Everyone seems to think AI is a magic wand you can wave at a pile of data to print money. I get it. As an investor in companies like Anthropic and Scale AI, I see the incredible potential every single day. But as an operator who has built and sold companies, I’ve also seen the ugly side. The side that isn’t in the glossy brochures.
I remember one of our early projects at RemoteTeam. We were building a predictive model to identify which new sign-ups were most likely to become long-term, high-value customers. We spent a quarter and nearly $100,000 building what we thought was a masterpiece. The model was complex, the platform was expensive, and the dashboard looked incredible. The only problem? It was dead wrong. The predictions were no better than a coin flip. We had to scrap the whole thing.
That failure was painful, but it taught me a series of lessons that have been worth millions. After drowning in messy datasets for over three years, we eventually cracked the code, boosting our predictive accuracy by over 40%. It wasn’t about finding a secret algorithm or a better tool. It was about embracing a few brutal truths that nobody likes to talk about. Here they are.
1. Your Data Is Probably Garbage
This is the biggest and most painful truth. We all like to think we have good data. We don’t. It’s a mess of inconsistent formats, missing values, duplicate entries, and hidden biases. Before you even think about an AI model, you have to become a data janitor. It’s not glamorous, but it’s where the war is won or lost.
I’ve seen startups burn through their seed funding because they built their entire model on faulty assumptions baked into their data. At MovieLaLa, we wanted to predict which upcoming films would be box office hits. We pulled data from a dozen different sources, and it was a disaster. One source used “Star Wars: The Force Awakens” while another used “Star Wars Episode VII.” Some budgets were in Euros, others in dollars. Release dates were in different formats. It was chaos.
We spent two full months doing nothing but cleaning and standardizing that dataset. It felt like we were falling behind, but it was the most important work we did. The old 80/20 rule of data science is real—you will spend 80% of your time cleaning data and only 20% analyzing it. If anyone tries to sell you a tool that magically cleans your data with one click, run away. It doesn’t exist. You have to do the hard, manual work of understanding what you have before you can ask it to tell you the future.
2. “More Data” Is Not Always the Answer
The prevailing wisdom is that more data is always better. Just feed the beast, and the AI will figure it out. This is a dangerous oversimplification. Sometimes, more data just adds more noise, making it even harder for the model to find the signal.
I once advised a startup that had collected terabytes of user interaction data. They had every click, every scroll, every mouse movement. They were convinced the secret to reducing churn was hidden in that mountain of information. But their models were useless. The sheer volume of irrelevant data was overwhelming the few important signals.
We took a different approach. We ignored the massive dataset and instead conducted a few dozen in-depth interviews with customers who had recently churned. We built a small, high-quality dataset based on their direct feedback. A simple model built on that tiny, qualitative dataset immediately outperformed the complex one trained on terabytes of noise. The lesson was clear: data quality and relevance beats data quantity every single time. Don’t ask how much data you can collect. Ask what is the right data to answer your specific question.
3. The “Black Box” Will Burn You
Some of the most powerful AI models are also the most opaque. They can give you a startlingly accurate prediction, but they can’t tell you why they made it. For a scientist in a lab, that might be acceptable. For a business leader, it’s a dealbreaker.
If your model tells you to double your marketing spend in a certain region, you need to know why. Is it because of a competitor’s weakness? A change in local demographics? A seasonal trend? Without the “why,” the prediction is just a gamble. And as an angel investor, I don’t bet on gambles; I invest in conviction.
This is why I’m so passionate about the work being done at companies like Anthropic. They are focused on building interpretable AI. We need models that can explain their reasoning. At RemoteTeam, we refused to deploy any model we couldn’t explain. We favored simpler, more transparent models like logistic regression or decision trees over complex neural networks for most business-critical decisions. A model that is 85% accurate and fully transparent is far more valuable than a model that is 95% accurate but operates like a black box. The moment you can’t explain a decision to your board or your team, you’ve lost control of your business.
4. Your Team’s Skills Matter More Than Your Tools
Every week, there’s a hot new AI platform that promises to solve all your problems. Companies spend millions on these tools, thinking they are buying a competitive edge. They’re not. The real edge is in the people you hire.
I would rather have one brilliant data scientist with a laptop and open-source tools than a team of twenty junior analysts with the most expensive AI suite on the market. A great data scientist understands the business context. They know how to frame a problem, how to interrogate the data, and how to question the results. The tool is just a means to an end.
Your budget should reflect this. Instead of spending $500,000 on a new platform, spend $200,000 to hire a top-tier data scientist and give them the freedom to work. Invest the rest in training your existing team. Teach your product managers, marketers, and finance people the basics of data analysis. When your whole company speaks the language of data, you create a culture of inquiry and evidence-based decision-making. That is infinitely more powerful than any tool you can buy.
5. AI Is a Process, Not a Project
Too many companies treat AI analytics as a one-and-done project. They build a model, launch it, and then move on to the next thing. This is a recipe for failure. The world changes, customer behavior changes, and your data changes. Your model will start to decay from the moment you deploy it. This is called model drift, and it’s inevitable.
Building an AI-driven organization means committing to a continuous process of monitoring, retraining, and refining your models. You need a system in place to track your model’s accuracy in real-time. You need to have a schedule for retraining it on new data. And you need to be constantly experimenting with new approaches.
At my portfolio companies, we treat our models like products. They have a product manager, a roadmap, and a regular release cycle. We are constantly working to make them better, more accurate, and more valuable to the business. It’s a long-term commitment, not a short-term fix.
The Real Work
Building a business that truly runs on data and AI is hard. It requires a cultural shift, a focus on fundamentals, and a willingness to do the unglamorous work. It’s not about chasing the latest hype. It’s about cleaning data, asking hard questions, investing in people, and committing to a process of constant improvement.
It took us years of painful mistakes to learn these lessons. But they are what allowed us to build models that didn’t just look good on a dashboard, but that fundamentally changed the trajectory of our business. Don’t fall for the hype. Focus on the brutal truths, and you’ll be on the right path.
Frequently Asked Questions
How do I know which items apply to my situation?
Start by honestly assessing where your biggest bottleneck is right now. The items that address that specific constraint will give you the highest return on your time and energy.
How were these items selected?
Each item on this list comes from direct experience, either from building my own companies or from patterns I've observed across the 200+ startups I've invested in. I prioritize practical, actionable items over theoretical concepts.
Which item on this list has the highest impact?
It depends on your stage and context, but in my experience, the items near the top of the list tend to have the broadest applicability. That said, sometimes the less obvious items create the biggest breakthroughs for specific situations.