Are all decisions programmable?
Decisions are at the heart of any company whether it is a public company, an SME, or a technology start-up. In this article, I present the main techniques that companies use to support or automate their decisions. To explore these techniques, I begin by asking where decisions are made. We will see that some are high stakes and therefore visible not only to the company but also to the market, customers, partners, and competitors. Others are so recurrent that they become almost invisible to the point that Bill Gates called them digital nervous systems.
Where are decisions made in companies?
In every business, there are three overlapping activities. Operations to run the business, projects to prepare for its future, and decisions to optimize both operations and projects. In other words, decisions are made at all levels, from the CEO who manages an entire complex conglomerate to the technician who operates a machine in one of the factories of the conglomerate or a local support agent who tries to resolve a contentious issue with a customer who is not happy with the product he has just purchased.
Modeling and automating a complete decision-making system would be too complex of a challenge — more complex than modeling and automating decisions made by one person. But the modeling and the total or partial automation of the decisions taken individually is possible as we will see later. To do this, I group decisions in companies into four categories: strategic decisions, tactical decisions, operational decisions, and technical decisions.
Strategic Decisions with Data and Real Options
By strategic decisions, I designate all the action plans or internal policies aimed at achieving global objectives. Strategic decisions are transformational in nature; examples include a merger or acquisition, a large capital raise, or an investment in an entirely new product, such as a vaccine in a pandemic crisis.
So far and in the foreseeable future, strategic decisions cannot be fully automated. Each situation is unique and requires creative thinking that is still beyond the capabilities of current AI. But this does not mean that tools are absent from strategic decisions.
In industries requiring large, high-risk investments such as energy, construction, and pharmaceuticals, companies use real options analysis. These are very sophisticated mathematical decision-making tools for risky projects as mentioned above. Originally, options are financial instruments that use historical data while considering uncertainty about the future. An option gives the company the choice to undertake certain large projects without committing all the necessary investment from the outset of the project. The strength of options is that the company can extend, delay, wait, or abandon the investment entirely based on data on changing economic, technological, or market conditions as the situation evolves. When the investment relates to a financial asset, we speak of financial options. When it comes to a real asset, we talk about real options.
Tactical Decisions with Data and Machine Learning
Tactical decisions are like strategic decisions but have narrower goals and less time span. Therefore, they do not rely on the long-term forecasting tools of the options briefly described above. On the other hand, they are big consumers of big data and machine learning.
For example, in the consumer goods, luxury, and beauty industries, data and predictive models are used to optimize advertising budgets across TV, OOH, radio, print, and digital. The same goes for forecasting the sales of a new product, by exploiting data on similar products, weather data, and even data on sporting and cultural events in the year to come. Big data and machine learning are also used for the computation of discount coupons to help customers save money, the brand to increase sales, and distributors to increase visits to their malls.
Thanks to big data and machine learning, tactical decisions such as those cited above become more measurable and improvable as companies execute them. We thus speak of learning companies in the sense that the data produced by the decisions is reinjected into the decision-making system to further improve future decisions.
Operational Decisions with Data and Business Rules
Operational decisions are the ones companies make in the thousands and sometimes millions in a single day. Financial services and insurance companies are typical examples of businesses where operational decisions are at the heart of the business. In each product they offer, there are cascading terms and conditions, legal constraints, eligibility criteria, risk levels, and other things to consider before deciding.
Operational decisions are normative insofar as they implement industry regulations, internal policies, or business strategies. Think of a branch manager at a bank deciding whether to lend to a borrower based on their repayment history, or an insurance agent calculating the premium a new policyholder should pay based on the brand of their car, its annual mileage and place and type of residence.
To automate operational decisions, companies in the US and China are increasingly turning to business rules in addition to data and predictive models. The grouping of data, predictive models, and business rules is often referred to by the acronym of decision intelligence by analogy with data intelligence.
Technical decisions with Data and Knowledge
Technical decisions are like operational decisions, but they are descriptive in the sense that they describe a procedure based on know-how often acquired through practice. Their modeling and automation were all the rage in the 80s and 90s under the acronym of expert or knowledge-based systems. As with operational decisions, automating technical decisions is easier and more efficient with rules accompanied by an engine that controls these rules.
Expert systems have automated many technical decisions where knowledge could be easily modeled, such as correlating alarms in a telecommunications network, configuring products, or diagnosing equipment failures. Nowadays, expert systems are embedded in the software that forms the backbone of large industrial, business, and financial systems. But they were unable to model and automate technical decisions where uncertainty was high, as is often the case when data is incomplete or even missing and expertise is still in its infancy. These limitations have given rise to several ways of improvement using the laws of probability but they are still in the state of academic research.
Micro-Decisions with Data and RPA
Although not part of the categorization of decisions, a fifth category of decisions deserves attention. I refer to it here as micro-decisions. They encompass all calculation and transformation automation operations, simple but numerous granular in the calculation of a price or an insurance deductible, for example. They are also found in the lower layers of data platforms for data ingestion, unification, and enrichment for data scientists and business analysts.
For their implementation, more and more companies use rules instead of scripts to implement their micro-decisions with the key advantage of easy maintenance during data changes or intermediate operations. Indeed, by their explicit side, it is easier to identify the rule(s) to modify than to browse and modify the code of a script.
And the humans in all this?
Interestingly, whatever the decision, whatever the level of sophistication of the tool and the level of automation it allows, at one or more moments, an expert, a designer, or a user must intervene. Either to define the problem, validate the decision, or correct it. A small error in a calculation or in a decision can led to big consequences in terms of brand image, losses, and even legal proceedings. Fortunately, modern decision management tools are all equipped with dashboards allowing the user to visualize and analyze the data on the one hand and to supervise the quality of the decisions taken on the other hand.
Going back to the title question, the answer is yes and no –It depends on which decisions. Companies use support tools for strategic and tactical decisions and automation tools for operational and technical decisions. Regardless of the category of decision support or automation tool, human intervention is always required to configure, monitor, refine, and justify the outcome of the tool. The day when all business decisions will be fully automated is not for tomorrow.
Conclusion
In this article, we have seen that unlike operations, not all decisions are programmable. This is the case of strategic decisions for which a complete and precise modeling would be a complex exercise without guaranteed success. But there are decision-making tools, coming from the world of finance, which make it possible to consider the uncertainty of the future for strategic projects. On the contrary, tactical, operational, and technical decisions as well as micro-decisions are programmable thanks to the combination of data, machine learning, and business rules.
Wrap-up
To find out more
The decision-making process, its modeling, and its automation has always been a vast field of research and application of academic research, particularly in operations research and artificial intelligence. Many books, journals, and articles have been devoted to decision-making in organizations. I nevertheless recommend the publications of Herbert A. Simon, Harry A. Klein and Daniel Kahneman. They cover the realm of the possible and the realm of the difficult or even the impossible when we think of automating decisions with current knowledge and technologies.
About the author and the article
Data consultant and interim executive for marketing, customer relationship management, and supply chain management in the retail, luxury, finance, and insurance sectors.
Although based on research, consulting, and implementation work, the views, thoughts, and opinions expressed in this article belong solely to the author, and not to his current or previous clients or employers.
An earlier version of this article appeared in English on VentureBeat.
