By Richard Boire
The concept of artificial intelligence (AI) is the sort of thing that sets imaginations ablaze. To the general public, AI evokes images of everything from automated contact centres to advanced robots intent on global domination. It sounds futuristic, but to experienced analytics practitioners in marketing—and direct marketing in particular—AI is not necessarily new but simply another predictive modelling methodology.
AI drives productivity
Direct marketers have been using predictive models for almost 30 years. Back in the mid-1990s, for example, a major Canadian bank developed the modelling methodology and process to target customers who were most likely to purchase something more expensive than what they already have. That early work yielded net savings of $120,000 from a single campaign. The only drawback: it required considerable staffing resources and the output wasn’t in any sort of presentable format.
In the early days of my career, analytics work that once took a dedicated team a week to complete can now be executed by a single person in a couple of days. Some of these gains can be attributed to advancements in computing power. But over the past five to seven years we’ve also seen a significant improvement in AI’s ability to produce accurate models, particularly in the area of image and language recognition. If we were able to achieve a given lift in marketing response of 20% with traditional predictive models, AI can potentially improve that lift to 30% and more, and therein lies its appeal.
How analysts use AI
The increasing ability to quickly process very large volumes of data and the emergence of more advanced software and tools are helping to automate the some of the manual-driven analytical tasks of analytics practitioners. But it’s important to understand how AI fits in. In essence, AI adds another arrow to the analysts’ quivers; it doesn’t replace the need to have companies or analysts help prepare and interpret the data.
In customer analytics, analysts are still required to process the data before AI systems can work with it. Someone also has to decide if it makes sense to use AI as part of the solution, since it doesn’t work for every model.
Here is how it might apply to direct marketing today. Let’s say a bank wants to get a certain segment of its existing credit card customers to charge more to their cards. The first thing the organization needs to consider is what type of modelling approach to use to get the desired results.
Traditional modelling techniques such as regression and decision trees are still the go-to approach for most applications. But for more complex analyses they may not be sufficient to yield the desired results on their own.
Traditional methods are easier to work with when there is a clearer relationship between variables, but these approaches consider fewer than a dozen variables at a time. While that’s more than sufficient to create a reliable predictive model, some organizations may require a little more analysis.
For instance, if you have extremely large volumes of data and are considering dozens of variables then an analyst might turn to AI for assistance. Sticking with our credit card example, in the past the bank might have taken a sample of their customer list to build predictive models, but with the new processing power available today it’s becoming just as easy to run the model on the full customer list. This offers another advantage of AI since more accurate predictions require larger volumes of data.
While traditional models may look for specific statistical relationships with the data, such as linear or log-linear, AI isn’t bound by such constraints. By using AI or neural nets the computer can run a limitless number of variables to come up with the optimal model, or combination of models, to find the best prospects to contact. This allows organizations to significantly lower their costs without compromising their overall engagement.
Avoiding AI overreliance
But a word of caution. AI isn’t a magic bullet. One of the severe limitations of using an AI predictive model is its interpretability.
With traditional models, it’s easier to understand the relationships between variables and understand their importance in the model. AI, on the other hand, exists in a “black box” and it can require considerable research to fully interpret the findings.
While AI can provide data-driven solutions, but in our experience telling clients the answer based on AI isn’t enough. Our customers want to understand how certain variables are impacting that answer before they act on it. Relying too much on AI can cause problems over time these models no longer work, since there is no way to know which variable inputs are affecting the model.
For now and for the foreseeable future, the ability to figure out the right analytical technique can now be augmented through the use of AI. At the end of the day, AI is another tool in toolbox. Organizations just need to know when to use it.
Richard Boire is senior vice president of the Innovation Hub at Environics Analytics, where he focuses on transforming data into insights to drive more effective CRM results. He is a recognized authority on predictive analytics and data science, and is the author of Data Mining for Managers: How to Use Data (Big and Small) to Solve Business Problems.