Is your business defensible against AI?

Published on 2025-06-25

I've been doing AI strategy audits for software companies lately, and the same conversation keeps happening.

A founder or CTO comes in wanting to talk about adding AI features to their product. Chatbots, copilots, summarization, the usual stuff. And yeah, those matter. But that's not the real question.

The real question is: could someone use AI to rebuild your product from scratch in a weekend?

Because if the answer is yes, or even "maybe," then adding a chatbot to your existing product isn't going to save you.

The four things AI can't easily copy

After running enough of these audits, I've landed on four dimensions that actually determine whether a company survives the AI wave. Not "thrives" or "leads" or whatever. Survives.

1. Proprietary data. Do you have data that wouldn't exist without your product? Years of user behavior, transaction history, domain-specific annotations? If a competitor starting today could recreate your dataset in a few weeks, you don't have a data moat. You have a database.

2. Relationships and dealflow. Exclusive partnerships, deep integrations into customer workflows, channel partners who sell your product without being on your payroll. The stuff that can't be vibe-coded into existence. If your customers could switch to a competitor in a week, your relationships are shallow.

3. Niche specialization. This is the big one for most SaaS companies. How specific is your product to an actual industry, with actual domain logic, actual compliance requirements, actual edge cases that only someone with years of experience would know to handle? "Small businesses" is not a niche. "Independent veterinary clinics with 2-5 vets" is a niche.

4. Physical and infrastructure moats. Hardware, custom manufacturing, self-hosted inference at scale. Most early-stage software companies score low here, and that's fine. But if you're spending $50K/month on API calls to OpenAI and a competitor decides to self-host, they now have a permanent cost advantage.

Where most companies fall apart

Almost every company I audit has the same weak spot: they're horizontal.

They built a tool that does something useful for a broad audience. Project management. CRM. Analytics. And it works, people pay for it, revenue is growing.

But here's the problem. AI is really good at horizontal. Give Claude or ChatGPT the right prompt and context window, and it can do 80% of what a generic SaaS tool does. The remaining 20% is where your defensibility lives, if you have any at all.

The companies that score well on these audits have gone deep, not wide. They have compliance workflows that took months to build correctly. They have visualizations that are purpose-built for their domain, not another markdown output. They have edge case handling that comes from actually working in the industry, not from reading about it.

If a user tried to replace your product with ChatGPT plus some spreadsheets, what would break? If the answer is "not much," you have a problem.

What actually works

I'm not going to give you a list of five things. That would be too clean. But here's what I keep seeing work.

Build things the AI chat can't replicate. Domain-specific dashboards. Preset reports that save your users 30 minutes a week. Automated workflows that encode best practices from the industry. This is the stuff that makes someone say "they really get our industry," and it's the stuff a horizontal AI tool will never bother to build.

Make your AI learn. Not just globally, across all users. Per account. Per individual user. The longer someone uses your product, the smarter it should get about their specific situation. That's a compounding advantage, and it's switching cost that builds itself.

Stop treating AI as a feature and start treating it as a colleague. The most interesting AI products right now aren't chatbots. They're always-on background agents that run continuously, surface insights, and take action. Think of them as digital employees. The pitch to customers is simple: instead of hiring another person to monitor this, try a digital employee first. That justifies higher pricing and creates real stickiness.

Be where your customers already are. Slack, Teams, email, ChatGPT, Claude. If your customers spend their day in those tools, your product should be accessible there. Don't make people adopt a new habit. Meet them in the habits they already have.

The uncomfortable truth

Here's what I tell founders who are nervous about AI: you should be. But not because AI is going to wake up one morning and replace your company. Because some 22-year-old with Claude and a good idea is going to rebuild 80% of your product in a month and charge half your price.

Your job is to make sure that remaining 20% is so deep, so specific, so embedded in your customers' workflows that it would take them years to replicate. That's defensibility.

The companies that are going to get crushed are the ones sitting on generic software, generic data, and shallow customer relationships, hoping that their head start is enough.

It's not. Not anymore.

If you want to know where your company actually stands, I do these audits. It takes a couple hours and you walk away knowing exactly which walls are strong and which ones are made of paper.

- Matt Ferrante