FounderDesk — Email your notes. Approve the exact article.Finding a moat: test it before you trust it
November 8, 2026
Finding a moat: test it before you trust it
Features, launch timing and better products all get copied. Edges that compound with use are the ones worth testing for, and this piece sets out how to test them before you have much to lose.
Start with the question, not the list
Most moat discussions offer a menu. I would start with a test. Roger Martin argues that a strategy needs a customer or cost benefit and a barrier to competitive replication S22.
So I would ask two things. What does the customer get that a rival cannot easily give them? And what would make that advantage harder to copy as the business grows?
I would treat compounding with use as a hypothesis to investigate, not proof that I have found a durable moat.
Why I would question the usual early edges
I would not count launch timing as protection. Lieberman and Montgomery's 1988 review, as described by UCLA Anderson, identified problems in pioneering research: samples included followers who entered after earlier pioneers failed, while market share was treated as evidence of success without enough attention to profit S18.
A better product also needs a replication test. Startup guidance warns that competitors can copy or improve it S8. That does not establish how quickly they will do so.
Platform owners can ship features that absorb thin AI wrappers; this is a risk to test for, not a documented pattern in our sources. For a thin wrapper, I would ask what remains valuable if the underlying platform ships this feature.
Fab: a cautionary case, not a moat success story
Copying can also become an expensive fight. The Samwer brothers cloned Fab.com. Fab expanded into Europe, and the clone was eventually shuttered. But Fab burned through most of its capital and was sold in late 2014 for $30 million, after a valuation above $1 billion .
The same account reports that Fab spent $40 million on marketing and lost $90 million in 2012. Its ad-acquired shoppers were less likely than its early customers to buy repeatedly or refer others S25.
I would use this as a prompt to compare acquisition cohorts. It does not isolate a missing moat as the cause of Fab's outcome.
The moats worth testing
Switching costs are the financial and non-financial burdens of leaving, including setup expenses, data transfer, retraining and the risk that a replacement falls short S12. In enterprise software, the friction often comes from integrations and data migration S3S12. I would write down what a customer would lose by leaving, in hours and money, and compare that with what they would gain by switching.
Roger Martin says he does not favour switching costs as a sustainable competitive advantage, because he prefers advantages that give customers something of genuine value S22. My aim would be value worth staying for, rather than making departure painful.
Network effects need more than a user count. NFX attributes 70% of technology-company value creation since 1994 to network effects S14. That is its reported association, not evidence that choosing a network moat improves startup survival. Research on social networks also cautions that structure, including clustering, can weaken network effects. Ease of multi-homing, meaning users on several platforms at once, can significantly reduce the strength of network effects S13. I would ask which users make the product more useful for whom, then test that specific mechanism.
For proprietary data, I would test whether exclusive, high-quality data becomes products, insights or decisions that rivals cannot match. One AI-vendor data-moat guide frames a data moat around exactly that idea S28, and I would treat its criteria as questions to investigate, not a proven checklist. Andreessen Horowitz, writing about enterprise startups, says it now wonders whether data network effects show up in practice at all, and suspects data scale has limited defensive value for many companies S27.
Distribution gets the thinnest support in this piece. I would not call a channel defensible until I had compared acquisition cost and repeat acquisition across channels, and checked whether a rival could reach the same buyers.
The other advantages I would put on the worksheet
To broaden the search, I would add the following hypotheses. These are proposed tests, not a ranking of moat types or claims that any one will work.
Scale economies: What cost per customer or transaction should fall as volume rises? What would stop a rival reaching the same cost?
Distribution: Which channel could become harder for a rival to access or reproduce? I would compare acquisition cost and repeat acquisition across channels before calling one defensible.
Brand: Would customers still choose us if a rival matched the product and price? I would test that preference rather than infer it from awareness.
Proprietary technology: What measurable benefit depends on technology a rival cannot readily reproduce or substitute?
Regulatory barriers: What specific permission or approval would a rival need? I would verify the applicable rules before counting them as protection.
I would keep efficient scale separate from the scale-economies question, rather than use the terms interchangeably. S3 links efficient scale to niche markets with limited room for competitors S3. This worksheet does not establish either advantage for a particular company.
A test you can run this month
1. Write the customer benefit in one sentence, including what you expect it to save them. 2. Name what a well-funded rival would have to reproduce. Record your estimated copying time as an assumption to challenge, not a fact. 3. Pick one candidate advantage and one observable measure, such as repeat use, migration hours or referrals. Write down what result would weaken your hypothesis. 4. Compare cohorts. Fab's ad-acquired shoppers had weaker repeat-purchase and referral behaviour than its early customers S25. I would investigate a similar gap, not treat it as an automatic moat failure. 5. Check your stage. The CRV guide frames moat ratings as a long-term benchmark for seed and Series A companies, not a standard they have already met S9.
I would use these steps as hypotheses to test. The supplied evidence offers no validated retention threshold or switching-cost audit protocol.
What this evidence cannot settle
The supplied material contains no comparative study establishing that startups deliberately pursuing a named moat outperform comparable startups that do not. It also supplies no measured copying window, validated ranking of moat types, or documented successful founder sequence for years one through three.
The requested successful early-stage case study remains an evidence gap. Fab is a cautionary case; I would not relabel it as a success. The broader worksheet above adds questions about scale economies, brand, distribution, technology and regulation, but the supplied excerpts do not establish successful early-stage cases for them.
For the next ninety days, which advantage would you test, what number would you track, and what result would make you change your mind?