Past The AI Hype: When Is AI Actually Profitable?

One of the most important value drivers of machine learning and artificial intelligence (ML/AI) technology isn’t the technology itself. It’s where a company deploys it. Our company’s experience is that when investors realize they lack the technical skills to evaluate ML/AI technology, they tend to throw in the towel and take a target’s inflated claims at face value. That’s a mistake. The good news is that it only requires general business knowledge to assess one of ML/AI’s main value drivers: profit proximity.

Profit Proximity

Turns out, the main value of ML/AI comes from which type of business problems ML/AI is applied to. I’ll cover a few categories here and explain how each differs in its proximity to profit, which is what really drives value. There are three profit proximities: The quickest payoff is when ML/AI is applied to customer behavior. Medium-term payoff typically applies when ML/AI improves a product that makes a user’s life noticeably easier. Lastly, the longest-term payoff applies when ML/AI’s largest effect is reputational.

I’ll explain more on each of those categories, but first a warning. Not all ML/AI fits neatly into these three buckets. There are some nuances we’ll get into below.

Customer Behavior: Here, ML/AI improvements translate directly into increased sales. The impacts can sometimes be near-immediate. Consider ML/AI that is applied to product recommendations. The release of a better algorithm this morning can result in more sales this afternoon.

Similarly, ML/AI applications related to advertising can mean a company gets smarter about which ads it shows, which also can have an immediate effect. Feeder data, such as customer geolocation intelligence and identity resolution, often exhibits tight ML/AI-profit proximity characteristics. The last common example are pricing algorithms whether for direct customers or affiliates. The defining characteristic here is that better ML/AI can immediately spur revenues.

Better Product: This is ML/AI that makes products or tasks in that product noticeably easier. The prototypical example in the business-to-business space is any ML/AI application that falls under the general category of robotic process automation, where the improved ML/AI immediately increases user productivity. While user productivity is immediately improved, increased profit from this improved product must wait until customers subsequently buy more or resist switching to lower-priced competition, which can be weeks, months or years away.

Reputational Driver: The ML/AI here results in improvements that are not as obvious as better product improvements. A classic example is security software. If ML/AI improves threat detection, how is a customer to know? Most customers only notice product quality when there’s a breach, and even then, it’s still not possible to know if a competitor’s ML/AI would have done better. For this kind of ML/AI, higher profit is largely dependent on magazine reviews or academic studies that compare products and conclude your company’s ML/AI is better than everyone else’s. The bad news here is there’s a much longer delay between ML/AI investment and the return on that investment. The good news is that competitors trying to overtake your hard-won reputation are in the same boat. Regardless, the ROI for this category is often years but then tends to be a durable advantage. Acquisition targets with sterling ML/AI reputations may actually merit higher valuations.


The categorization is not always obvious, and there are times when a product applies ML/AI to more than one of the above categories: say, a retail product that uses ML/AI for customers behavior insights (customer insights) and more automation for back-office processing (better product). Or software-as-a-service products that bill based on usage. Here, a better product might result in immediate increased usage; if that usage is a factor in billing, revenue immediately goes up.

Recommended Steps

Note that ML/AI profit proximity is orthogonal to the ML/AI tech (natural language processing, image analysis, time series, etc.) involved. And that’s a good thing, because it means the following analysis can be done early on before expensive ML/AI diligence starts. I recommend taking these three steps:

Step 1: Draw up your own assessments of a target’s ML/AI profit proximity. Do this for each ML/AI application within the target.

Step 2: Ask the target for its own take on its ML/AI profit proximity. This step has two benefits: 1) It’s a gut check for the results you drew up in step 1, and 2) you’ll quickly expose a target who says they track their ML/AI’s ROI (most targets claim they do but don’t). This isn’t usually a deal breaker, but this step can often reveal enough detail to help bring a target’s ML/AI puffery back down to earth and allow for more fact-based conversations and valuations.

Step 3: Put the results of steps 1 and 2 into financial models. In particular, ensure that models correctly delay or accelerate profits based on relevant ML/AI profit proximity. While a fuller vetting of a target’s ML/AI competitive moat and ROI often requires a deeper tech dive, the above can quickly help model the timing of when various ML/AI investments will generate returns.


Original post:

Leave a Reply

Your email address will not be published. Required fields are marked *