WGMaTW: SaaS comes back to Earth, LLMs vs. marketplaces, and why Main Street shrugs at AI
What Got My Attention This Week, as a product leader
Here’s what got my attention this week — a short list of things I read and watched recently that felt worth pausing on, as a product leader. The common thread this week is that the fundamentals are critical, especially now.
1/ SaaS valuations are officially back on Earth 🌍
The latest SaaS Valuation Multiples (2015–2025) report from Aventis Advisors confirms what many of us have been sensing for some time.
The SaaS boom — launched into outer space during the pandemic — is no boom anymore.
Three signals that matter:
Multiples have dropped sharply.
2016 called — and it wants its valuations back. Median public SaaS EV/Revenue multiples have fallen from the 2021 fantasy zone (20–30×) to roughly ~5× today.Growth ≠ gravity-defying anymore.
Median SaaS growth is settling around 11–12% YoY. The market is rewarding durable growth, not ‘growth at all costs.’ Nothing lasts forever.Profitability is no longer optional.
Margins are improving, Rule of 40 is trendy (but tough), and efficiency is cool again. The narrative for a SaaS business needs to be strong — and it needs to be real.
Why this matters for product leaders:
Build products customers need — and won’t churn from
Design business model, monetization, and tangible differentiation very deliberately
Craft products (and teams!) for trust, efficiency and effectiveness, not excess
In some ways, it feels like a different universe than just a few years ago when your mom and my mom were downloading Zoom and everyone was buying SaaS stocks.
I suspect it will get tougher before it gets easier for some SaaS businesses.
The good news? We needed a back-to-basics reality check back then. Today, after a much needed wake up call, we can probably say SaaS is healthier.
And the companies that get it right might come out 10× stronger.
“Bad companies are destroyed by crisis. Good companies survive them. Great companies are improved by them.”
— Andy Grove
2/ LLMs don’t replace marketplaces — they pressure-test them
This essay from Dan Hock explains where LLMs are a real threat to marketplaces — and where they aren’t.
Marketplaces are most exposed when:
Supply is easy to aggregate (standardized data).
The product stops at search or comparison.
The purchase is high-consideration but low-frequency.
In these cases, LLMs sit in front of the marketplace. They don’t just ‘capture value’ — they break the business model by cutting off the free repeat traffic marketplaces rely on to be profitable.
“Superior search combined with the ability to transact lets AI systems like ChatGPT push further into the marketplace stack than Google ever could — doing much of what transactional marketplaces already do.”
That’s especially bad news for Expedia-style businesses.
Travel is high-consideration (requires research).
Inventory is widely available.
Differentiation lives mostly in search.
Exactly where LLMs are strongest.
Dan’s draws a clear line on how to survive the AI threat:
Go Deep: Manage the messiness of the real world.
Go Frequent: Build habits that bypass search entirely.
The defensible ones:
Underwrite transactions: Manage returns, guarantees, and financial risk.
Own the physical layer: Build logistics and fulfillment (LLMs won’t drive cars or flip burgers).
Drive frequency: If usage is high-frequency and low-consideration (like Uber), users won’t ask a chatbot… they’ll just tap the app.
Build AI search inside the marketplace: Invest in AI within.
If a marketplace is mostly search, AI will eat the surface. If it runs operations or owns the habit, it still has a moat.
3/ Why AI hasn’t really hit Main Street yet
On ‘Mostly Growth,’ I watched CJ Gustafson + Kyle Poyar‘s conversation with Daniel Lang, CEO of Mangomint, on why AI still hasn’t meaningfully landed for SMBs.
His point is simple: Real-world SMBs don’t care about AI. They care about what the product actually does.
"In the real world, like with salons and spas, they don't care what's on the label. As long as you solve a real business problem for them, as long as you deliver value... The real economy on Main Street America, they don't care if it's AI or something else."
Just like you instantly know what ‘Dave’s HVAC’ does from a van, software needs to be clear. It needs to be obvious, not impressive.
Lang argues that most standalone AI wrappers are dead on arrival because they treat the system of record as an afterthought.
Instead, he describes a contextual, ‘layer cake’ approach to vertical SaaS AI:
System of record first — own the core data.
Communication layer next — be where decisions happen (texts, calls, chats).
Agents last — only then does AI become helpful.
Without deep access to historical data and metadata, AI agents behave like a junior employee on day one — OK at the basics, pretty bad at nuance, and incredibly risky in real customer interactions.
Mangomint earned trust before AI, without contracts, data lock-ins, and relentless focus on ‘boring’ fundamentals and high quality workflows. What worked for them might seem contrarian and may not work for everyone, but there are lots of insights and hot takes on this one:
Bonus: In 2026, AI will start to prompt you more than you prompt it
That’s what Gumloop’s CEO Max Brodeur-Urbas posted on LinkedIn this week.
Ps. I’m glad this post got your attention! If you made it this far, please let me know if this is valuable to you. Feel free to send suggestions for future posts. You can find me on LinkedIn.
