Platform Risk: What Happens When the Platform Ships Your Feature

Platform risk is the chance that a company your startup depends on (Apple, Google, OpenAI, Meta, Amazon, Shopify, Reddit or X) changes its rules, cuts your access, or ships your product as a free built-in feature. When that happens to an app, developers call it getting "sherlocked." The more you depend on one platform, the larger the risk.

The term comes from Mac history. Karelia Software released Watson, a Mac app for searching web content through plug-ins, in November 2001. In September 2002, Apple shipped Sherlock 3 with Mac OS X 10.2 and similar functionality, and Watson's development stopped by 2004 (Wikipedia). Twenty years later, the pattern is the same, only the platforms now include AI model providers.

Why platform risk matters more for startups

A big company with many products can absorb one platform change. A startup with one product and one channel often cannot.

Platforms have three advantages a startup on top of them never has. They control the default (the app is preinstalled, the answer sits at the top of search, the feature is in the menu). They see demand data (which apps are installed, which products sell, which prompts are popular). And they set the rules, including API prices and what third parties are allowed to access.

Ben Thompson's essay on the Bill Gates Line is a useful frame. Gates defined a platform as something where "the economic value of everybody that uses it, exceeds the value of the company that creates it." When a platform starts capturing most of the value for itself, the room for businesses built on top shrinks.

For a startup, platform risk is a timing problem as much as a strategy problem. The more successful your category becomes, the more attractive it is for the platform to absorb it. Success makes the risk worse, not better.

Types of platform risk

Platform risk shows up in four main forms:

1. Feature absorption (sherlocking). The platform builds what you sell and ships it free, preinstalled or one click away.
2. Access cutoff or repricing. The platform changes API terms, raises prices, or bans your use case.
3. Rule changes that break your economics. A privacy, ranking or fee change makes your acquisition or monetization model stop working, even if no one copies you.
4. Self-preferencing. The platform promotes its own product above yours in search, the store, or the default settings, often using data generated by your users.

14 real examples of platform risk

1. Karelia Watson (Apple). The original case: Sherlock 3 shipped features close to Watson's inside Mac OS X, and Watson stopped development two years later (Wikipedia).
2. f.lux (Apple). f.lux pioneered night-time screen color shifting; Apple blocked its sideloaded iOS version, then shipped Night Shift in iOS 9.3, and the f.lux team asked Apple to open the API (iDownloadBlog).
3. Day One (Apple). Apple's Journal app arrived with iOS 17, preinstalled and able to pull suggestions from data across the phone; Day One's founder confirmed the app was being sherlocked (9to5Mac).
4. Tile (Apple). Before AirTag launched, Tile told Congress that Apple "owns the stadium, the ball, and the league" and had not let Tile use the iPhone's UWB radio (9to5Mac).
5. Tweetbot and Twitterrific (Twitter). In January 2023, Twitter cut off third-party clients and then changed its developer terms to ban building substitutes for its own apps (TechCrunch).
6. Apollo (Reddit). Reddit's new API pricing would have cost Apollo about $20 million a year, and the app shut down on June 30, 2023 (TechCrunch).
7. Zynga (Facebook). Zynga's S-1 said it generated "substantially all of our revenue and players through the Facebook platform" (Practical Ecommerce), while Facebook's own S-1 showed Zynga was 12 percent of Facebook's 2011 revenue (TechCrunch).
8. Snapchat (Meta). Instagram launched Stories in August 2016, and Kevin Systrom said of Snapchat, "They deserve all the credit" (TechCrunch). Snapchat was not built on Instagram, but Meta's larger network made the format its own.
9. Meta (Apple). Even giants carry platform risk: Meta's CFO said Apple's iOS privacy changes would be a headwind "on the order of $10 billion" in 2022 (Macworld).
10. Genius (Google). Genius said Google's search lyrics boxes reproduced its transcriptions and hid "red handed" in Morse code to prove it, but a court dismissed the case because Genius did not own the copyrights (TechCrunch).
11. Yelp (Google). Yelp's 2024 antitrust suit alleges Google places its own local results above organic results, so users rarely reach Yelp (CNN).
12. Amazon third-party sellers (Amazon). A Wall Street Journal investigation reported that Amazon employees used individual seller data to decide which private-label products to launch, despite a policy against it (CNBC).
13. Shopify email apps (Shopify). In November 2019, Shopify launched Shopify Email so merchants could run campaigns "natively inside Shopify," a category previously served by third-party apps (Shopify).
14. GPT wrapper startups (OpenAI). At its first DevDay in November 2023, OpenAI launched custom GPTs, a GPT Store, and an Assistants API for agent-like apps that retrieve knowledge and run code (TechCrunch), covering ground many thin AI startups were selling.

The pattern across these cases: the platform rarely needs a better product. It needs a good enough product in the default position.

What happens when OpenAI ships your feature

AI startups face the fastest version of platform risk. Model providers ship new capabilities every few months, and each launch can turn a standalone product into a checkbox inside ChatGPT, Gemini or Claude.

The test is simple. If a user could get 80 percent of your value by typing a good prompt into a general assistant, your product is exposed. If your value comes from proprietary data, a regulated workflow, integrations into systems the model provider does not touch, or a customer relationship you own, the launch is less threatening, and sometimes helpful, because it makes your underlying model cheaper. This overlaps closely with our AI moat test.

How a startup reduces platform risk from zero

You cannot remove platform risk entirely; almost every product runs on someone's cloud, app store or model. You can make sure no single platform could end you with one decision.

• Multi-home from early on. Support more than one platform: iOS and Android and web, more than one model provider, more than one sales channel. Zynga's problem was not Facebook; it was having nearly everything on Facebook.
• Own the customer relationship. Collect the email, run the billing, keep the account on your side. If users only know you as a tile inside someone else's interface, the platform holds the leverage. See owned customer relationships.
• Own the data. Build on data that normal use of your product creates and the platform cannot see. A platform can copy a feature; it cannot copy years of outcome data it never collected.
• Go deep, not wide. Platforms ship broad, generic features for hundreds of millions of users. A product that handles one industry's edge cases, compliance and integrations is a poor target for a general feature.
• Pick markets too small or too regulated for the platform. Apple will build a journaling app. It is far less likely to build dental practice billing, insurance claims software, or a compliance tool for a specific regulator.
• Avoid features the platform needs for its own strategy. If your product sits on a path the platform's roadmap must cover (search, messaging, photos, core AI chat), assume it will be covered.
• Read the terms as strategy documents. Twitter's 2023 terms explicitly banned substitutes for its own apps. Any clause that lets a platform reprice or revoke access is a risk you are carrying.

For a broader view of how rivals with more money attack a product, see copycat resistance.

Common mistakes and fake versions of platform independence

• "We have a partnership." A partner manager or a featured placement is not a contract that prevents the platform from competing with you. Apollo was featured at WWDC 2023 weeks before shutting down.
• "They would never bother." Platforms bother when a category gets big. Your own growth is the signal that attracts them.
• "We're multi-platform" when 95 percent of revenue comes from one. Diversification counts by revenue and users, not by logos on the homepage.
• "Our UX is better." Sometimes true, and sometimes enough for a niche of power users, but a default free feature usually wins the mainstream.
• "We'll build a moat at scale." If the platform ships the feature before you reach scale, the future moat never arrives.

How Moat Check scores it

The Moat Check methodology asks one question for this test: "Is it safe from Apple, Google, OpenAI, Meta, Shopify or another platform shipping it as a feature?"

The reasoning is direct: products built on a platform can be replaced when the platform ships the same feature. A pass needs to be true today or by design; "once we're big" is a partial at best. Moat Check uses AI and web research to apply this framework and names the company most likely to copy your idea, often a platform. Treat the result as a starting point for your own thinking, not investment advice.

Platform risk is one of ten tests. Read the full list on the moats hub, or start with what a moat is.

Check your startup's moat

Questions

Keep reading

• AI Moats: What Survives When Everyone Has the Same Model
• Network Effects: What They Are, the Main Types, and Real Examples
• Atomic networks: win a small market before you try to win a big one
• All 10 types of moats
• What is a moat in business?