The AI Productivity Paradox

This week, I analyzed feedback from 800+ companies explaining why they joined our Growth Program. One founder wrote:

"All our growth has come magically and almost nothing we did seems to work."

Another said they were running:

"Organic, App Store Search Ads, Google Ads, Facebook Ads, Affiliate Marketing, YouTube, Snapchat..." but still struggling to scale.

This is the pattern I see everywhere now, amplified by AI tools.

Look, I've been there. When you're just starting a company, you inherently don't know what works. To find out, you need to test. And generally speaking, testing more gives you more chances to unlock a winner.

But there's a difference between systematic experimentation and spraying and praying at 10x speed.

Companies using AI to run 20 tests per month when they used to run 3 end up with a lot of shallow data but very few real deep learnings that inform how they can adapt their strategy and double down on what's working.

Not to mention, even if you can run 20 tests per month:

• Are you adequately budgeting those tests?
• Are you letting them run long enough?
• Are you documenting what you learned?
• Are you building on previous insights?

The answer is almost always no.

​Your Foundation Is Your Multiplier

Think about it this way, if you just got your driver's license, getting a Ferrari won't make you a good driver. It'll just make you more dangerous on the road. And if you're heading in the wrong direction, you're going to end up 100 miles off course, when if you rode a bike, you'd only end up a mile off course.

How I think about it:

• Strong fundamentals × AI = Compound growth
• Weak fundamentals × AI = Compound confusion

Here's what strong fundamentals actually look like. Make sure to have these in place before amplifying your growth efforts with AI.

1. Know Your 5 Fits (Not Just Product-Market Fit)

After working with hundreds of startups at our agency and thousands of startups through our Growth Program, we've identified 5 critical fits. We wrote a whole newsletter about this in May, but it's worth revisiting.

• Market-Product Fit: Your product actually works. People are pulling it out of your hands. They use it and tell friends.
  ​
• Market-Model Fit: Your pricing aligns with how your market buys. (One company we worked with was trying to sell $50K enterprise contracts through self-serve.)
  ​
• Market-Brand Fit: Your brand resonates with your specific audience (e.g., Stripe's technical excellence appeals to developers).
  ​
• Market-Channel Fit: You're where your customers actually make purchasing decisions (e.g., B2B SaaS on LinkedIn, not TikTok).
  ​
• Model-Channel Fit: Your unit economics work in your chosen channels (e.g., $10/month products can't afford $100+ CACs).

Miss any of these, and AI will enable you to spin your wheels faster than ever, until you run out of cash.

2. Implement a Minimally Viable Experimentation Framework

One Growth Program student told us their biggest challenge was...

"Knowing what's working/not working...and consequently, knowing where to focus."

When you're running a startup, it seems so appealing to have all your tests cleanly organized and meticulously tracked.

But like wasting time on worthless AI-fueled tests, overbuilding your experimentation framework can pull valuable time from running the actual experiments.

I can't tell you how much time I've wasted in Notion, Clickup, Asana, etc...

Even a minimum level of organization helps. What matters is learning intentionally to inform your strategy.

This is the difference between testing loops and learning loops:

• Testing loops ask: "Which version won?"
• Learning loops ask: "What did this teach us about our customer?"

You don't need a complex system. Create a simple Google Doc with these questions:

Ideating tests:

• What metric are we trying to improve?
• What are all our hypotheses for improving it? (List them all, even wild ones)
• What evidence or customer insight supports each hypothesis?

Prioritizing tests:

• Which test has the highest conviction based on customer data?
• Do we have enough traffic/users for a meaningful result?
• What's the effort vs. potential impact?

Planning tests:

• What exactly are we changing?
• What does success look like? (Define win criteria upfront)
• What's our budget/time limit?

Learning from tests:

• What happened? (Just the facts)
• Why did it happen? (Your best explanation)
• What's our next test based on this learning?

Not that you need it, but I'm giving you permission to literally do this in a Google Doc. The best system is the one you'll actually use.

Three well-designed experiments that you track beat 20 random (un-tracked) tests, every time.

3. Max Out What Works

Most early-stage startups abandon their channels too early.

I worked with a DTC company spending $6,000/month on ads. They had barely scratched the surface of their serviceable addressable market.

Their CAC was $40, and their unit economics could support up to $80 CAC before they would be unprofitable on every sale.

As they pushed spend above $6,000, their CAC crept up to $50, then $60. They panicked and pulled back.

But here's what they missed. Even at $60 CAC, they were still profitable on each sale. The real questions they should have asked:

• What's our payback period at the higher CAC?
  This is how long it takes to recoup your customer acquisition cost. If you spend $60 to acquire a customer, when do you get that $60 back? For subscription businesses, it might be 3-6 months. For this DTC company, it was immediate on first purchase.
  ​
• Can our cash position handle the float?
  "Float" is the money you need upfront before customers pay you back. Even if you're profitable long-term, you need enough cash to cover acquisition costs while waiting for revenue. A startup with $50K can't float thousands in ad spend if payback takes months.
  ​
• Is our contribution margin still healthy?
  This is what's left after subtracting all variable costs (product, shipping, transaction fees) from revenue. As long as this margin exceeds your CAC, you're adding profit with every sale.

For this company, all three answers were favorable. In other words, they were leaving cash on the table by pulling back ad spend.

Rising CAC isn't always bad. Only start testing new channels when your marginal CAC in the current channel exceeds what you could realistically achieve elsewhere.

Most startups never get there. They see CAC increase 20% and immediately jump to the next shiny channel, spreading themselves thin instead of maximizing what's already working.

Have you actually pushed your best channel to its limits? Or did you just get scared when the numbers changed?

The Bottom Line

Before you add another AI tool to your stack, test another channel, or pull the plug on an existing test, ask yourself:

1. When's the last time you went through the 5 Fits framework with your co-founder or a trusted advisor?
2. Do you have a simple, systematic way to document learnings from tests?
3. Have you maxed out your best-performing channel?

Depending on your answers above, AI may actually be delaying your success.

Focus on your fundamentals first. Build real learning systems. Max out what's working.

Then, and only then, let AI amplify your efforts.

Kevin

​Kevin DePopas​
​Chief Growth Officer