After just a month of launching our own SaaStr AI (trained on a decade of content and hundreds of founders, speakers, and events), we’ve already had over 34,000 (!) conversations with our AI in just the first 30 days.
Our AI has ingested 10,000+ SaaStr posts, 2,000+ YouTubes, 800+ Podcasts, and 100,000 of our tweets. And trained on all that — it’s pretty darn good.
It’s completely opened my eyes to what AI can truly do in the real world – not just in coding and support, but what it will do for all our apps. It’s run by Delphi.ai and try it here:
I did a deep dive live on Workshop Wednesday with my Top 10 Learnings, and brought in Simon Farshid, founder of Assistant UI, which offers the leading platform to add ChatGPT-like interfaces to your own app and to add some color. With 4,000 GitHub stars and an impressive customer base, Simon’s seeing firsthand how this framework works across industries.
and slides here:
The Scale is Mind-Blowing, And Radically Accelerating
As Sam Altman noted, ChatGPT went from 1 million users in 5 days to 1 million users in an hour. That’s several orders of magnitude difference in just 26 months.
With SaaStr AI, we’ve had almost 35,000 founder conversations in less than 30 days. There’s no human way we could have this many conversations – not me, not our team. This isn’t about replacing humans or efficiency. It’s about doing things at a scale where there’s simply no human alternative.
10 Key Learnings From Our AI Roll-Out:
1. Users Will Tell AI Things They Won’t Tell Humans
We discovered people are asking our AI about sensitive topics they’d never bring up with me or our team directly. Topics like “I’m thinking about quitting” or deeply personal founder struggles.
This has massive implications for B2B SaaS. Your AI can become a channel for feedback and insights you’d never get otherwise. Customers will openly discuss problems they’d never bother reporting to your support team. This creates an unprecedented feedback loop that can transform your business.
2. The AI Is Shockingly Good – It’s Often Better Than Me (For Some Things)
One of our fastest-growing portfolio companies had all board members and investors provide feedback on their investment memo – and the AI won. It beat me and the other investors in feedback. All the feedback was good, but the AI overall was the best.
Why? Because it remembered 12 years of my thinking and applied it without forgetting anything. It gave the best feedback on TAM, team, positioning, competition, and markets. Importantly, it was vastly better than what I would have gotten from a generic ChatGPT prompt. More on the below.
3. Hallucinations Aren’t a Big Deal for Our AI – And Getting Rarer
Everyone was skeptical about AI hallucinations initially, and many founders still mention this as a blocker. But the reality? In our experience, hallucinations are now:
- Correct ~80% of the time
- Mediocre/generic ~15% of the time
- Wrong only ~5% of the time
Even when it does hallucinate, it’s typically about edge cases or personal details outside the core domain knowledge. The hallucination rate continues to drop dramatically, and within 12 months, we might be looking at sub-1% error rates.
Also training is easy. When I see a mistake (e.g., the AI got the dates wrong for SaaStr Europa 2025, which is probably because we haven’t launched yet so it’s “guessing”), I just make a quick voice note and the model is almost instantly updated. And the hallucination goes away.
4. The Digital + Human Combo Creates a Superpower
One of the most interesting paradigms we’ve discovered is when there are three of us in a conversation: two humans IRL plus digital Jason (or digital Brian Halligan sometimes). This creates an IRL co-pilot that’s transformative.
I ask Digital Jason what his thoughts are on things I forget or am foggy on … when the rest of the SaaStr team is also in the room. And our AI, Digital Jason, is usually right with his thoughts on the matter.
I’m shocked we don’t already have this for sales calls. By the end of the year, every single sales call should have a digital representative that can fill in when the human doesn’t know an answer. This will utterly change how we do sales.
If your digital sales engineer joins the Zoom and gets product questions 95%+ right … that’s better than 95% of AEs and SDRs 😉
5. Voice vs. Text: The Data May Surprise You
Despite all the hype about voice being the future, our data shows a different story. In the first 5000 or so sessions:
- Web text: 2,000 sessions
- Voice: 500 sessions
- Embed (chat on our homepage): 500 sessions
That’s only 20% voice, 80% typing on keyboards. Voice clearly matters, but we shouldn’t assume it’s taking over everything. Different use cases demand different modalities.
6. Custom-Trained AI Vastly Outperforms Generic Models That Have Ingested The Same Data
One of the biggest surprises was how much better our specialized SaaStr AI is compared to ChatGPT or Claude with the same questions – despite those models having already ingested most of our content through web crawling.
When asked classic SaaStr questions like “How do I hire a great VP of Sales? What answer would SaaStr give”, our AI gives answers that are 50X better than generic models. There’s something about the dedicated training, the interface, and the relationship with users that creates a dramatically better experience.
7. It’s Already a B+ BDR Out of the Box
Without any specific sales training, our AI can already handle basic qualification questions like a B+ BDR. When asked about exhibiting at SaaStr Annual, it provided correct information about sponsorship offerings, the expo hall, contact details, and audience dynamics – all without waiting for business hours or human intervention.
The key insight: users will quickly use your AI for everything – not just what you designed it for. Support, sales, marketing, research – they’ll expect it to handle it all.
8. People LOVE Good AI
I was genuinely surprised by how much people love our AI – especially technical people who know how it works! When smart founders ask nuanced questions about their specific situations, the personalized responses create what Simon calls “magic moments” – something increasingly rare in traditional B2B SaaS.
People are building different kinds of relationships with AI than they do with traditional interfaces. This isn’t just about higher conversion rates – it’s unlocking entirely new ways of interacting with your content and product.
9. The Price/Value Equation is Insane. And a Threat to B2B Vendors That Charge More for Far Less.
The economics of AI tools are mind-blowing. For $29-399/month, we’re getting 6,000+ additional interactions, surprise and delight, video conversations, and more. Compare that to $60/month social media tools that schedule three tweets.
This has profound implications: you must provide much more value in your app than you did a year ago. Don’t view AI as an upsell – view it as how you can make your product 10X better, because that’s what your competition is doing.
10. The AI Itself Probably Isn’t Defensible. The Software Layer Around It Probably Is, However. How Much, We’ll See.
As amazing as our AI is, someone smart could build a clone in a week, especially if they used the same platform. 99% of our content is public. Still, when I tried a different but seemingly similar vendor, I couldn’t get it to really work well at all.
The same rules of B2B SaaS continue to apply. Build a good product around the AI, focusing on the complete experience rather than the underlying technology itself too much. I know our AI uses different LLMs (I think I can see the Deepseek when it pokes out), but I also don’t care.
And the amount of data needed for good AI is decreasing as models improve, which means you might face 10X more competitors than last year.
Key Takeaway for Every B2B Founder
Every single app should have some sort of ChatGPT-like interface, some sort of co-pilot. I wasn’t even sure about this 30 days ago, but now I’m convinced. Although it has to be great.
This is how users want to interact – they don’t want to input data manually, hit refresh, or figure out complex UX anymore. And if nothing else, they will soon expect you at least offer an AI of the level of quality of ours or better.
Look at Assistant UI and other frameworks – your engineers can implement this in a week. Don’t wait to launch your AI for a year. If you don’t have anything like this very soon, you’re going to look pretty old-school.
And try it by clicking on the chat bubble at the bottom right of SaaStr or by clicking here.
