The essence of strategy is making choices. Choices about scope, products, organisation or about the direction of the company for the future. In the age of Datanomics, more and more choices are automated. Is it the end of strategy?
The three layers of machine learning in the strategy process
Data generate value when they are used to make a decision or perform an action. This is where machine learning plays a big role. Machine learning can be defined as the ability of statistical models to develop capabilities and improve their performance overtime without the need to follow explicitly programmed instructions. Computer vision, speech recognition, natural language processing or robot control are typical applications of machine learning where a system is trained by showing it examples of desired input-output behavior.
What does this have to do with strategy? Well, machine learning is present at three layers in the strategy process.
Layer 1: Implementing strategy with data and AI
Big data and machine learning have been massively used to support strategy implementation. Whether on revenue increase or on cost reduction, their impact on value creation is unquestioned. In particular, automated and customized content have a positive impact both on revenue increase and cost reduction. Though not perfect, Toutiao and GPT-3 recently demonstrated that the creation of original content can be automated. So, the first layer is when machine learning is used to implement a strategy.
Layer 2: using data and analytics to define strategy
The research report on the use of big data and analytics to define a strategy or at least contribute to that definition. Two recent pieces of research reveal such practices. Pröllochs and Feuerriegel (2020) analysed 250 000 8-K fillings documents from companies in the energy sector and they used topic modelling to identify the strategies used by the different companies over time. One could easily imagine the benefit for a company in the same industry to tap in that research to define their own and distinctive strategy. On a similar note, Tidhar and Eisenhardt (2020) used machine learning techniques to analyse the revenue models of 66 652 applications and answer a typical strategy question for a new application: what is the optimal choice of revenue model? As more and more data is available, the second layer is when machine learning is used to contribute to defining the strategy.
Layer 3: communicate a strategy that fits with the categories of the machine learning algorithms reading it
As a complement to the previously mentioned research, Cao & al. (2020) investigated how companies adjust their language and reporting in order to achieve maximum impact with algorithms that are processing corporate disclosures. According to that study, companies go beyond machine readability and manage the sentiment and tone of their disclosures to induce algorithmic readers to draw favourable conclusions about the content. For example, companies avoid words that are listed as negative in the directions given to algorithms. So the third layer is when decisions or communications are partly influenced by how the machine learning algorithm will analyse them.
Companies also appear to adjust their use of words associated with potential stock market reactions, such as those that the alternative dictionary labels as litigation-related, uncertain, or demonstrating too little or too much confidence. (Cao & al, 2020)
What does it change?
So, if business and strategic decisions can be automated and if they are designed to please the algorithms analysing them, where is the choice, where is the strategy? Does it signal the end of strategy?
Let first admit this situation is not totally new or original. In fact, the three layers we just described can be observed in other functions than strategy. Marketing and communication have been through that path already. Generative design, search engine optimisation, content mills and click farms illustrate the three layers we just described. Of course, marketing and communication changed but they didn’t vanish, quite the contrary. It is also clear that one function of strategy, as a discourse, is to communicate to stakeholders to get their support. The third layer of machine learning in the strategy process just means that a new stakeholder needs to be taken care of: machine learning algorithms that read corporate speeches.
Even if it’s not totally new or original, the use of machine learning in strategy is changing how strategy is designed, implemented or communicated:
- as corporate speeches are more and more read by machines (nearly 80% of the readers of the 8K fillings are machines according to the study I mentioned earlier), companies are starting to consider that stakeholder as seriously as market analysts and industry experts
- as data is widely available and machine learning techniques better mastered, the gap between those who use them intensively and those who do not is widening. Data and analytics capabilities are critical
- like for marketing, the race between the content producer and the algorithm designer has started. The more a company knows about how algorithms work, the more chances it has to strike a favourable balance, using the terms valued by the algorithm
- with automation and retro engineering, the probability of manipulations and misconduct increases. Companies that act in a transparent way are likely to see their brand image reinforced, contrary to the ones whose bad behaviour is exposed
But one thing doesn’t change: part of the success of a strategy lies in its difference and unpredictability. However, machine learning models are based on analysing past events, understanding past correlations between variables. Though important and useful, reproducing the past limits somehow the ability to build a different future or to adapt to a radically different environment. For that reason, companies that design and communicate original and unexpected strategies, even less numerous than the ones adjusting to anticipated expectations, are likely to benefit from an advantage.
In fact, machine learning doesn’t signal the end of strategy, it signals an evolution of how it’s performed and on the key capabilities to master. It also reinforces the necessity to balance two opposite demands in strategy: adapting to the environment and building a unique position.
Questions you should ask yourself to leverage data and analytics in your strategy:
- layer 1: what is the role of machine learning tools and techniques to implement your strategy?
- layer 2: what large datasets do you analyse to contribute to your strategy definition?
- layer 3: what topic modelling or natural language processing tools have you used to test how your strategy is interpreted by a machine?
Questions you should ask yourself to build a unique and valuable position:
- what is the core challenge of your company?
- what do you want to change in the market and environment?
- what do you do differently than our competitors?