Three things to know about predictive AI

Some problems are easier to solve with predictive AI than with generative AI. Many business operations require predictions rather than the generation of new content to run more efficiently. That's why predictive AI is the type of AI that companies rely on to improve the effectiveness of large-scale processes. Predictive models decide who to contact, approve, test, alert, investigate, detain, set up dates, or prescribe medication. They target operational decisions: market to people who are likely to buy, approve loans to people who are likely to pay on time, and screen people who are likely to get sick.

But the world isn't making it easy for predictive AI to succeed. First, genAI's current popularity has made predictive AI a relatively unsung hero. Second, predictions are, after all, probabilities, and probabilities aren't sexy. Cultural aversions keep businesses from embracing probability. To many business professionals, the topic of probability can seem boring at best, and arcane and complicated at worst.

But there's no way around it. Embracing predictive AI means becoming a business that operates probabilistically. Generally, there are no absolute, ultra-confident predictions. There is no magic crystal ball. But there is the next best thing: a number between 0 and 100 that represents the expected likelihood of a particular outcome or action. That's a probability. And it is exactly this probability that you get from a predictive model generated by machine learning. You could even call this model a “probability calculator.”

Culture strongly resists probability. The Empire Strikes Backour trusty android C-3PO nervously cries, “Your chances of successfully navigating the asteroid field are roughly 1 in 3,720,” but our beloved hero, Han Solo, risks his ship in danger as he retorts, “Don’t tell me the odds!” Perhaps the screenwriters were echoing a common sentiment: It would feel uncool to admit that if I were the captain, I’d want to know what our chances were.

In contrast, beloved films MoneyballThe book is a compelling example of successful quant-stakeholder collaboration with a true success story: Working closely with data scientists, the general manager of the Oakland Athletics baseball team unexpectedly outperformed and won.

The problem is, despite being a crowd-pleasing drama, Moneyball This is a classic example of underestimating math, and it does little to teach leaders how to work with quants.

But if you take a deep breath and look closely, you'll see that predictive AI isn't difficult to understand. The upskilling required is accessible, not esoteric. Business professionals working with predictive AI need to develop a semi-technical understanding that boils down to three things: 1) what to predict, 2) how accurate the prediction is, and 3) what to do about it.

For the first three, you work with data experts to determine what outcomes or actions you want to assign probabilities to, such as whether a customer will click, buy, lie, die, cancel a subscription, commit fraud, etc.

For the second, you need to decide what metrics to report to determine if your ML model is production-ready. This could include simple business metrics like the improved profits or savings you expect to bring with your deployment. Spoiler alert: accuracy is usually a rude and misleading metric.

For the third, you need to establish how to act on the prediction: for example, if the model predicts that a customer will purchase if contacted, include them in a marketing campaign; if the transaction is predicted to be fraudulent, block or audit it.

Most predictive AI projects never get adopted. I believe this is primarily due to a lack of close collaboration with business stakeholders. To improve this dismal track record, business professionals need to be involved in the details of these three aspects of predictive AI projects, provide an informed perspective, and help keep the project on track not just technically but practically as well. After all, if you don't get your hands dirty, you may find yourself with cold feet when it comes time to approve the adoption of predictive AI.