Decision Management
In an early ‘90s sketch by German comedy group Badesalz, a woman enters a music store to buy a drum computer as a present for her husband. The shop owner talks her through the details of an expensive device, demonstrating many of the pre-configured rhythms, which sound all the same. At one point, the woman inquires if the machine can also “do the Lambada”. The shop owner, after consulting the machine’s manual and going through many more of the identical rhythms, finally declares one of those to be Lambada. The women, overwhelmed by the advanced features of the device, decides to buy it. “But if it hadn’t had the Lambada,” she says when leaving, “I wouldn’t have bought it.”
When some vendors or organisations claim they use machine learning or artificial intelligence to enhance an operational process or even a business model, I sometimes feel reminded of this Lambada story. Today, it seems sufficient to do some AI voodoo to turn what used to be a payday lender into a fintech. And with some operational software systems, all other features seem to become irrelevant, if they come with some machine learning.
Machine learning has become accessible, and many libraries are available as open source. It is easy to attach some machine learning functions to an existing system or process. That alone does not deliver value.
Machine learning and data analytics are powerful methods, but typically the benefits do not come without effort, and careful considerations are required to make these tools efficient. When you purchase a tool or service, make sure it covers your core requirements before you start thinking about machine learning. If intelligence is claimed to be inside, make sure you understand how the respective technology helps to solve your business problem. If the mechanisms cannot be explained, or the benefits can’t be monitored and quantified, it is probably just another Lambada.
Getting the Value from AI and Machine Learning
If you really want to get the most out of AI and machine learning, think about your business problem first. And do your analogue homework, as machine learning is unlikely going to fix a broken process. Then think about where machine learning can help most: Where is the value in your business process, and which decision improved makes the biggest impact? Which knowledge and which predictions can help to make this decision better?
For example, if you have a business process that involves contacting customers either via phone or through digital channels, using AI to find the best time in the day to contact your customer digitally might not be the best place to start. First, the unit costs for calls will most likely be much more expensive than your digital contacts, and second, in an automated communication you could just ask your customers when and through which channel they would like to communicate, rather than using machine learning to predict.
In this example it is much more important to understand if and for which accounts a human contact provides an uplift, and what the additional costs are for a call over a digital contact. Machine learning can certainly help to find the right customers to call, but even a judgemental segmentation on whom to call might carry you further than using a complicated AI to find the best time to spin up your communication bot. So when using AI, make sure you use it on a high value problem.
Can You Fully Automate Model Development?
Among the strengths of machine learning is that the process of data preparation and model training can be automated. This is specifically useful in situations where models need to be re-trained on a frequent basis, for example because behaviour patterns change. However, the need for solid data preparation, model design and model development does not go away just because you use and automate machine learning. Skipping that effort might seem impressive, and is certainly easier, but is unlikely to deliver a great value.
If you really want to get the value from machine learning, you need to do it right, a point FICO’s chief analytics officer, Dr. Scott Zoldi, has made in blog posts about data scientist “cowboys”.
As with any other tool, machine learning and AI do not make all other problems go away. Take the time to structure your problem, and to understand which decisions in your process create the biggest value. Make sure to invest the time and effort to get machine learning right, and don’t rush for the fancy buzzwords otherwise most likely, you’ll get to your Lambada.
