SaaStr用AI Agent自动化续约谈判：全量账户定制Deck与叙事审批流

Every company cuts the same corner on renewals.

You go top down. Your biggest accounts get a real custom deck with their actual results in it. Everyone else gets a templated follow-up email. Not because you don’t care about them. Because nobody has a week to build 25 custom decks, and the week you’d spend has to come out of new business.

Amelia built a renewal agent on top of 10K, our AI VP of Revenue, in about half a day. That corner is gone.

"Our latest AI agent? A renewal agent.

Because it knows so, so, so much about every renewal, it builds renewal emails, renewals decks, and renewal data better than any human ever could.

And it can do it … for every single renewal. Not just the big ones." https://t.co/pAPqzBtiKt pic.twitter.com/stMSEIQRNX

— Jason ✨👾SaaStr.Ai✨ Lemkin (@jasonlk) August 31, 2026

Why we built it instead of pointing an AI SDR at renewals

We already run four agentic sales tools. Artisan, Agentforce, Qualified, Monaco. We like them, we pay for them, we recommend them.

None of them could do this. The reason is instructive if you’re building in the space.

Our renewal pitch depends on data that has never lived in a CRM. How many times we mentioned a sponsor on the podcast. How many articles we wrote about them on saastr.com. Their social impressions from our accounts. Their actual badge scans and lead counts from our event platform. What Jason said to their CEO by email three weeks ago.

Third-party sales agents write a decent email. Better than 99% of humans, which is what matters most days. They aren’t great, because none of them have that context or can reason on top of it. So we built on top of them.

What it pulls

The agent starts in Salesforce, which is running headless for us and reachable entirely through the API: who the AE is, what they signed, what they’ve signed historically, lifetime value, email opens, their Qualified chat conversations, their Momentum call history.

Then it goes to everything that was never in Salesforce:

• The saastr.com WordPress API, for every article we’ve published about them
• Our social accounts, through their APIs
• Our podcast archive, for every mention
• Bizzabo, for the real event numbers: leads, companies, seniority mix
• Amelia’s Gmail, through the Replit connector, because we don’t log every email to the CRM

That last one required a decision. Amelia thought about what could go wrong with an agent reading her email, decided the contacts in there were worth it, and let it in.

The deck gets built through the Gamma API, and that choice was not incidental

The agent generates the deck through Gamma’s API rather than any of the generalist tools.

Replit, Canva, Claude, most others will all make you a deck. What they also do is generate an approximation of your logo and your brand imagery, and you can spot it instantly. A renewal deck with a hallucinated version of your own logo on it is unusable.

Gamma has an API path that preserves our real images and our real templates, which is why it won here. They’re at $150M in revenue doing this one thing, so they enforce the guardrails that one thing needs.

What comes out

The Replit deck opens with their results. Number one sponsor at the show. 5.9 million social impressions across our channels. 76% of their leads at VP level or above, which is very high for an in-person event. A sample of the 214 articles we’ve written about them. Who they met at SaaStr Annual, so their sales team can go check what closed.

It goes to the whole account rather than one contact. 10K already has every contact from Salesforce and from PandaDoc through our CPQ flow, so it tells us who to send to and writes a different note for the CEO than for the person who actually signed the contract.

Twenty to thirty have gone out. Before agents we would have built five, for the diamonds.

The Silver sponsors (our ‘lowest’ tier) replied at a higher rate than the Diamonds (our highest)

Our silver sponsors are the smallest checks and, historically, our lowest renewal rate. Smallest and least likely to come back. They got more replies to this than our biggest accounts did.

A $25K sponsorship is real money to the company writing it, and often a real bet on their year. The same $25K to Google Cloud is a rounding error. Back in 2022 when things were crazy we actually discouraged the smallest sponsors, because if $20K was that consequential to you, the stress on both sides wasn’t worth it.

You should treat every customer like kings and queens whether they paid you $25K or $1M. The reason nobody does is that the labor never scaled. It scales now, and the accounts that appreciate it most are the ones that used to get the template.

The agent writes the pitch before it writes the deck

The first decks were methodical and flat. Here’s what you bought. Here’s what you got. Thank you.

Amelia had to teach it to sell. This is a renewal. Put the heavy hitters in the first three slides. Make the case for doing more. Nobody reads the whole deck, no matter how good it is.

Then she added a step, and took the term for it from Westworld. Before the agent generates anything, it writes out the narrative it intends to tell, and we agree on it first.

On one silver sponsor, its proposed narrative was “you’re a silver, upgrade to gold.” Reasonable, and wrong for that account. They had come out of stealth right before the event and had grown a lot since, so the right pitch was three options including a media tier with content, since they’re speaking a lot more now. Amelia corrected the narrative, not the deck. Then it built.

Every agentic sales tool writes roughly the same pitch every time because it has no context. Approving the narrative takes a minute and gets our context in without us building the deck.

It still makes up numbers

Amelia wrote DON’T MAKE UP NUMBERS into the agent’s instructions four separate times, in caps. It still occasionally invents one.

Check every figure before it goes out. Every number in a renewal deck is a number you’re putting in front of a customer who knows exactly what they got.

Where the human still is

We ran it both ways to find out where the human should sit.

Amelia split the list. On one group, she gave the AI SDR enough context to lead with the custom results in the first email. That didn’t work. On the other, the AI SDR ran a normal sequence to get the open and the reply, and then she sent the custom deck herself as the response.

The second one worked much better. The deck is overkill for someone who isn’t ready to hear from you, and the replies to that first email gave us feedback we could fold into the pitch before generating anything.

So the agent does the top of the funnel, and I never let it send the follow-up unsupervised. Then the deck goes out as a human reply. Everything we send is tracked, and 10K picked its own heat-mapping tool, so we get an alert when the person actually reads it.

The next version already exists in the pipeline

Two things came out of building this.

• The first was much, much better (and automated) champion tracking. We wanted to find contacts who had left a renewal account and see where they went. 10K did a poor job of it, so Amelia handed it to Claude directly, which ran Clay through its connector, decided the browser was faster, finished the job in Cowork, and then suggested running ZoomInfo inside Clay. That produced the best enrichment list we’ve had. Those people are now in a sequence that pitches their new company, and when they reply they get a custom deck too.
• The second was 10K’s own suggestion: run the whole thing in reverse on inbound. Today you can book a meeting instantly with Amelia AI, which beats a contact-us form by a thousand times. Next, every inbound that books or downloads the media kit gets an automatic custom proposal back. Who’s similar to them, who’s had success, what they should do with SaaStr and why. No human in the loop.