Companies are about to waste billions on AI – here’s how not to become one of them.

“This is venture money, not adventure money.” This was the loving response a dear friend got from a VC when he pitched an idea. But when we’re in the hype cycle phase of a new technology, that caution goes out the window. After all, VCs have to invest all the capital they’ve raised, and the cost of missing something big outweighs the downside of swinging and missing, especially when everyone else is taking the same swing. .

A similar dynamic plays out inside most companies — and the technology of the moment is AI and anything remotely related to it. Large Language Models (LLMs): They are AI. Machine Learning (ML): This is AI. That project you’re told has no funding for every year — call it AI and try again.

Billions of dollars will be wasted on AI over the next decade. If it sounds counterintuitive, it shouldn’t be. Every big technology wave comes with excitement—before we know how real and transformative it is. Search, social and mobile have all had a broad and lasting impact, but virtual reality (VR) and crypto have been much more limited.

Although, you wouldn’t know it from reading the headlines from five years ago. Right now, everyone is rushing to show how much they are spending on AI and how it will change everything. This shotgun approach to investing inevitably results in some big wins and a lot of misses. For VCs, the same dynamic that drives companies’ leadership to greenlight investments in the name of AI is optimistic, idealistic, misguided, and adventurous more often than not.

This does not take away from the fact that LLMs are a game-changing technology. Just look at how much faster ChatGPT reached 100 million users than other conversion companies:

Almost every enterprise company has some work to leverage LLMs and AI. So, how should you decide where to place your bet and where you are entitled to win?

Keep a clear eye on these three things, and you’ll cut 80% of wasted costs:

1. understand the total cost over time;
2. Ask why someone else can’t do the same.
3. Set a few conditions that you want to follow.

1: Understand the total cost over time.

As you think about saying yes to that next AI project, consider the cost of resources required to sustain that project, today and over time. Your data science team’s ten hours of work is often 5X the time of Engineering, DevOps, QA, Product and SysOps. Companies are littered with pieces of projects that were once a good idea but lacked the ongoing investment to sustain them. It’s hard to say no to an AI initiative today, but too many yes’s often come at the cost of fully funding the few things worth supporting tomorrow.

Another cost dimension is the incremental marginal cost that AI drives. These large models are expensive to train, run and maintain. Overuse of AI without a corresponding increase in downstream value chews up your margins. Worse, pulling back released or promised functionality can lead to customer dissatisfaction and negative market perceptions, especially during hype cycles. Look at how quickly a few mistakes have tarnished Google’s reputation as an AI leader, not to mention the early days of IBM’s Watson.

2: Ask why can’t someone else do it?

It is easy to forget what you learn from textbooks. We have all read about commoditization. The same lessons learned from knocking in real life stick with you. When I worked as a chip designer at Micron, our primary product was close to the perfect thing – a memory chip. No one cares what brand of memory chip their laptop has, how much it costs. In this world, scale, and cost are the only sustainable advantages over time.

The tech industry can be bimodal. There are monopolies and commodities. When you say yes to the next AI initiative, ask yourself, “Why us?” Working on something that gets commoditized over time is no fun, especially when you don’t have a scale/cost advantage. take it from me The only ones that will definitely benefit are Nvidia and AWS/Azure. The only way around this is to focus on something where you have a defensive moat. Priority access to data, proprietary insights around a use case, or an application with strong network effects is where you start.

3: Place a few bets you want to see.

The easiest bets are the ones you already have in your business. BASF’s old commercial comes to mind: “We don’t make the things you buy, we improve the things you buy.” If applying AI gives you speed in the products you’re already building, it’s the easiest bet to make and scale. The other easiest conditions are those that allow you to move up and down the value chain or later expand to other sectors.

The most difficult but important prerequisites require you to disrupt your existing business with new technology—if you don’t, someone else will. Double down on a handful of bets that pass these two tests, and be prepared to watch those bets. Leave the rest to VCs and startups.

So while the hype around AI is real and legitimate, if there’s one lesson we’ve learned over the years, it’s that with these cycles comes not only good investment, but also a lot of waste. By following a few of the tips outlined above, you can ensure that your investment has the best chance of bearing some algorithmic fruit.

Mehul Nagrani is currently the Managing Director of North America.