Business Analytics and AI
And still the seminars and articles extolling the promise of AI continue to tempt the unwary. It appears that all problems known to mankind can be solved with AI – or is it the ‘snake oil’ salesman at work to enhance the ‘fear of missing out’ (FOMO)?
An example is surveys that indicate a high percentage of participants that already use AI in their business. But what is included under the banner of AI? Even though the term Artificial Intelligence was first identified as an academic discipline in 1955, quantitative techniques within business continued to be called business analytics. All have since moved, and now reside under the catch-all term of AI. Business Analytics contains three types:
Descriptive Analytics, which includes Diagnostic Analytics: Used when managers consider past activities and look, through analysis of the historical data, for the reasons behind success or failure. Management reports utilise this form of analytics. Descriptive Analytics commenced in the 1950s to answer the question “what happened”. It incorporated initial developments of linear programming and computer based modelling and then optimisation in the 1970s.
Predictive Analytics: Data is analysed to provide information about what may happen in an organisation, with probabilities. Predictive Analytics commenced in the 1980s. It uses historical data combined with rules, algorithms and sometimes external data, to detect patterns (called pattern recognition). The computer learns the patterns experienced in the data (called machine learning) and determines the likelihood of a similar event occurring and the most likely outcome of an event, should it occur.
Prescriptive Analytics: Is a term introduced in 2010 by IBM when ‘digital transformation’ raised its head. Using Prescriptive Analytics, data is analysed to provide information, not only about what may happen, but how the objective can be improved to optimize a process or service. It uses scenarios to identify the likely outcomes of different decisions and the inter-related effects of each action. Prescriptive Analytics incorporates structured (numbers, categories) and unstructured data (text, sounds, images, videos) and can continually absorb new data to improve prediction accuracy and decision options.
So analytics is not a new concept. What is new are developments in computing processing speeds, data storage, cloud computing, and programming, together with machine learning models. These have enabled information technology to increase its capability for AI.
Simplify before AI
Prior to writing a CAPEX proposal to implement and use AI in the hope of solving complex problems, think about the problem. If it is demand unpredictability and supply shortages, the solution is not AI. If the users cannot analyse the solution to a problem based on supply chain principles, then AI will not be able to work through the problem!
Instead, put the effort into simplifying the planning of your organisation’s supply chains. Focus on designing the organisation’s supply chain model, utilising segmentation, constraints and capabilities; the Supply Chains Map to understand constraints; Supply Markets Intelligence to understand supply and Supply and Operations Planning (S&OP) to balance demand and supply.
But, if a case for AI is considered as possible, then AI of itself has limited value. The approach is that while AI is technology-driven, for long-term success the implementation project must be people centred and collaborative.
Use cases of AI in Supply Chains
While there may be use cases written about implementing AI within a Supply Chains group, they are impossible to find. In part, this can be because manufacturers and distributors (shippers) are not early adopters of ‘new’ technologies. There are too many stories of businesses unable to operate due to a ERP system malfunction or automation failure.
As an example of the assistance provided to a business using Predictive Analytics, the supply chain risk advisory firm Resilinc has written about the techniques used within its business:
- Predict Purchase Order Delays: Regression models – to estimate the expected delay in supply that will be caused by an event. Identify how a supplier will perform by analysing past events and on-time delivery data.
- Commodity Tracking and Predictions: Time Series Forecasting – predict commodity price fluctuations and supply constraints more than three months out.
- Supply Chain Disruption Alerts: Classification Models – determine whether or not a news item can be considered a supply chain disruption from about 8,000,000 rows of data received daily.
- Supplier Network Predictions: Recommendation Models – this AI is tailored for data structures as graphs, such as knowledge graphs, to help companies gain visibility concerning where suppliers’ factories are located.
- Natural Disaster Supply Chain Disruptions: Simulation Models – for simulating the effects on supply chains from real-world scenarios such as extreme weather and other disruptive events.
Note that the business does not use Prescriptive Analytics. They may be waiting for the hype surrounding AI to subside or have decided that the techniques are not required. In the company’s favour for using the techniques are that the data used is either supplied by clients or is otherwise controlled and therefore has a high confidence level for accuracy. Also, as a data intensive business, the company employs data specialists, so management of the process is a key outcome to provide assurance to their clients.
It is possible that in a Supply Chains group (Procurement, Operations Planning and Logistics), there is potential for AI within: demand forecasting, production scheduling, inventory management, predictive maintenance and warehouse management. In transport it is freight management, fleet management and route optimisation. AI can also help to improve the administration of purchase orders and invoices, and the increasing requirements of compliance and associated records.
In supply chains operations it is better that AI is not used to develop the perfect operations schedule or transport route within what is a ‘complex, adaptable system’. Instead, AI should ask the knowledgeable planner or scheduler to consider options B or C or D and provide the evidence. This enables the specialist to make an improved decision, while considering the many options and constraints that may be occurring.
As always, the main challenge of selecting and implementing the software technologies is understanding your organisation’s business challenges and capabilities and how the application will address them.