A health AI startup had two U.S. patents, a published study proving it saved 11 minutes per visit, real paying customers, and seven years of work behind it.
One of its best customers still copied it in a few weeks. No patents. No knowledge graph. Just a basic form and some backend logic.
The startup shut down five months later.
What happened
Cydoc was founded in 2018 by Rachel Draelos, an MD/PhD candidate at Duke who'd watched doctors ask patients the same repetitive questions during clinical rotations. The idea: an AI system that automatically took a patient's history before a doctor walked in, saving real time on every visit.
Over seven bootstrapped years, Cydoc built something genuinely impressive — a HIPAA-compliant platform across web, iOS, and Android, powered by a proprietary knowledge graph that could add an entire medical specialty's worth of questions in hours. It won two patents, a research grant, and four paying customers, including a telemedicine urgent care practice that loved it.
Then that same urgent care practice built a stripped-down version of Cydoc's core feature themselves — hardcoding the questions into a plain form with some text-generation logic behind it — and quietly stopped needing Cydoc. Draelos shut the company down in August 2025.
Why it actually happened
Draelos wrote her own detailed postmortem afterward, and named this failure precisely: "Cydoc's 'moat' was more of a rain puddle."
Here's what that means, underneath the joke.
Cydoc's technology was genuinely hard to build. The knowledge graph, the dynamic question generation, the graph-traversal algorithms — all real engineering, all patented. But the customer was never paying for any of that. They were paying for one outcome: a faster patient intake with less typing.
The urgent care practice didn't need to replicate Cydoc's sophisticated backend to get that outcome. They only needed to replicate enough of it — a form, some hardcoded questions, a bit of text logic — to make the sophistication irrelevant. Their version wasn't as elegant, wasn't extensible to other specialties, wasn't defensible the way Cydoc's was on paper. It didn't need to be. It got the doctor 80% of the value with none of Cydoc's engineering.
This is the trap that catches technical founders specifically, and catches them precisely because they're good at building things. "Hard to build" and "hard to copy" feel like the same fact. They aren't. Something can take years of real skill to build for you, while being trivially approximable by anyone who's only seen the output — because they never had to solve it the hard way. They just had to solve it well enough.
The founder principle
A moat protects the method. A customer only ever pays for the outcome.
The moment someone else can deliver most of that outcome without your method, your method stops being a moat — no matter how hard it was to build, how many patents cover it, or how proud you are of the engineering underneath it.
The Rain Puddle Test
Named for Draelos's own phrase — three questions to check if your technology is a real moat or a rain puddle:
Outcome vs. method. What is your customer actually paying for — the specific way you solved this, or the result they get? Be honest; these are usually not the same answer.
The copy test. Could someone with basic tools — a form, a spreadsheet, an intern, a weekend — approximate 80% of your outcome without touching your actual technology?
Name the real moat. If the answer to #2 is yes, your technology isn't your moat. Something else has to be: proprietary data nobody else has, a distribution advantage, a regulatory approval, a switching cost you deliberately built. If you can't name it specifically, you don't have one yet.
Do this today
Write one sentence describing what your last paying customer is actually paying for — not what you built, what they get. If you're not sure, ask them this week; don't guess.
Try to "fake" your own product with the crudest tool available — a form, a spreadsheet, a basic script — and time how long it takes. That's roughly how long a motivated competitor, or your own customer, needs before they stop needing you.
If your technology fails the copy test, list what would actually be hard to replicate. Data you're the only one with. A relationship. A regulatory approval. A switching cost. If the list is empty, that's more urgent than your next feature.
Before you add a sophisticated feature to defend a weak moat, ask who it's actually for. Does it make the outcome better for the customer, or does it just make your solution harder to copy on paper? Only the first one is worth building.
Look at your last lost customer, or your biggest churn risk, honestly. Did they leave because a competitor out-built you — or because they realized they could get most of the value without you at all?
Closing thought
Two patents didn't stop a competitor with a basic form and an afternoon. Patents protect how you built something. Customers only ever pay for what it gets them — and the moment someone else can get them most of that without your engineering, the rest of it isn't a moat. It's just the bill nobody else has to pay.
Reply and tell me: if you ran the copy test on your own product right now, honestly — how many weeks would it survive?