The chatter about Artificial Intelligence (AI) and its recent companion, machine learning (ML) is ever-present. In a wide range of economic activities, AI/ML methods are now standard practice. For example, if I search for soccer balls, soccer ball advertisements appear when I am reading the news online. For supply chain management (SCM) it is clear AI/ML has value, but the exact nature of its value and executive perceptions are murky. A recent presentation by Dr. Wright reviews executive perceptions of ML with regards to marketing. In that review, some of the critical barriers identified were (a) where to use AI/ML and (b) data issues or anomalies. Arkieva is certain that one of the best uses of AI is “data cleaning and profiling”. Based on its years of experience, getting the data right dramatically improves demand estimation outcomes. Pivoting from a proposal in the early 1980s by Dr. Gerry Hahn (“More Intelligent Statistical Software and Statistical Expert Systems: Future Directions”) Arkieva has just completed a major initiative in this area for use by its customers. Future blogs will explore this critical area.
The chatter about Artificial Intelligence (AI) and its recent companion, machine learning (ML) is ever-present.
In a wide range of economic activities, AI/ML methods are now standard practice. For example, to find a specific show or channel I talk to the remote. My 4 ½-year-old granddaughter talks to Alexa to get her favorite song to play. If I search for soccer balls, soccer ball advertisements appear when I am reading the news online. Although the overall value of AI and ML is clear, slowly an understanding of AI’s limits is emerging.
For supply chain management (SCM) it is clear ML has value, but exactly the nature of its value is murky since the various methods in ML toolkit have roots in other disciplines and almost all successful SCM analytic efforts must pull from various toolkits (analytics without borders – AWB). That said, since ML is the new analytics method with the largest chatter, it is critical to understand the executive expectations of ML. This blog will review an excellent presentation on this topic from the spring INFORMS conference and identify one of the best emerging areas for AI to improve SCM decision making is “data cleaning and profiling” to avoid data-driven disasters.
What are Executives Expecting from Machine Learning (ML) for Marketing Decisions
Each spring INFORMS runs its Business Analytics Conference. INFORMS is the premier professional analytics organization worldwide with roots in the 1950s and the early use of computers and analytic methods to improve supply chain decision making and performance. The 2020 event was virtual and one presentation that caught our attention was “Executive Perception of Machine Learning for Marketing Decisions.”
Overview of the presentation
Dr. Beverly Wright provides this overview: Industry executives hear about the buzz and promise of machine learning for solving tough marketing questions. We sought to gauge understanding, perceptions, and future direction of machine learning from a corporate perspective.” Based on the results of 20 in-depth interviews among senior business leaders, three main themes arose:
- Analytics leaders seem to believe that machine learning has had a positive impact on their company; however, quantifying the effects is difficult. Distrust and fear of job loss is a concern among some organizations.
- Barriers to implementing machine learning include people who do not understand the concept, analysts that are not trained, and data that is not ready for machine learning.
- Leaders appear generally optimistic about the future and growth of machine learning. While the exact nature is unclear, there is consensus among those surveyed that the role of the analyst will change.
Negative impacts and process change concerns include black box apprehension, fear of organizational disruption in who owns and benefits from machine learning, job loss, and hiring changes toward analysts who know machine learning. Positive machine learning attributes were increased confidence in decision making, mathematical rigor, demand for more data-driven decisions, and general excitement on revelation, discovery, and the future of innovation.
Barriers to machine learning implementation were identified as lack of understanding, poor data ecosystem, lack of talent, and knowledge of where or how to apply, general counter to gut lack of acceptance, regulatory constraints, and general beliefs of being too complicated.
Definition of machine learning with a data science orientation. The ability to obtain, process, and build models out of data for decision assistance.
Executives associate these terms with machine learning
- Identifying patterns
- Data, big data
- Using computer algorithms
- No human programming
- Example, you tell your child to clean up their room and let them figure out the rest as opposed to a step by step set of tasks: pick up the shirt, fold it, pick up dirty socks and put them in the laundry bask
- Teach machines to analyze and identify patterns without having to code specific rules
Primary Application to the Customer
- Propensity to buy
- Predicting the impact
- Fitting ads to an audience
- Lack of understanding
- Data not clean and available
- Lack of talent
- Uncertainty on where to apply
- Results counter to gut expectations
- Will not get used
- Unnecessarily complex
The work done and presented by Dr. Wright is very valuable to anyone who is working in this area today. It is critical to have a sense of what folks in charge are thinking and then work to control expectations Additionally, it identifies a critical gap that the technologists must answer – where can it be applied. In this presentation the focus was marketing, and another area where Arkieva is certain AI can help is automatically helping clean up data issues – called anomalies. Originally proposed in the early 1980s by Dr. Gerry Hahn in “More Intelligent Statistical Software and Statistical Expert Systems: Future Directions.” Arkieva has recently put in place AI modules that detect gaps and anomalies in your data, resolving issues without manual intervention and prescreening which forecasting methods are most appropriate to your data set and the best levels to forecast at.