Predictive AI Too Hard To Use? GenAI Makes It Easy

Problem: While anyone can use genAI, predictive AI is generally out of reach. Solution: Use genAI to ..

. More steer your predictive AI project. So much potential.



Predictive AI often delivers more value than its heavily publicized sister, generative AI. It’s the technology you turn to for improving your largest-scale operations. Its value proposition is straightforward and it has accumulated a decades-long track record of proven deployments.

But to date, predictive AI has achieved only a fraction of its great potential because it is hard to use. Anyone can use genAI, since it’s trained to respond to human-language prompts, but predictive AI isn't readily accessible to business users in general. To use it, a business professional needs the assistance of data scientists as well as a semi-technical understanding of how ML models improve operations .

Since this understanding is generally lacking, most predictive AI projects fail to deploy – even when there are data scientists on hand. GenAI Chatbots Make Predictive AI Accessible An AI chatbot does the trick. With the right configuration, it puts into the hands of the business user a virtual, plain-spoken data scientist that helps guide the project and answers any question about predictive AI in general.

It serves as an assistant and thought partner that elucidates, clarifies and suggests, answering endless questions (without the user ever fearing they’re pestering, overtaxing or asking “stupid questions”). Here's an example that demonstrates how well this approach works. Consider a project to target a marketing campaign that shows the following profit curve .

A profit curve estimating the profit of a marketing campaign targeted with a machine learning model. ..

. More The horizontal axis represents the number of customers contacted – ordered from most likely to buy if contacted down to least likely – and the vertical axis represents the marketing campaign's profit. The gray vertical line represents the decision boundary.

If set to 25% (as shown), the campaign's profit is estimated to be almost $1 million, depending on the settings shown at the bottom: an average $20 profit for each positive customer response and a $10 cost per contact. This example is based on the data set used for The Second International Knowledge Discovery and Data Mining Tools Competition. Thanks to Dean Abbott for the predictive models used for this example.

When given the image above – and prompted “under the hood” with several pages of plainspoken, explanatory background information – we get the following (this example uses Anthropic’s Claude Sonnet 3 large language model). CHATBOT: The chart displays three profit curves representing the forecasted profits for three different predictive models across varying percentages of customers contacted. With the given parameter settings, Model 1 achieves the highest peak profit around 25% of customers contacted, while Model 3 has the lowest profit potential.

The curves provide visibility into the tradeoffs between contacting more customers and diminishing returns. ME: What's the purpose? CHATBOT: The purpose of this profit curve chart is to help determine the optimal percentage of customers to contact for a targeted marketing campaign by visualizing the trade-off between the number of customers reached and the expected profit for each predictive model. Continuing this dialogue, I asked a couple questions about why the curve goes up then down, and then asked: ME: Explain that for a 10-year-old using a story.

CHATBOT: Let’s imagine you have a lemonade stand and you want to make as much money as possible. You have a special machine that can predict which kids walking by are most likely to buy your lemonade. At first, you use the machine to call out to the kids it says are really likely to want lemonade.

Since these kids really want lemonade, most of them buy from you, and you make good money. But as you keep calling out to more and more kids, eventually you start calling kids who don’t want Lemonade that much. Some of them buy, but many don't.

And for every kid you call out to, you have to use your allowance money to make the lemonade. So while you make some money from the kids who do buy, you’re also spending money on lemonade for kids who don't buy. At some point, you’ve called to so many kids that you’re spending more of your allowance making lemonade than you're making from the sales.

That's when the profit curve starts going down - you're losing money by calling to too many kids. This elucidation is stellar. Most business professionals have yet to realize that the core value proposition of predictive AI is well within their grasp of understanding – it isn’t “rocket science” and comprehension hinges only on arithmetic.

A genAI chatbot can ramp them up – and then remain available moving forward as they revisit the concepts to reinforce their understanding and become comfortable with them. Beyond the example above, LLMs also handle many other kinds of user questions well, such as: Why does the curve start steep? How is the displayed profit calculated ? Where should I move the decision boundary cutoff? What parameters should I consider changing? How do I navigate tradeoffs between competing KPIs and how do these charts help me do so? How do I decide when the project is ready for deployment? Well-defined, finite-scoped tasks like this are exactly the kind of thing that LLMs can handle well. For this use of a chatbot, it only needs as background information the equivalent of a few articles’ worth of written material (seeded to the LLM as a prompt “under the hood”).

This stands in contrast to more ambitious use cases for generative AI, which are plagued by reliability challenges. Often hailed as a panacea that can – or will soon be able to – solve most any business problem, genAI is meant to handle projects that assimilate much larger knowledge bases, work with and selectively divulge sensitive data, and automatically enact consequential transactions, such as purchases, medical prescriptions or flight changes. While it’s an open question whether and when genAI will advance far enough that its most ambitious incarnations can be released to interact with consumers at large, predictive AI may be able to realize a great deal of genAI's promise of autonomy .

But the article that you’re currently reading focuses on the reverse: genAI helping predictive AI. GenAI can handle the scope of this task indeed, thereby meeting a critical, outstanding need: bridging the costly tech/biz divide that hampers predictive AI projects at large. Beyond genAI assisting predictive AI – and predictive AI addressing genAI’s reliability problem – the two fields are also destined to blend in other ways.

GenAI can empower predictive AI projects by itself acting as a predictive model , and large database models complement large language models, tapping a company's tabular data and serving for predictive AI projects. This June, I will present a keynote address on this topic, “ Five Ways to Hybridize Predictive and Generative AI ” (also at this online event on April 8, 2025)..