How to Tell If a Startup Has a Moat

A three step test for whether a startup, or your own product, has a real compounding advantage.

CheatsheetLayout, Section, NumberedItem, ExampleSection, Callout, } from "@/components/CheatsheetLayout";

Step 1: find the compounding advantage

Speed, distribution, brand, a strong team and lots of funding are fuel, not moats. A compounding advantage is a self reinforcing loop that gets stronger with every turn. Only four matter, and every major tech company has at least one.

Switching costs

The deeper a product is embedded in how a customer works, the harder it is to remove. Watch for a roadmap full of integrations and new use cases. Rippling started with onboarding and expanded across HR, payroll, finance and BI on shared infrastructure.

Network effects

Each new user makes the product better for everyone, and features matter far less than the network (Craigslist still pulls traffic in 2026). At Faire, most of the roadmap was more retailers, more brands, better discovery, and less risk for retailers trying new products.

Scale economies

A big upfront investment (a factory, a warehouse, an LLM training run), then volume drives unit cost down. Zipline's cost per delivery has fallen from $300 at launch to $12.

Data effects

Usage generates data, data improves the product, a better product drives more usage. Stripe Radar has seen over 70 trillion data points and $1.9 trillion in volume; 92% of transactions use a card it has already seen. Ignore the strategy deck. Look at what the team ships every day and which teams are largest. A compounding advantage cannot be bolted on; it is the core loop of the business.

Step 2: gauge how strong it could be

Every marketplace has some network effect, and most still fail. Powerful advantages pass three tests.

High value: the benefit is why customers choose

ServiceTitan wins because small businesses want scheduling, dispatch, invoicing and payments in one system. Zoom keeps adding chat and docs, but that is not how people pick video conferencing. Costco holds gross margins around 11% because value for money is what members come for; Salesforce runs at 78% because price is not how its customers choose. One question per loop: Switching costs: "Would customers still choose the product if it only did one thing?" Network effects: "Would the product still be worth using if you were the only user?" Scale: "Is value for money the main reason customers choose the product?" Data: without data from other customers, would they still buy?

High headroom: the benefit keeps improving

How much scale before costs stop falling? How much data before the product stops improving? The longer the curve, the harder it is for a rival to catch up.

High durability: no shortcuts for a competitor

Can a rival buy the advantage (LLM training data is largely public, and Mercor and Handshake sell the rest), borrow it (multi tenanting, like drivers running Uber and Lyft at once), or bypass it (Evernote had 250 million users until Notion built an importer that moved everything in a few clicks)?

Step 3: measure whether it is working

By around Series B the advantage should be materially improving the business. For each loop ask for one input metric (is the advantage itself growing?) and one output metric (is the business performing better because of it?). Absolute values matter less than direction. The tell is metrics that improve over time as the loop compounds. In most businesses later cohorts activate worse as the product reaches a broader audience. In a marketplace with a strengthening network effect the opposite happens: newer cohorts activate better and spend more. If nothing shows up, either there is no real path to an advantage, or the team is not working on the right things.

What working looks like: four public benchmarks

Every extra product adopted makes Datadog harder to rip out. Since 2020 the share of customers using four or more products has more than doubled, and gross retention is in the mid to high 90s. More riders attract more drivers, which means shorter waits and better prices. Since 2021 monthly users are up 70%, and each one takes 25% more trips. More driverless miles make the cars safer. From June 2025 to June 2026 Waymo nearly tripled miles driven and is now 95% less likely than a human driver to be involved in a serious crash. Content costs the same whether 10 million or 100 million watch. As membership doubled, content costs fell from 46% to 36% of revenue and operating margins more than doubled. Treat these as the shape of a healthy curve, not as targets. The source's appendix adds rough "good" and "great" ranges, with a caveat not to get hung up on them.

How this applies to AI products

Data effects are the moat AI makes common. Cursor gets better with every accept or reject, and vertical AI in law, healthcare and customer service can do the same. Historically the rarest loop, now the first one to look for. Be wary of "multiplayer" features on an AI product. Canva has sharing, but that is not why people pick it over Figma. Unless interacting with other users is the main reason customers come, it is a feature, not a network effect. AI lowers migration cost, not switching cost. The bigger cost is the workflow tax: the time a team spends learning a product and building it into how they work. AI also lets teams ship across more surfaces, so embedding can get deeper. Durability is the weak point for frontier models. Much training data is public and the rest is for sale, which thins the data effect. Scale now includes the frontier labs. A training run is the same kind of capital bet as a factory, and the same value for money test applies.

Questions to ask, and how to read the answers

After an offer, sit with someone who knows the numbers. In your own product review, ask the same questions of yourself. Which of the four loops is the company actually building, judged by roadmap and team size, not the pitch? Would customers still buy if the loop's benefit disappeared tomorrow? Where does the curve flatten: how much more scale or data before the benefit stops improving? Can a rival buy, borrow or bypass the advantage? What is the one input metric and one output metric for the loop, and are both improving cohort over cohort? If it is not in the numbers yet, how does leadership describe the path to defensibility, and does it show in daily priorities? All three boxes ticked is a strong buy signal, worth more than a better title or a bit more comp. None ticked, with the team talking mostly about brand, traction or hard work, is a red flag. The messy middle is where the questions above earn their keep, and the conversation usually breaks the tie. For the wider map of where AI moats come from, see AI Strategy and Competitive Moats for PMs. A lottery ticket's odds never change. A startup with a compounding advantage gets better odds every month, and if it is real, the metrics will show it. Source: Dan Hockenmaier with Lenny Rachitsky, "How to tell if a startup has a moat", Lenny's Newsletter (Oct 2026). );