacing is a form that sits in a queue.

The agent also filters noise well. Early in our run, the funnel looked like this:

• 668,591 total sessions
• 45,188 sessions where the agent engaged, only on intent signals like the pricing page, sponsor pages, or repeat visits
• 1,025 real conversations
• 91 meetings booked and 99 Salesforce records created

Most visitors are reading content and don’t need to be qualified. The agent leaves them alone. When someone hits the sponsor prospectus for the third time, it shows up in two seconds. A human team doing live chat might handle 100 to 200 of those sessions. The agent covered 45,000.

Compare that to outbound, where you need clean lists, segmentation, warmup, and a lot of iteration before anything works. Inbound agents start producing in weeks.

Amelia AI (Our Contact Me Agent) Gets Every Release Our Other Agents Get

Amelia AI is the most-trained agent in our stack and has one of the biggest knowledge bases. That’s not an accident, and it’s most of why she converts.

What we do:

• She crawls saastr.com and the Annual site every day. Pricing, agenda, speakers, sponsor packages, all current.
• Every release we push to 10K (our AI VP of Marketing), QBee (our AI VP of Customer Success), or Annie (our Annual website agent), we push the same context to Qualified. She’s never a version behind the rest of the company.
• Two buyer types, trained separately from day one. Ticket buyers and sponsors are different conversations with different questions. If you have self-serve and enterprise, or SMB and mid-market, the agent needs to know which conversation it’s in.
• A separate, smaller brain for in-person attendees. At the event, people ask “where is this session” and “where do I pick up my badge.” A tight, venue-specific context answers that fast without dragging in 14 years of saastr.com.

An inbound agent answering from stale data is worse than no agent. It answers confidently and gets it wrong.

One tip from Qualified’s own demand gen team at SaaStr AI that matched our experience: write the agent’s guidelines in plain human language. Using another LLM to generate the instructions performs worse. And give it the same onboarding materials you’d give a human SDR, including the PDFs, decks, and competitive notes that never made it onto your website. An hour of setup gets you off-brand answers.

Routing on Who Actually Closes, Not Round Robin

Round robin is fair to reps. It’s not optimal for revenue.

Amelia AI routes by weighting our Salesforce data. She books most deals with the rep who closes that deal type best, and sends the deals that fit a specific closer to that person.

It’s not set-and-forget. For a while she over-indexed one of us on certain accounts, and we had to correct the weighting. But even with the occasional drift, it beats a queue where whoever’s next gets the $150K sponsor.

The Jobs Beyond Chat: Abandoned Sponsor Pages and VIP Ticket Codes

The chat window is the least interesting part. The agent earns its keep on what happens when people don’t finish.

• Customer / sponsor page abandonment. If you visit the sponsor page, don’t book, and we know who you are from Salesforce, she follows up with a meeting offer and a few lookalike companies in your space who already sponsor. She excludes anyone who’s already a sponsor so we don’t pitch existing customers.
• Ticket abandonment. If you visit and don’t buy, she sends a VIP code, and follows up if you don’t use it. That one campaign has sold hundreds of thousands of dollars in tickets.
• Discounting. Humans are bad at this. We don’t love discounts, but years of data say it’s better to mark up 20% and offer a 20% discount, because that’s how buyers actually behave. A human rep under pressure slides from 20% to 34% the moment a deal wobbles. The agent follows the guardrails on a schedule, every time.

Full CRM Context Is Why She Converts

When someone lands on a sponsor page, the agent already knows if they’ve attended before, whether a colleague has talked to us, what their company sponsored in prior years, and what size they are. She’s not starting from zero, and the prospect can tell.

The downstream effect: the AE gets a full dossier before the call. We skip discovery because the agent already did it.