Markets 
Last week a Blackstone-led group took control of Medallia from Thoma Bravo, and Thoma Bravo wrote off close to $5 billion. By one widely cited count it is the second-largest loss in private equity history. Thoma had taken the customer-experience software company private in 2021 at $6.4 billion, near 9x forward revenue, on cheap debt. Orlando Bravo has since called it a “big mistake” and conceded the firm bet on fast growth that never showed.
Medallia is not a one-off. What is the driver?
Rates are the usual explanation. PE bought software at pandemic-peak multiples, loaded with debt, underwritten on growth that assumed the world of 2021 would hold.
But for me, the question this and other transactions trigger relates to AI disruption, and enduring moats.
Look at Chegg, which had no buyout and no leverage to pin it on. The homework-help company was a profitable public business worth about $14 billion in 2021. It had spent a decade paying contractors to build a library of tens of millions of step-by-step answers, then rented that library to students. ChatGPT arrived, free and better.
Within months the CEO was telling analysts the chatbot was hurting growth. Google’s AI summaries then choked off the search traffic that fed new sign-ups. By late 2025 Chegg had cut almost half its staff, citing the “new realities of AI”, and its market value sat barely above $100 million.
That is the variable the 2021 underwriting missed. When the cost of generating an answer, a summary, a design, or a line of code falls toward zero, a recurring-revenue moat built on producing that output stops holding.
The market has named this the Saaspocalypse, though I think this is too generalized a phrase.
AI does gut some categories. Customer-experience and feedback software is squarely one, which is why Medallia’s closest comparable, Qualtrics, went private too, in a $12.5 billion Silver Lake deal in 2023. But “AI kills SaaS” is not a law. The useful question is which software keeps its moat once generation is free.
For fintech, my view is that durable businesses tend to hold some mix of three advantages I have called the 3Ds: distribution, data, and delivery. AI does not retire the framework. However, it cuts each leg in half. The replicable half, the part a frontier model can now do, stops being a moat. The half AI cannot reach gets more valuable.
Most of the job now is telling the two apart.
Distribution
Distribution is the leg that strengthens. When anyone can ship a capable product over a weekend, building is no longer scarce and demand is the whole fight. The forms that last are the ones a model cannot conjure: sitting inside a system of record a customer already runs, holding a license or a payment rail a competitor cannot cheaply copy, owning the trust that lets a stranger move money through you.
Chegg’s distribution was Google referrals. When Google changed how it answered queries, that channel went dark. Rented distribution disappears the moment the channel reprices.
One complication: the buyer itself is starting to change. Over the past year Visa, Mastercard, Stripe and Google have all shipped rails for payments made by AI agents, and Visa now frames the coming year as the one where agents stop assisting purchases and start completing them. McKinsey puts the potential U.S. market at up to $1 trillion by 2030. When an agent does the buying, the things you tuned for a human shopper matter less: store ranking, paid traffic, a frictionless signup the agent will never see.
Therefore, new questions are required: Are you the option the agent picks? If so, why? Do you own the rail it settles on? Are they protected?
Data
Data is where founders fool themselves most often. A pile of historical data is less and less valuable.
But in fintech or health, what survives is a live, exclusive, usually regulated flow you keep generating and no one else can buy. Lending off a borrower’s own transaction history is a real moat: proprietary, real-time, consented. Local fraud and credit patterns tied to one market hold up too, because a foreign model cannot cheaply acquire them. The question is not how much data you have. It is whether you sit on a flow that compounds.
The fintech line here is fairly clean. Proprietary data still beats a general model in credit and fraud. It also has to be built inside systems of work that today can’t be done by foundational models alone.
Ask yourself, which side your product sits on, and whether it is a business AI rebuilds from one it leaves alone.
Delivery
A delivery advantage can take many forms.
Sometimes it is about delighting a customer in a unique way. Sometimes it is about having a unique delivery path, like being embedded in the payment flow, obfuscating what was once a complex experience (itself a delight arguably).
This too is changing. Delight is now cheap; an AI-native rival can stand up a polished, focused product in any market within weeks, so a great experience is table stakes rather than a moat.
The durable part of delivery is what sits beneath the delight: trusted, accountable execution. Money actually settles, compliance holds, and someone is liable when the model gets it wrong.
This is one of the reasons I’ve been focused on trust and verification as an example.
Own the flow
Taking a step back, in many ways the conclusion of the above is the importance of “owning the flow” as an enduring moat.
Sit inside the transaction and distribution, proprietary data and accountable execution arrive as a set. Distribution is accelerated. The flow generates the data, and the flow is where settlement actually happens. Rent the flow instead, as a thin layer on someone else’s rails, and you hold none of it. You are a feature, priced at zero the day the engine beneath you improves.
In a market where a hundred teams chase the same idea in San Francisco and three chase it in São Paulo or Jakarta, the ones that own their flow, in places where the data and the trust are anchored to the ground, are the businesses AI makes stronger instead of obsolete.
Read the full article here
