How to Manage Products in an AI Frost
As hype wane and costs increase, product leaders should achor to ROI and efficacy
Lately a skeptical note has been introdced into the business conversation around AI. A version of “is all this AI stuff actually worth it?” The conversation has especially centered on rising AI costs; people trading stories about companies burning eye-watering sums on tokens, wondering out loud whether we’ve leaned in too hard. That’s the start of a clear shift… As bills climb and returns stay fuzzy, the questions get sharper and we have to lean into demonstrabable efficacy, proof points and ROI. Before we look at how to adjust to a frostier AI product climate, let’s look at the pattern in more detail, because it’s happened before, multiple times.
In the early 2000s, when I was still in academia, we were careful not to call our work “artificial intelligence.” The term still carried baggage from earlier cycles of grand promises that collapsed, so we said “machine learning” or “statistical methods” and let the results speak. I started building AI products in 2011. By the middle of that decade, the caution was gone and the claims were loud again.
One whitepaper from around 2015 has stuck with me as the high-water mark. A cybersecurity company was pitching its product as a kind of biological immune system for the network, and the technical section piled up every impressive term available: recursive Bayesian estimation, particle filters, sequential Monte Carlo, stochastic networks rewiring themselves over time. It was a wall of sophisticated-sounding mathematics. What it never did was explain, in plain language, what the product did for a customer or why those particular methods were right for the problem. The vocabulary was meant to signal mastery, but to me it signaled the opposite.
Then the mood changed. Starting around 2016, buyers got harder to convince. The questions got sharper. Prove it. Show me it works on my data. The companies running on a good story and a big marketing budget started to struggle, and a number of them folded or pivoted. The buying cycle got longer and more demanding.
That same chill is what we seem to be starting to feel now, so it’s worth being precise about what kind of cold snap it is, because there have been two different kinds in AI, and the impact on product development is very difficult.
Two kinds of cold
The severe kind is a true AI winter. We’ve had a couple of those. Most recently, in the late 1980s and into the 1990s when the technology of the day, expert systems, couldn’t deliver what it promised. The disappointment ran deep enough that research funding collapsed, companies shut down, and the survivors rebranded their work to get away from the word “AI” entirely. The field went quiet for years. That’s a capability collapse. The cold comes from the technology failing to do the job, in a dramatic fashion. For example, the product INTELLECT, released by Artificial Intelligence Corporation in 1981 was pitched as a way you could “talk to your computer” to get whatever data you needed. IBM licensed the technology, got 1,700 trial installations, and a sales close rate of only 1%. The technology just wasn’t there (forms were way easier) and people gave up trying for years (we’ve arguably only really cracked this use case recently with GenAI).
The milder kind is a buyer frost. That’s what set in around 2016. The research never stopped. The building never stopped. What stopped was the buyer patience for vague and unprovable claims. Across tech, the funding environment got pickier, the marketing got quiet, and weak companies washed out. The grounded ones kept right on growing. DataRobot, Dataiku, and H2O.ai, for example, were all building real machine learning platforms through that stretch. DataRobot raised straight through 2016 and reported triple-digit revenue growth year after year; Dataiku and H2O.ai kept raising and shipping too, and all three grew into substantial companies. A frost clears out the field. It doesn’t kill the crop.
Today seems more like a frost
The big reason I believe this is that AI today is delivering business value - it’s just not as deep and wide as the hype would have you believe.
That wasn’t true in the 1980s, which is why that correction led to a true AI winter. Today, the results people are getting from AI are real (even with the older tools used in most publishd research), and in some cases they are stunning. I’ve personally seen teams compress work that used to take weeks into hours with a reasonable level of quality (in an environment of solid systems and infrastructure). There are clearly products that can do things that were simply not possible a few years ago. The fundamental capability is not in question (though it is argubly over its skis in some cases). What is in question is whether a given company can turn that capability into a product its customers will pay for and be able to rely on over time. That’s a very different problem.
So the skeptics asking “is this worth it” are half right. They’re right that a lot of AI spending hasn’t produced clear value. They’re wrong if they throw the baby out with the bathwater. The technology is the most capable it has ever been. The gap for products is in execution, in positioning, and in proof.
How should products position in a frost?
The response to a frost is the same set of habits that make you stronger in any market. Buyers in a frosty AI market want proof, demonstrable value to a problem they really care about, reasonable total costs of ownership, and reliability.
Know your customer and their problems. In a frosty market, if buyers don’t have a problem they really need solved, they won’t buy your product. The “it’s got AI” pitch just stops working. Buyers don’t care. If you know your customer and what they need, you can still use AI where it actually makes sense and it becomes a selling point again.
Use the right tool for the job. When AI is expensive (as it is becoming again), you should only use it when it’s actually needed. Many problems don’t actually require AI - they can be solved with workflows, chron jobs, and just plain old software. If you are thoughtful about the problem you’re solving and the needs of your customer, you can blend traditional technologies with AI in a way that can dramatically reduce TCO and strengthen your value proposition.
Think about demonstrating ROI. If you’ve done the first two things on this list, then all you need to do is measure. How are you solving the problem? How is it faster, better, more efficient than the other products they can buy to solve the problem? If you can demonstrate that, your product will sell.
Stay grounded and don’t be afraid to say what your tech can’t do. A big thing that happens at the height of the hype cycle is that people stop talking about the limitations of their products. A lot of silver bullets emerge and fly blindly around. Trust in frostier times is built by being clear about what you can and can’t do. So be clear and keep your feet on the ground.
Don’t get spooked. If you have a solid understanding of your customer, a great solution to their core problem, and you know your solution benefits from AI, then go forth and don’t be detered. You can build a great AI company even in a frosty moment if you stick to product fundamentals.
The through line
No matter what happens to the AI market, this technology is useful and has a lot of applications in business products. In the past, work has continued through frosts and freezes. Great companies have been built even in times of buyer chill. Across my career, there has been a massive increase in AI capabilities that has not tracked exactly to the business hype cycles. So, stick to your guns, know your customers, and build tech that is useful, maintainable and makes sense.