Recent advances in data-driven technologies have unlocked the potential of prediction through artificial intelligence (AI). However, forecasting in uncharted territory remains a challenge, where historical data may not be sufficient, as seen with unpredictable events such as pandemics and new technological disruptions. In response, hypothesis-oriented simulation can be a valuable tool that allows decision makers to explore different scenarios and make informed decisions. The key to achieving the desired future in an era of uncertainty lies in using hypothesis-oriented simulation, along with data-driven AI to augment human decision-making.
Can data-driven analytics predict the future?
In recent years, AI has undergone a transformative journey, fueled by remarkable, data-driven advances. At the heart of AI’s evolution lies the astonishing ability to extract profound insights from massive datasets. The rise of deep learning models and large language models (LLMs) have pushed the field into uncharted territory. The power to leverage data to make informed decisions has become accessible to organizations of all sizes and across all industries.
Take the pharmaceutical industry as an example. At Astellas, we use data and analytics to help inform which business portfolios to invest in and when. If you are developing a business model focused on a common and well-understood disease area, the power of data-driven analytics enables you to derive insights into everything from drug discovery to marketing, which can ultimately lead to more informed business decisions.
However, while data-driven analytics excels in established domains with ample historical data, predicting the future in uncharted territories remains a formidable challenge. It is difficult to make data-driven predictions in areas where sufficient data is not yet available, such as areas where extraordinary change or technological innovation has occurred (it would be very difficult to predict the impact of a sudden pandemic of an infectious virus or the rise of generative AI on a particular business in its early stages). These scenarios underscore the limitations of relying solely on historical data to chart a course forward.
A typical example in the pharmaceutical industry, and one that Astellas regularly confronts, is the valuation of disruptive innovations like gene and cell therapies. With so little data available, trying to predict the exact value of these innovations and their far-reaching impact on the portfolio based solely on historical data is like navigating through dense fog without a compass.
Peering into the Future: Hypothesis-Oriented Simulation
One promising approach to navigate the waters of uncertainty is hypothesis-oriented simulation, which mimics real world processes. If you are a business that’s venturing into unknown areas, you need to adopt a hypothesis-oriented approach when historical data is not available. The model represents how key factors in the processes affect outcomes while the simulation represents how the model evolves over time under different conditions. It enables decision-makers to test different scenarios in the virtual “parallel worlds”.
In practice, this means laying out a smorgasbord of key scenarios on the decision table, each with its own probability and impact assessment. Decision makers can then evaluate critical scenarios and formulate strategies for the future based on these simulations. In the pharmaceutical industry, this requires making assumptions about a range of factors such as clinical trial success rates, market adaptability, and patient populations. Tens of thousands of simulations are then run to illuminate the murky path ahead and provide invaluable insights to steer the course.
At Astellas, we have developed a hypothesis-oriented simulation, which creates scenarios and makes a deductive guess, to help inform strategic decision making. We are able to do this by updating the simulation hypothesis in real-time (at the decision-making table), which helps improve the quality of strategic decisions. Project valuation is one topic where the simulation method comes in. First, we build possible hypotheses on various factors including, but not limited to market needs and success probability of clinical trials. Then, based on those hypotheses, we simulate events that occur during the clinical trials or after product launch to generate the project’s possible outcomes and anticipated value. The calculated value is used to determine which options we should take, including resource allocation and project planning.
To dig deeper, let’s look at a use case where the method is applied to early-stage project valuation. Given the inherently high level of uncertainty that comes with earlier-stage projects, there are an abundance of opportunities to mitigate the risks of failure to maximize the rewards of success. Put simply, the earlier a project is in its lifecycle, the greater the potential for flexible decision-making (e.g., strategic adjustments, market expansions, evaluating the possibility of abandonment, etc.). Evaluating the value of flexibility is, therefore, paramount to capture all the values of the early-stage projects. That can be done by combining real options theory and the simulation model.
Measuring the impact of hypothesis-oriented simulation requires an evaluation from both the process and the results perspectives. Typical indicators such as cost reduction, time efficiency, and revenue growth can be used to measure ROI. However, they may not capture the entirety of decision making, especially when some decisions involve inaction. Furthermore, it’s important to recognize that the results of business decisions may not be immediately apparent. In the pharmaceutical business, for example, the average time from clinical trials to market launch is over 10 years.
That is, the value of the hypothesis-driven simulation can be measured by seeing how it is integrated into decision-making process. The more the simulation results have impact on decision-making, the higher its value is.
The Future of Data Analytics
Data analytics is expected to diverge into three major trends: (1) An inductive approach that seeks to identify patterns in large data, which works under the assumption that the patterns found in the data can be applied to the future we want to predict (e.g. generative AI); (2) An analytical approach, which focuses on interpretation and understanding of phenomena where sufficient data cannot be utilized (e.g. causal inference); and (3) A deductive approach, which relies on business rules, principles, or knowledge to see future outcomes. It works even when there is less data available (e.g., a hypothesis-oriented simulation).
LLMs and other data-driven analytics are poised to significantly expand their practical applications. They have the potential to revolutionize work by speeding up, improving the quality of, and in some cases even undertaking human work. This transformative shift will allow individuals to focus their efforts on more important aspects of their work, such as critical thinking and decision making, rather than more time-consuming activities, such as data collection/arrangements/analysis/visualization, in the case of data analysts. When this happens, the importance of which direction to move in will increase, and the focus will be on augmenting human decision making. In particular, the trend will be to use data analytics and simulation for strategic decision-making while managing future uncertainties from a medium- to long-term perspective.
In summary, achieving a harmonious balance between the three approaches above will maximize the true potential of data analytics and enable organizations to thrive in a rapidly evolving landscape. While historical data is a tremendous asset, it’s important to recognize the limitations. To overcome this limitation, embracing hypothesis-oriented simulation alongside a data-driven approach enables organizations to prepare for an unpredictable future and ensure that their decisions are informed by foresight and prudence.
Credit: Source link
