In late May 2026, CVS Health invested $40 million in a data startup called H1. The deal turned heads — SaaS startup funding had been visibly contracting in the post-AI-boom environment — but it was something CEO Ariel Katz said while announcing the round that left the deeper impression.
"AI can replicate workflow SaaS.
It cannot replicate H1's proprietary physician data."
That sentence compresses a shift playing out across the entire software industry. As AI tools proliferate, the line between what retains durable value and what doesn't is sharpening fast. And CVS just spent $40 million confirming that the line falls in a surprisingly simple place.
What AI Can and Can't Catch Up To
H1 aggregates information on U.S. physicians and sells it to pharma, biotech, and medical device companies. The product covers individual doctors' specialties, prescribing patterns, academic publishing records, conference speaking history, and influence within hospital hierarchies — all in a structured, queryable format. H1 has sales automation and marketing features, but the company's real pitch has always been the data itself, not those features.
That distinction matters now more than ever. As general-purpose AI models like GPT-4o and Claude have become standard infrastructure, the prediction that "AI will replace most workflow software" has been gradually proving out. Scheduling, email drafting, contract summarization, meeting notes — AI already handles a meaningful share of these tasks. The so-called tool layer is commoditizing fast, and implementing any specific workflow feature now takes a fraction of the time it once did.
H1's physician database sits on an entirely different tier. The prescription histories, academic networks, and hospital decision-making roles of doctors across the United States require years of individual-record collection and consistency validation to build. You can't assemble it overnight. Legal constraints exist. And trustworthiness — the key driver of adoption in clinical settings — takes years to earn. CVS's $40 million was a bet on that accumulation process itself.
Why the Moat Has Shifted from Features to Data
Before AI, software competitive advantage was typically described in terms of feature superiority: a more intuitive UI, deeper integrations, faster processing. That logic started to break down when connecting a general-purpose AI through a single API allowed competitors to replicate a significant portion of any product's feature set within weeks. Whether you're a startup or an independent developer, open-source models now compress timelines that once spanned years into timelines that span weeks.
In business strategy, the standard measure of sustainable competitive advantage is imitation cost — how much time and capital would a competitor actually need to copy your edge? The higher that cost, the thicker your moat; the lower it is, the faster you get undercut. H1's moat is, in practical terms, equal to the age of its physician data. A feature advantage can be replicated in six months. Three years of clinical and prescribing records takes three years.
The same structure is visible in the Korean market. Some real estate platforms earn more trust than major portals for specific use cases — not because of feature differences, but because of the density of transaction data that users have entered and verified over years. Some fashion e-commerce players have survived competition from far larger platforms largely thanks to community-built size data and specialized reviews accumulated over time. When features look similar on the surface, data density changes the actual experience. I don't think this is purely a technology trend story. It's a question of fundamental business architecture.
That Said, "Just Have Data and You'll Survive" Is an Overstatement
This is the place to be honest about the counterargument. H1's logic is compelling — but stretching it into a general rule, "hold proprietary data and you'll maintain competitive advantage in the AI era," introduces some real caveats.
First, data without the capability to use it just sits there. Several large Korean conglomerates hold years of customer data they've never managed to translate into new services or better decisions. Accumulating data and refining it into something genuinely valuable are entirely different organizational capabilities.
Second, the data-moat strategy isn't an immediately available path for someone starting a business from scratch. Replicating H1's multi-year physician database with a small team in a short timeframe is essentially impossible. This is a defensible position for those who already have accumulated data — it is not necessarily an offensive strategy available to someone just entering a market.
Third, as data privacy regulations tighten globally, the cost and legal risk of building and maintaining proprietary data assets rises with them. The model H1 built within the specific regulatory structure of U.S. healthcare does not automatically transfer to other countries or other industries.
Even so, the problem is not one you can simply set aside. When AI is rapidly claiming the tool layer, a business whose entire value proposition lives in features and workflow automation is occupying ground that keeps shrinking.
The Question Left for Solo Operators
When you translate H1's story into the context of independent operators and small practitioners, a few categories are worth examining.
Relationship data. Information about specific clients' decision-making styles, the report formats they prefer, and who actually holds influence within a given industry. If it lives only in your head or scattered across a notes app, it can't be reused on the next project. Documented systematically, it becomes a starting point that competitors in the same sector simply don't have.
Domain-specific case archives. Patterns you've repeatedly observed within a specific industry, region, or client type. Things like "independent cafés under 30 pyeong (roughly 1,000 sq ft) consistently show these characteristics in their P&L during the first three months" or "B2B deals where the approval chain runs more than three levels deep average six additional weeks to close" — that kind of granular, field-verified knowledge does not appear in any general AI training set. As it accumulates, the nature of your evidence base changes, whether you're doing consulting or content work.
Raw customer feedback. Not survey averages — the actual sentences customers wrote. "I got confused here at first, because of this" — when statements like that pile up, you develop a fundamentally different basis for product improvement and marketing direction. Averages obscure signal. Originals restore it.
All of this runs on a single prerequisite: documentation. Even the best relationships and sharpest experience, if unrecorded, can't be reused when a team grows or when AI tools get wired into your workflow. Redesigning how you run your business — treating daily operations as an ongoing data-accumulation process — is what makes this idea actionable rather than abstract.
The logic H1's CEO offered to explain the CVS investment is simple: accumulated data, not software features, is the moat. In the business you're running right now — where is the place a competitor couldn't quickly replicate, even given time? Are you investing in documentation and data at the same rate you're investing in features and tools?
