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 call-centres to advanced robots that will one day set out to dominate the Earth. It sounds futuristic, but for experienced analytics practitioners in marketing—and direct marketing in particular—AI is not necessarily new.
AI represents another predictive modelling option and direct marketers have been using these 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 manual intervention to build and the output wasn’t in any presentable format.
Besides predictive modelling, AI has been used to develop highly sophisticated voice messaging systems, which is taking automation to a new level. These systems purport to achieve better customer service. While the success of that goal is debateable, from the organizational standpoint no one disputes the significant cost savings it offers over call-centres.
Early in 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. This change didn’t require advanced artificial intelligence; all it took was the birth of the PC and the internet to bring this increased level of automation into the office environment.
In the past five to seven years there has been a significant improvement in AI’s ability to produce accurate models particularly in the area of image and language recognition. The question now is if we were able to achieve a 20 percent lift in marketing response using traditional predictive models, does AI have the potential to improve that lift to 30 percent or more?
Software and other tools continue to automate the many manual-driven analytical tasks of the analytics practitioner, giving them the ability to process larger, more complex data files in less and less time. These advancements are helping more organizations see AI as a viable data science solution that can be deployed to help solve the business problem at hand.
With the evolution of artificial intelligence becoming more mainstream, one may ask what the impact will be in automation and in particular to the more knowledge-intensive tasks such as the discipline of data science. Increasing levels of automation that continue to replace labour may result in outsourcing becoming a moot point as the technology costs become cheaper than lower-cost labour from Third World countries.
But what skills will AI replace within the data scientist’s arsenal? In theory with AI, choosing the right mathematical algorithm becomes obsolete as the machine determines the right technique. The machine, through its artificial intelligence algorithms, outputs the solution which can be immediately applied to a given business problem.
Say a business needs to develop an upsell response model and has data from three different sources (customer file, purchase file and campaign/contact management file). Next, data from these three disparate sources are integrated into one analytical file. Like before, the data scientist would consider a variety of techniques to find the one that offers the best outcome for an upsell campaign, such as logistic regression, decision trees, etc. In this example, AI represents another technique. The exploration of multiple machine learning options, including AI, is typical of the open-minded approach adopted by most data scientists.
The important question marketers need to ask now, is whether these improved algorithms translate to improved performance in predicting consumer behaviour? In my experience, much of this behaviour is difficult to predict with a high level of accuracy due to the degree of random error or variation. But this does not mean that AI should always be discarded when looking at consumer behaviour.
In fact, if we are building marketing response models with millions of records, AI’s ability to leverage large volumes of data may offer increased model performance despite the fact that there tends to be more randomness in consumer behaviour.
Data scientists also need to consider the limitation of using an AI predictive model, specifically its interpretability. Amongst the more traditional predictive modelling techniques, the data scientist can easily identify each final model variable in terms of its impact on the target variable as well as its overall importance within the model. The so-called “black box” nature of AI and neural nets has created the need to conduct research in methodologies that will yield better model interpretability. The deployment of AI and neural models will always be minimized if the end user has no understanding of at least the key inputs that comprise the solution.
For now and for the foreseeable future, AI will not help us in the areas of identifying business problems, creating the analytical file, and implementation/measurement. It also won’t automate the data science practitioner.
What AI will do is shift the data science role within the organization. This shift will emphasize the ability to identify more business problems alongside the ability to create the right analytical environment that will address the given business problems. The data scientist will still need to have a deep understanding of the technology components, but will need to be more of a hybrid in being able develop more domain business knowledge. Having more of these hybrids is the key towards using AI more effectively thereby allowing companies to solve more business problems.
About the author: Richard Boire
Richard Boire's experience in data science dates back to 1983, when he received an MBA from Concordia University in Finance and Statistics complemented with career experience at leading-edge organizations such as Reader’s Digest and American Express.
Through his work at Boire Filler Group and most recently at Environics Analytics, Richard has become a recognized authority on predictive analytics in Canada with unparalled practical experience and expertise across virtually all business sectors. Richard published a book in2014 entitled “Data Mining for Managers: How to use data(big and small) to solve business problems” which was published by Palgrave McMillian of New York City.
Richard is currently senior vice president- customized analytics at Environics Analytics.