Aug 4, 2025

From SaaStr AI Wednesday with Jacob Efron, Managing Director at Redpoint Ventures If you’re building in B2B today and haven’t figured out your AI strategy, you’re likely falling behind. But if you think integrating basic AI features is sufficient, you’re missing the bigger opportunity. This was the key insight from Jacob Efron, Managing Director at Redpoint Ventures, during our latest Workshop Wednesday. Redpoint has $8B under management and has backed winners like Snowflake, Stripe, Twilio, and Ramp. More recently, they’ve led investments in AI breakouts like Abridge ($5.3B valuation) and Lorra. Here’s what Efron shared about how top-tier VCs are evaluating AI investments—and the key realities facing SaaS founders in 2025. The Economics Have Fundamentally Shifted The data tells a compelling story: AI companies are scaling significantly faster than their traditional SaaS counterparts. Stripe’s data shows AI applications hitting product-market fit and scaling at rates that exceed historical SaaS benchmarks. “When these startups find product-market fit, they’re just scaling way faster than traditional SaaS counterparts,” Efron explained. “This pace of adoption breaks a lot of rules about traditional startups.” The reason? Model costs are plummeting faster than cloud costs ever did. Efron showed data demonstrating that for any given benchmark of capabilities, the cost per token is dropping dramatically year over year—a decline rate that exceeds what we saw during the cloud era. This means: Gross margins that look challenging today will improve rapidly The “AI tax” on unit economics is temporary Focus on end use cases, not current margin profiles Additionally, Efron noted that usage of AI models is growing at rates significantly faster than cloud adoption. “There’s this pervasive feeling in the investment community today that even if you froze the model capabilities, there’s trillions of dollars of applications just waiting to be discovered.” The Founder Takeaway: If you’re dismissing AI applications because of current unit economics, you may be overlooking the same trajectory that made cloud companies successful despite early margin challenges. The Four AI Use Cases Actually Working at Scale Through Redpoint’s portfolio and market analysis, Efron identified exactly four categories of AI applications that have achieved real product-market fit: 1. Conversational Interfaces (Chat) Beyond ChatGPT, this works exceptionally well in customer support. Companies are seeing 10x improvements in response times and resolution rates. 2. Document Search and Summarization This is exploding across verticals—from Glean in horizontal search to legal AI companies like Lorra, to construction companies like Trunk Tools. 3. Speech Processing (Text-to-Speech and Speech-to-Text) Healthcare is the obvious winner here. Abridge, which transcribes doctor-patient visits, has reached a $5.3B valuation by solving “pajama time”—the hours doctors spent typing notes after work. 4. Code Generation The breakout category. From Cursor to Cognition to Poolside, coding AI is seeing explosive adoption because the 10x improvement is immediately obvious to users. The Founder Reality Check: If your AI initiative doesn’t fall into one of these four buckets, you’re probably not building on proven product-market fit patterns. The Redpoint Three-Question Framework for AI Investments Efron shared Redpoint’s exact framework for evaluating AI applications. Every AI investment decision comes down to three questions: Question 1: Is There a Really Effective Wedge for AI? “The bar for what product-market fit looks like has gone up in AI because you’ve seen explosive growth for companies when they do have them,” Efron noted. The key insight: There’s massive top-down pressure for companies to experiment with AI tools. Boards are asking CEOs, “What are we doing in AI?” But you need to distinguish between companies that are just “messing around” with AI versus those where end users genuinely love the product. Enterprise AI Budgets: The Data Behind the Hype Efron shared compelling data on enterprise AI adoption. Morgan Stanley estimates that 25% of global software spend in the next few years could be directed toward AI use cases. This isn’t just speculation—enterprises are already demonstrating real commitment. “In the early days post-ChatGPT, there was skepticism from enterprises about what these models would actually be able to do,” Efron noted. “But the combination of actual real-world impact these models have had, plus everyone watching the pace at which they’ve gotten better, has led to deep interest from enterprises.” The practical result? AI budgets are expanding rapidly, and the traditional rules about enterprise sales cycles are being broken by the magnitude of improvement these tools provide. Question 2: How Much More Can This Company Do From This Wedge? This is about market expansion potential. Efron prefers large industries (healthcare, legal, finance) over niche markets because there are more follow-on use cases to build. He gave a telling example: “You could make a great AI product to write sourcing emails for venture capital firms. This would probably be a great wedge that could get you some initial scale pretty quickly. The issue is there’s only so many venture capital firms.” Founder Insight: A great wedge in a small market is still a small outcome. The best AI companies expand from one killer use case to become the AI platform for their entire industry. Question 3: Will Quality Matter in This Market? “In any category we look at, there’s always going to be somebody who’s going to say, ‘Hey, for 10% of the price, we’ll sell something that’s 50% as good,'” Efron explained. The markets that work are where customers will pay for the higher-quality product. Healthcare and legal are obvious examples—no one wants the “cheap AI” handling their medical records or legal documents. Strategic Implication: If you’re building in a market where “good enough” wins, you’re going to face a race to the bottom on pricing. The Competitive Reality and Market Data Here’s what SaaS founders need to understand: In every AI category Redpoint examines, there are 10+ startups building similar solutions. The funding environment reflects this activity—Redpoint’s data shows AI companies are raising both larger rounds and at higher valuations compared to non-AI companies at similar stages. Specifically, when Redpoint analyzed all Series B and C companies they evaluated, AI companies showed: Larger average round sizes Higher growth rates justifying the premium pricing However, there’s an important pattern: “In any category we look at, there’s really two to three of them that rise to the top,” Efron observed. The differentiator? Building effective AI applications is more complex than it appears. Companies need to build custom infrastructure around foundation models, and that infrastructure needs to evolve rapidly as models improve. The Speed Imperative: “Velocity is probably the most important thing we look for,” Efron emphasized. “The market changes so fast—it’s both a race to build the breadth of what these models can do, but also a race to translate whatever GPT-5 can do to an end industry.” Two Case Studies: How Winners Actually Win Abridge: The Perfect AI Wedge Abridge eliminates “pajama time”—the hours doctors spent typing visit notes after work, which was a major contributor to physician burnout. The wedge is compelling because: Measurable user experience improvement: Doctors reclaim 2-3 hours of personal time daily Clear expansion path: Everything in healthcare flows from the doctor-patient conversation Quality matters: Healthcare systems prioritize accuracy over cost savings The quantified impact: Abridge has reached a $5.3B valuation by solving a problem that affected physician retention and patient care quality. Lorra: The Successful Second Mover Lorra launched after Harvey in legal AI but caught up by: Starting in the Nordics: Built an end-to-end product with leading Nordic law firms before expanding globally Shipping velocity: “The fastest team you can find in terms of shipping new capabilities” End-to-end platform: Not just a point solution, but a comprehensive legal workflow platform Key Insight: We’re still so early in AI capabilities (today’s products are “probably 5% of what they’ll be able to do down the line”) that there’s room for leapfrogging, even for second movers. The Moat Question: What Actually Matters When asked about competitive moats, Efron gave a refreshingly honest answer: “We may have somewhat unrealistic expectations around moats. If you think about reasoning models, they’ve been out since September. People have been building on them for 9-10 months. In SaaS world, nobody would say after 9-10 months you should have this defined moat.” Instead, focus on: Quality differentiation: The thousand little things that make one product delightful versus another Speed of execution: First to market with new model capabilities What This Means for Traditional B2B amd SaaS Companies The workshop revealed a striking reality: Redpoint has collapsed the distinction between “AI companies” and “SaaS companies” in their CRM. “Any SaaS company is building their own AI features,” Efron noted. The Red Flag Test: If a SaaS company hasn’t deeply thought about what AI models can do in their domain, that’s concerning. Even if there isn’t a 10x use case today, there will be as models improve. The Integration Challenge: Companies are ending up with 10+ AI tools that don’t talk to each other. The winner in horizontal AI tooling will be whoever solves the integration problem. The Vertical vs. Horizontal Debate Efron believes vertical AI applications are relatively safe from foundation model companies building competing features. The reason? “So much of the magic of these vertical AI companies is tailoring what these models can do to the specific workflows of how these different places work.” However, horizontal tooling is “much more in the strike zone of what models can do” directly. Strategic Implication: If you’re building horizontal AI tools, you’re competing directly with OpenAI, Anthropic, and Google. If you’re building vertical AI, you’re competing with other startups—a much more winnable game. Marketing AI: The Surprising Laggard Despite the hype, Efron doesn’t think marketing AI is saturated—quite the opposite. “I haven’t seen a really great AI marketing tool yet that can write the email, sort contacts, and figure out the best time to send,” he admitted. There are also entirely new categories emerging, like “the equivalent of SEO for ChatGPT and Perplexity”—what companies like Profound and Bluefish are building. Opportunity Alert: Marketing AI feels surprisingly behind sales AI, customer success AI, and other verticals. There’s still room for category-defining companies. The Bottom Line for B2B Founders Timing is important: AI-first companies are scaling faster than traditional SaaS companies historically did. Having a clear AI strategy is becoming table stakes. Focus on proven patterns: Concentrate on the four validated use cases (chat, document processing, speech, and coding) rather than speculative applications. Wedges matter more than features: Find the one area where AI delivers substantial improvement in your industry, not incremental gains across multiple areas. Speed beats perfection: Model capabilities change every 3-6 months. Teams that can quickly adapt and ship new capabilities capture more value. Quality differentiates: In a market with 10+ competitors per category, the companies with the best user experience tend to win. The AI transformation of SaaS isn’t coming—it’s happening now. The question is whether you’re building the future or adapting to it. Jacob Efron is Managing Director at Redpoint Ventures and hosts the “Unsupervised Learning” AI podcast. Redpoint has invested in AI leaders including Abridge, Lorra, and numerous other applications across healthcare, legal, and enterprise software. This deep dive is based on SaaStr AI Wednesday, where we bring leading investors and operators to share tactical insights with the SaaS community. Join us live every Wednesday for more insights like these. Screenshot