Advanced Analytics versus Artificial Intelligence
For firms with Advanced Analytics or Artificial Intelligence (AI) in their future, Gartner has just published a useful report with three findings that clarify the difference between the two technologies and offers insight on employing them for augmenting versus automating supply chain decision making.
Advanced Analytics is broader than Artificial Intelligence
People often use these two terms interchangeably, but Gartner says that they are not synonymous. Their recent report entitled Augment and Automate Supply Chain Decision Making with Advanced Analytics and Artificial Intelligence (30 March 2018, Noha Tohamy) says that advanced analytics is the umbrella term for a variety of underlying technologies, whereas AI is a subset of advanced analytics.
As shown in the chart above, there are three types of advanced analytics technologies. The more basic, and most widely used, is predictive analytics which employs technologies such as statistical modeling and simulation. Gartner says that statistical modeling is more common because simulation requires more effort to develop and maintain models.
The next step up is prescriptive analytics which employs optimization, heuristics and rules-based “expert systems” with business rules defined by humans to solve a supply chain problem.
Artificial intelligence is the most leading-edge form of advanced analytics which includes machine learning, deep learning, natural language processing and “cognitive advisers” which are AI-based solutions that interact with business users through natural language. Gartner finds that deep learning is still only emerging due to its intensive data science requirements. Interestingly, Gartner’s poll of 260 users found that except for deep learning, all other categories of artificial intelligence are already now more widely used than heuristics and expert systems.
Advanced Analytics are more commonly used on the demand side
The Gartner report identifies a wide range of supply chain functions where advanced analytics are being employed, including supply side activities such as production scheduling and supplier management. But the top use areas are all focused on demand – demand forecasting, sensing and shaping. Demand forecasting was by far the most common use, with demand analytics slightly more popular than other well-known supply chain planning applications such as supply planning, production planning, factory scheduling, and sourcing/supplier management.
Gartner’s research matches what we’re seeing. Firms are looking for ways to employ machine learning to sense demand, asking “What data is out there? Weather. Social sensing (Facebook, Twitter). Consumer sentiment. Macroeconomics. Demographics. How can I take advantage of it? They are looking for cause and effect for short term forecast accuracy and longer term supply chain resiliency. A recent discussion with a retail industry supply chain analyst also aligned with this finding. He said that while the language around AI was ratcheting up in multiple ways, the area of most activity was promotional forecasting.
Augmenting and Automating are two different goals
Gartner makes a distinction between using technology to augment decision-making by generating insights and recommends actions, compared to automating decision-making to also execute decisions without human intervention. Which takes precedence, Gartner says, depends on the circumstances. “Areas like order fulfillment, production planning and demand forecasting are strong candidates for increased automation,” Tohamy says, “while collaborative processes like S&OP and risk management will continue to be better fits for decision-making augmentation.”
Again, Gartner’s research matches our own experiences here. More firms are asking why planners need to spend so much time nursing their planning system. Large enterprises ask, “What’s wrong with my process that I need armies of planners?” Growing mid-market growth companies ask, “Why do I need to keep adding so much overhead?” Worse yet, firms are asking if all this non-value added effort is preventing them from reaching higher levels of maturity. We see much less of this kind of discussion in S&OP and Integrated Business Planning (IBP).
Right now, advanced analytics is still used more to augment, not automate, process decision making. However, Gartner says that difference in use will narrow significantly in the next two years. This matches the message we have heard from other market analysts – automated decision-making is top of mind with every analyst we brief.
Tohamy thinks that the shift towards automation is attributable to “anticipated improvements in technology and data quality and increased organizational openness toward process automation.” She concludes that supply chain executives should “work with business leaders to understand and craft the vision for supply chain process automation and keep on top of advances in technologies to support the goal of accelerated AI-enabled process automation.”
Click below for a Supply Chain Brief on using machine learning to automate supply chain decision-making.
A Note on Gartner’s Survey Methodology
Gartner’s Advanced Analytics study surveyed organizations in five countries between July and September 2017. Two hundred sixty respondents participated, with about half coming from the United States. Other well-represented countries included Canada, Germany, Ireland and the U.K. Gartner says that country, industry, revenue and reporting team quotas were established. Qualifying organizations were from the retail, consumer products, chemical, industrial, high-tech and life science manufacturing industries with at least $500 million equivalent in total annual revenue for fiscal year 2017. Qualifying companies had to have already implemented advanced analytics capabilities for at least two of three categories (prescriptive analytics, predictive analytics and artificial intelligence).
Gartner’s Definition of Terms used in this Blog
- Advanced analytics is the analysis of all kinds of data using sophisticated quantitative methods to produce insights that traditional approaches to business intelligence — such as query and reporting — are unlikely to discover. Advanced analytics spans predictive analytics, prescriptive analytics and artificial intelligence.
- Artificial intelligence is a set of technologies that seeks to mimic human ability to understand data, find patterns, make predictions and find recommended actions without explicit human instructions. What distinguishes AI technology from traditional predictive and prescriptive analytics is (1) its ability to self-learn and (2) its ability to process natural language.
- Cognitive advisors are AI-based solutions that rely on domain-specific data and interact with the business user through natural language to support decision making.
- Decision-making augmentation is using technology that generates insights and recommends actions for business users but leaves it to business users to analyze those insights and approve and execute the recommended actions.
- Decision-making automation is using technology that generates insights and recommended decisions and executes those decisions without human intervention.
- Deep learning is a type of machine learning inspired by neural networks that is capable of uncovering hidden data layers to identify complex patterns. A common application of deep learning is in identifying facial images in large sets of pixel data.
- Machine learning is a type of technology that self-learns based on data and its own experience without being explicitly programmed by humans.
- Natural language processing is the ability of a computer program to understand human speech as it is written and spoken.
- Optimization: creating a mathematical model with a defined objective (like minimizing cost) and finding the best solution that meets that objective, taking into account costs and constraints.
- Predictive analytics is a set of technologies that predict future scenarios. Examples: demand forecasting and predictive maintenance.
- Prescriptive analytics is a set of technologies that recommends actions. Examples: calculating optimal safety stock and calculating optimal transportation route.
- Rule-based expert systems: business “recipes” or rules defined by humans to solve a supply chain problem.
- Simulation: building and observing a computer model that represents the supply chain, taking into account variability, to better understand ongoing performance.
- Statistical modeling is a mathematical approach to approximate reality and then use that approximation to make predictions. For example, statistical modeling can be used to generate an equation that describes actual demand levels and then use that equation to predict future demand.