It’s now a decade since Nassim Taleb formalised the powerful concept of Antifragility - the ability to gain from disorder. The author noted that while most things break when subjected to shocks (fragility), some thrive and grow when exposed to volatility, randomness, and disorder (antifragility).
In a world where disruption and randomness are increasing, should companies consider this notion explicitly in their value creation strategy? Which steps should they consider taking?
The Greeks encoded the concept of antifragility in their mythology. The bird Phoenix, whenever destroyed, is reborn from its own ashes back to where it was (robustness). But Greek mythology has a more powerful metaphor: the Hydra. The Hydra is a serpent-like creature with numerous heads. Each time one is cut off, two grow back. This effect results in a very odd creature that actively seeks out disorder - an obvious choice for a system benefiting from harm (antifragile).
We can recognize the same phenomena in our daily lives: our teacups don’t like turbulence, and (most) politicians won’t survive a minor scandal (fragile); but rockstars will see skyrocketing Spotify plays with every controversy.
As we turn the page on the pandemic, we see a similar effect: some companies had to close their doors; others were resilient to the changes; yet a third group benefited disproportionately from the disorder and increased their market share. It might be an interesting timely exercise to reflect on the conditions that allow some companies to become stronger when facing adversities.
This post discusses how AI is becoming an essential part of the puzzle and why you should consider antifragility as a guiding principle (beyond cost reduction and efficiency gains) that motivates and shapes how your AI journey will help you achieve your strategic goals.
Antifragile companies embrace volatility and errors. These stressors include disruptions and disorders such as natural disasters, financial crises, customer demand randomness, volatile customer tastes, and other unanticipated factors.
What’s the mechanism behind this process? Firstly we need to recognize that stressors carry information. The fact that your supplier is taking twice as long to deliver your order tells you something about the environment. Every plane crash makes the design of the next one more robust.
So we hit here the first characteristic of organisational antifragility. A company with static business processes and workflows won’t be able to learn with the information contained in the stressor. We need dynamic, adaptive capabilities to improve when exposed to faults.
From the above discussion, it is evident that agility - the ability of a business to mobilise and respond quickly - is a required but not sufficient property of antifragility. A truly antifragile company explicitly wants randomness and disorder (remember the Hydra). So what is the necessary ground for a company to be in a place where it can gain from disorder?
Here I’ll assume that antifragility manifests itself in a closed system, i.e. the antifragility of one thing comes at the cost of the ultimate fragility of something else (it’s an interesting question whether this is true in general).
In light of the above, we see that an antifragile company is one that can read and absorb the information contained in stressors more effectively than its competitors. Such a company will desire volatility in the environment - since it will benefit disproportionately from the disorder relative to its competitors.
While AI is not a silver bullet for every business problem, it certainly helps embrace volatility in operational conditions - by allowing the integration of disparate data sources into decision-making.
Amazon strikes us as a clear winner of the pandemic. Net sales rose by 40% in 2020 compared to the previous year. Profits have tripled. Part of the answer to the enormous benefits reaped by Amazon has to do with its unique relentless innovation and automation culture.
If we look more closely, though, we can understand the critical role of AI. The company has argued that "Machine learning (ML) and artificial intelligence (AI) technologies played a crucial role in the company’s ability to respond".
We all remember the empty shelves and shortages triggered by the first COVID-19 lockdowns. Consumer habits changed dramatically, with online shopping becoming a reliable alternative way to stock up. Consider the violence of the stressors created by the pandemic, with many variables changing at any given time.
As a company used to acknowledge demand volatility, it has been leveraging machine learning forecasts to plan comfortably ahead. These models accommodate all sorts of contextual information, allowing Amazon to keep an ear out and listen to the state of the world. Note that during COVID-19, different cities conducting different styles of lockdowns affected spending patterns and purchases in unique ways. Anticipating when customers were going to place orders based on their location, buying patterns, and even what those orders may be, permitted effective scenario planning.
The key observation is the following:
- Having in place all sorts of powerful models integrating rich-enough data sources to listen to the state of the world allowed Amazon to absorb and learn from information contained in the stressors created by the pandemic. Competitors lacking this ability could not exploit the information-loaded stressors, and the shockwaves rippling through their processes ultimately exposed their fragility.
Furthermore, the continuous investments in computer vision technology enabled Amazon to quickly launch an AI-based tracking system to enforce social distancing at its offices and warehouses to help reduce the risk of spreading the virus - ultimately reducing workforce shortages.
Conscious of its superior capabilities, one could argue that Amazon was waiting for the right levels of volatility and disorder affecting its competitors to consolidate its position. The pandemic heralded "a golden age" for Amazon.
From early in the global pandemic, Nike, the sportswear giant, responded quickly to a consumer shift to digital engagement and transformed its supply chain to serve consumers more directly. The volatility felt in supply chains all over resulted in increased lost sales due to product unavailability and increased costs in inventory inefficiencies. Nike is now using AI and machine learning technologies to predict and order the products that will be popular among consumers. This capacity enables fast, agile actions because the models anticipate demand changes rather than just respond to them. Novel algorithms, powered by the relevant contextual data, perform decision-making at a global level rather than at a local level, allowing potential disruptions to be addressed in real-time. Forecast changes in demand are automatically factored into processes and decisions along the chain, back to inventory, production planning, scheduling, and material procurement.
As reported in March this year, Nike’s sales have shown accelerating numbers as the company capitalises on the benefits of its digital businesses, effectively seizing the pandemic disruption to accelerate and transform its operations.
Many companies still rely on manual forecasting. And note that much forecasting is done - understanding demand is common to several internal functions like risk assessment, capital-expenditure planning, and workforce planning. Conventional approaches to demand forecasting depend on constant manual updating of input data and adjustments to forecast outputs. These interventions are typically time-consuming and do not allow for agile responses to immediate changes in demand patterns.
The two previous examples illustrate how AI can be used to absorb and learn from information-loaded stressors. To reinforce the connection between AI and Antifragility, we note the following:
- In general, AI systems benefit from exposure to novel data, which helps improve their performance.
- AI-driven processes can become capable of handling non-stationary contexts by using dynamic models that continuously integrate new stressors.
- Errors (or incorrect predictions) become a first-class object of analysis and integration (through feedback mechanisms) within the organisation. Overcompensation is encouraged by focusing resources on improving operational conditions where the process is more vulnerable.
- AI-driven approaches offer a natural context where a great variety of disparate data sources are integrated using powerful algorithms - enabling the prediction of events that were not possible to model before. Capturing the proper data context is essential in extracting information from stressors.
Designing and executing a strategy to build the right AI capabilities to support a business is no trivial task. In Mesh-AI, we help our clients deliver transformative outcomes. One of the ways we do this is through data democratisation and reimagining business process to become AI-driven.
It all starts with the identification of core and supporting business processes. These are broken down into constituent tasks. Within each task (e.g., connecting a solar farm to a national electricity transmission network takes 7 distinct steps), we identify the relevant decisions supporting it - allowing us to understand where predictions can be inserted. Most decisions are based on predictions and judgments informed by data. These could be anything from a university deciding to admit an application to a visual inspection of a product within a quality testing workflow.
It’s important not to lose sight of the economics – innovation can wither without a business case. So we estimate the business value of building a system that can automate each decision, prioritize each opportunity and create a delivery roadmap aligning the benefits of each item with the company’s strategic goals. This requires a continuous and interactive refinement process that is best handled as an agile backlog.
The number of opportunities within your organization is potentially huge. The Department for Business, Energy and Industrial Strategy (BEIS) in the UK recently estimated that 18% of existing UK occupations face a high probability of automation over the next 10 years. The economic implications for growth are even more staggering if we consider partial automation, i.e., certain activities within occupations and productivity boosts. AI is increasingly being used for problems that were not traditionally prediction problems - which calls for the ability to see a problem and reframe it as a prediction problem.
To characterize the potential role of a prediction system within an organization we produce an AI canvas for each one. Filling out a canvas introduces structure into the process. It forces us to be clear about all data sources and flows required. It also impels us to articulate precisely which predictions are required, the potential actions we can take based on the predictions, how we assess the relative value of different actions, and the desired outcomes.
Getting started early not only helps produce results quickly but also helps speed up an organization’s journey toward embracing the full potential of AI.
Many companies delay serious investment in AI capabilities, endlessly waiting for better-quality data. We should reverse the equation and use AI to drive demand for data quality improvements in targeted segments of the data with significant pockets of value.
The AI transformation doesn’t require a large up-front investment, but an agile methodology is an enabler. We advise small and quick iterations through incubators and lighthouse projects (see Mesh-AI Orbital Framework).
Interested in seeing our latest blogs as soon as they get released? Sign up for our newsletter using the form below, and also follow us on LinkedIn.