We talk a lot on The Agents pod about what our agent stack does. This episode Amelia walked through what it actually, exactly looks like, screen by screen, with the real inputs and the real outputs: the leads, emails and decks that went out this week.
Sponsorship revenue has doubled in the last 12 months. Three humans, 21+ agents in production. Below is where that came from, what’s working, and the one part of the funnel that still isn’t solved.
The numbers, before anything else
- Sponsorship revenue: 2.1x year over year, with the year just started
- New business from inbound: up 60%, directly attributed to the inbound agents
- Renewals: 60% ahead of last year year to date. We’ve already passed all of last year’s renewal total in roughly three months since Annual
- Outbound revenue: up 124%
- ~3M website sessions in the last year, 17,000 conversations with our inbound agent, ~600 meetings booked for SaaStr AI Annual, at roughly $90K ACV
A 60% increase in new business is a big number at any company. At three people it’s a different number entirely.
10K is the backend, and Salesforce is headless underneath it
Everything runs through 10K, our AI VP of Revenue, built on Replit. He is the dashboard we open every day, the queue, the renewal tracker, the AP tracker, and the thing that writes to Salesforce.
We didn’t start there 6+ months ago when we built the very first version of 10K. We started with a basic Replit dashboard, hooked Salesforce into it, and it turned into headless Salesforce almost by accident. Today our full-time human sales exce logs into Salesforce … sometimes. Amelia logs in to click a button as necessary. I never log in at all. Nobody really uses the UI.
Salesforce is still the system of record. Headless 360, Momentum for call recording, Qualified, Marketing Cloud (now on Agentforce), classic Sales Cloud, Slack. All of it feeds 10K, and 10K is plugged into about 30 other things on top of that. That’s why headless works for us: the agent can pull from all of them at once, and the Salesforce UI only shows you Salesforce.
Inbound door one: the agent on the site has had 17,000 conversations
Thirteen months ago the customer / contact us / sponsor page was just … a page with a long contact form. It came to Amelia, she round-robined it to herself or David, and someone followed up within about a day.
The follow-up email was always some version of: “Hey [company], you look like a great fit for SaaStr, let’s book a time.” Amelia’s take on that email, unedited: it’s the worst email on planet Earth. If you’ve raised your hand and reached out to a company, the last thing you want back is a form letter telling you that you’re a great fit.
Now the sponsor page has two doors. Door one is Amelia AI, our avatar on Qualified. She answers questions in real time, qualifies (budget, lead gen vs. brand vs. speaking, which competitors they’re watching), and books the meeting on the spot.
The results over the last year:
- 3M sessions
- 17,000 conversations
- ~600 meetings for Annual
- A couple million in influenced and closed pipeline
- Real logos, including OpenRouter
Some of those 17,000 were goofball conversations. Doesn’t matter. The conversion path is clear: they do their own discovery, they talk to the agent, they book their own meeting, they close. For a tech-centric buyer, that’s how they want to buy. And for AI-native sponsors specifically, seeing agents in the sales process is part of the evaluation. If you’re pitching yourself as one of the top AI events in the world and the buying experience is a PDF and a two-day email lag, they notice.
Inbound door two: a tokenized prospectus that rewrites itself 10 minutes after you download it
For a while we debated killing the self-serve path entirely. We didn’t, because a real share of buyers won’t talk to an avatar. They want the packages and the pricing, not a conversation, at least not yet.
The old self-serve path was a download link to a Google Slides prospectus. It converted worse than Amelia AI for a year. Here’s what replaced it.
- Step 1. Shorter form. On submit, they get a tokenized version of the prospectus on our site, unique to their company, not a static PDF.
- Step 2. Microsoft Clarity (10K picked it! we’d never heard of it) heat maps what they do on it. 10K picked Clarity as the vendor itself.
- Step 3. The agent waits 10 minutes. We landed on 10 because most self-serve visitors click off after 5 to 7, so 10 means the session is done and we have the full heat map.
- Step 4. The agent builds the narrative. It runs first-party signals first: have they hit the site, are they in an outbound sequence, have they seen our LinkedIn or X ads, are they on the newsletter, have they attended, have they spoken, have we written about them on SaaStr.com. Then competitors as the second-biggest signal. That takes 30 to 60 seconds.
- Step 5. It emails Amelia, Slacks her, drops the lead into a live dashboard queue, and writes it to Salesforce with the heat map and the custom link attached. That last part used to be a quarterly CSV upload.
- Step 6. She approves or edits the pitch. The email goes out from her, personalized. And the prospectus link they already have updates in place.
That last piece is the one worth copying. The prospect downloaded something generic. Ten minutes later they go back to the same link and it says “Marlin, here’s why Base44 should be at SaaStr,” with their competitors and their recommended packages in it. Same URL, and now it’s a living document we can keep updating through the whole cycle.
The Base44 example in the episode was a real lead from the day before. The heat map showed him on the package overview, briefly on Super Gold, mostly on Gold, then a long stretch on the contact form. He’d also self-reported Gold and Super Gold. The pitch the agent wrote led with their competitors who were at Annual and the public, non-proprietary result that Replit was top of the leaderboard as the number one sponsor.
He replied to say it was a great follow-up and it was fast, and booked for Friday.
We rebuilt Calendly in 20 minutes because the calendar was the last silo
Before: David sent his Calendly, Amelia sent her Read AI link, two different tools, neither connected to anything.
10K’s suggestion was to just build our own. It took the agent 20 minutes. Now the booking link carries the company name, the agent tracks whether they hit the page and didn’t book, and it ties back to the prospectus they were looking at. When someone opens the booker and bounces, the agent tells Amelia and offers to draft the next email.
Amelia’s first reaction was that it’s just a calendar, who cares whether it’s part of the flow. At 20 minutes of build time it was worth doing anyway, and it closed the last tracking gap in the funnel.
Routing is agent-decided too. Base44 got routed to Amelia rather than David because she owns Replit and Lovable, and the agent weighted that volume over David owning Vercel.
The renewal agent: different slides for the CEO than for the events team
The renewal agent is about a month old and it’s the reason renewals are 60% ahead.
Two mechanics explain most of it. First, we now contact every renewal. We used to work top down and spend our time on the big accounts. Agents follow up with every customer at every ACV. Second, depth of context: anything worth reporting (leads, impressions, coverage) becomes a real-time or monthly touch instead of a conversation that only happens at re-up time.
The decks are built in Gamma. We tried a few other tools and got AI-generated festival imagery and malformed people. Gamma held to the data, the stats, and the real photos we gave it, including photos of the customer at their own booth.
The audience segmentation is what surprised us. On the Replit renewal, the agent said: lead with top of the leaderboard at Annual for the events and marketing team, but the CEO gets a single slide, and the slide is the 6M impressions across everything we’ve done together all year, not the event. Two audiences inside one account, two sets of proof. Amelia didn’t know the full promotional footprint until the agent assembled it.
Then the harder case. A three-time sponsor was down this year and we didn’t know it going in. The agent flagged it while building the deck, and instead of a happy-path narrative it wrote a rebuttal: you were number one in your category the last two years, this year you were down both as a company category and within your sponsorship level, so let’s figure out why and fix it for 2027. That account re-engaged instead of taking a year off. Assembling that comparison by hand would have been impossible.
Outbound is always being improvedf
Outbound was the first agent we ever ran, a year ago, and it’s now the weakest surface relative to the other two. Revenue from it is up 124%, so it works. Inbound and renewals just got so much better that outbound looks like the laggard.
It’s still a lot of data cleanup and a lot of manual follow-up. So we reversed the model. The agents used to run the full sequence on autopilot and we’d reply when needed. Now they send one to three emails maximum, and the moment someone responds, a human plus the agent takes the reply, because the reply is where the customized deck and the customized prospectus go. “Here’s a link to my calendar” is a wasted response.
Three tools, three segments:
- Artisan for warm outbound: people we know, people our enrichment flagged as having moved
- Agentforce for website visitors we know through Qualified, and anyone we’ve ghosted, since they’re already in Salesforce
- Monaco for net-new accounts that look like our best accounts
We also do manual outbound to companies our agents are already working. We do both. No single tool is at the point where you drop everything else for it.
No single enrichment tool wins, and the agents want all three
We have more than 500,000 people in the database between newsletter signups, event attendees, and everyone who has run through these flows. People in this industry change jobs constantly, so champion tracking is a permanent job.
Asked to pick one tool, both 10K and Claude said Clay, because the waterfall gets the highest hit rate. But neither wants to pick one:
- ZoomInfo runs the first pass. Cheapest credits, and it confirms who’s still where
- Sumble catches what ZoomInfo misses. Recovers roughly 50% more people, added in the last week
- Clay cleans up everything left, because it’s the most expensive
It’s now a Claude skill that calls all three to verify any list before it goes into an outbound agent. Nobody talks about enrichment much, and it decides whether the rest of the outbound stack has anything real to work with.
Measure once, cut twice: the renewal agent paid for inbound and outbound
The renewal agent was the first time we hooked up everything: all of SaaStr.com, all the social, the podcasts, every agenda from the past year, on top of Salesforce.
That was a one-time cost. The inbound agent uses the same corpus. Outbound now uses it too, generating the same class of deck and custom page for anyone who responds. The inbound build was the agent’s idea, not ours: it looked at what we’d built for renewals and said you already have most of the pieces, do this for inbound.
So do the expensive data plumbing once, on the surface where you have the most information. For us that was renewals, where we have years of proprietary touchpoints. Every other surface then gets cheap.
When 10K went into stupid mode
For a few weeks the agent got noticeably worse. It told us it had too much in it. Too much data, too many APIs, too much surface. We cleaned it up and modularized, and the quality of its ideas came back.
Two things were probably happening at once. One was the app getting too big. The other was a model transition. Fable 5.1 landed September 1 and its advice on what to build and what to improve on complex systems is better than what came before. If you tried this six months ago and it felt too hard, try again. It’s not as hard as you think. It does have to be built stair-step, bit by bit, rather than all at once.
10K’s current pitch: be the event that agents recommend to other agents
The ideas have changed character. A year ago it was suggesting a referral program for tickets. Now it’s telling Amelia she should make SaaStr the event that other agents recommend to other agents, with an actual game plan attached, and the argument is that we’ve been marketing to humans for a decade and the buyer composition is changing.
It’s out there. We’re all-in on it anyway.
The other thing it wants to build is a version of itself for the sales team. David logs into Salesforce more than the rest of us because he doesn’t have the same backend access to 10K. So the agent proposed a rev-ops copilot built around what the sales team needs rather than what Amelia needs. Most of what we get out of 10K now is product recommendations rather than completed tasks.
Why we built our own, and why we still tell everyone to buy
We’d rather buy all of this. It can’t be bought today.
Every third-party outbound tool narrows the data it uses for customer communications by design. That’s how they make one product work for many customers. Their signals (LinkedIn posts, web data from Exa or Parallel or their own database) are still better than 95% of humans. But they’re outside signals.
The most valuable asset we have is our own data. The email that remembers you sponsored three years ago, names your three competitors and two partners who will be there, and shows your 2024 ROI even though you skipped 2025 and 2026, is not something any of them will build, and pushing it all into Salesforce doesn’t help because the products don’t use it. Amelia has raised this with Agentforce, with Sam at Monaco, and with Jasper at Artisan. It isn’t their core.
The outside signals turn out to be the easy part. Core Signal or Exa and you’re most of the way there. We assumed for a long time that was the hard half. The proprietary half is the part nobody can sell you, and it’s what moved our conversion rates.
What to copy first
If you’re starting from a contact form and a round-robin, the order that worked for us:
- Put an agent on the highest-intent page and let it book meetings directly. Ours has had 17,000 conversations and booked 600 meetings.
- Keep the self-serve path. A real share of buyers won’t talk to an avatar yet, and killing that door costs you deals.
- Make every asset tokenized and living. One URL per prospect that you can rewrite after you learn something is worth more than a better PDF.
- Do the data plumbing on renewals first. It’s where you have the most proprietary context, and everything else inherits it.
- Wire the boring surfaces in. The calendar took the agent 20 minutes and closed the last tracking gap in the funnel.
Three humans, 21 agents, 2.1x on sponsorship revenue. Copy what’s useful, then keep improving it each month and each quarter.
