Simple rules for complex times
Making big decisions is a daily task for business executives. Dan Goldstein asks if you have given much thought to how those decisions are made. It may be that the way you or your company reach decisions is both time-consuming and wide of the mark when it should be fast and frugal
People make decisions on a regular basis, but people in certain fields - such as business, medicine and government - have to make decisions that impact others, often in major ways. Yet the art and science of decision making is just beginning to receive, properly so, new and greater scrutiny. For example, in a recent article in Medical Education, some elemental questions were posed that illuminate the reason for such scrutiny:
Imagine the following situation: a man is rushed to hospital with serious chest pain. The doctor suspects acute ischaemic heart disease and needs to make a quick decision: should the patient be assigned to the coronary care unit or to a regular nursing bed for monitoring? How do doctors make such decisions? And how should they?
Those without years of formal medical training (and I suspect even many with such education) would probably address the last question with some variation of this answer: "In such a situation, a good doctor would collect as much information as possible about the patient, weigh it against the backdrop of the current emergency, consider all possible options, and then make the best decision." Such advice is commonly found in decision making textbooks. What the authors of the article concluded - and this squares with my own research and thinking - is that such an answer may, in fact, be wrong. Collecting as much information as possible could be time-consuming when every second counts, and considering all possible options could burden the doctor with choices that are far-fetched, unrealistic and, quite possibly, unhelpful to the health of the patient. Think this doesn't apply to you in the business world? Think again. "Heuristics" may not be a word you use every day, but they enter into most decisions you make.
Undue diligence
Heuristics is a term derived irregularly from the Greek word heuriskein, which ties to our word for find. Heuristics are rules of thumb. They help us find solutions to problems. When applied to decision making, they therefore help someone make a decision under limited time and knowledge. Executives often insist that their management teams generate sophisticated forecasts of what the future will bring. But are "sophisticated" forecasts really any better?
I and a number of others have explored the question of how simple heuristics compare to optimizing methods in the science of forecasting, and results are startling. In reviewing the studies of forecasting techniques tied to predicting the daily temperatures for next year based on this year's, tracking where criminals might strike next, projecting the outcome of Wimbledon or World Cup tournaments, or (perhaps closer to your interests) if current customers will purchase from a business again in a given time frame, it became clear that a more complex forecasting strategies had no clear advantage for many everyday problems. Phrased differently, we found that collecting and analysing all available data may turn out to be undue diligence.
Prediction competitions
Consider the important task of forecasting future best customers. In 2008, two researchers, Markus Wübben and Florian Wangenheim, tested two dominant ways to predict (from historical data) 1) which customers are inactive, that is, not going to repurchase, and 2) which customers will emerge as the "best customers" in the future, that is, purchase the most. Customer data from the airline, apparel and online music industries were used to test the models. The apparel data, for example, contained initial and repeat purchase information for 2,330 customers of a retailer over a period of 80 weeks.
The authors staged a prediction competition. First, they considered a relatively simple "hiatus heuristic" that they knew managers actually used: If a customer had not purchased within a certain number of months in the past (the "hiatus"), the customer was classified as inactive. Others were considered active and, therefore, probable repeat customers. For the question of predicting the best customers, they used a simple persistence heuristic: They simply predicted that the top X per cent of best customers in the past would be the top X per cent of best customers in the future.
Pitted against the heuristics were two complex stochastic models, which cannot operate without estimating a handful of parameters from training data. While such models can perform admirably when conditions are right and data is abundant, they provide no particular advantage under the all-too-common conditions of having limited knowledge.
How did the competition turn out? The surprising result of the competition was that the sophisticated forecasting techniques fared no better than the extremely simple rules of thumb articulated by the managers. How could this happen? Without needing to estimate a host of parameters, the heuristics avoided wildly inaccurate results that may arise from over-generalising from relatively small samples. The heuristics, on the other hand, have been around for a long time. They hold within them the wisdom of decades of data, and they no doubt evolved and improved as they were passed from manager to manager. In these analyses, the intuitions of actual managers were surprisingly close to optimal. For example, for the apparel industries, the managers' intuition for the length of the hiatus was 40 weeks. The optimal figure turned out to be 39 weeks. For this industry, choosing the optimal hiatus over that from the managers' intuitions would yield an improvement of less than one per cent in terms of correctly classified customers.
For the question of finding the "best customers", customers in the same airline, apparel, and online music industry datasets, were categorised as either in or out of the top 10 per cent or 20 per cent by number of transactions. The key statistic of managerial interest in this domain is the percentage of predicted best customers who actually turned out to be best customers. On this criterion, out of 12 tests (two competitor models, three industries and two levels of best customer being top 10 per cent or 20 per cent), the heuristic beat the stochastic models all but three times. Further analyses showed that varying the length of the time period studied had little impact on the results.
Fast and frugal
The phrase now being used to identify heuristics that are quick to employ because they do not require vast amounts of data collection and analysis - and are yet predictably accurate - is "fast and frugal". But it would be an enormous mistake to believe that the simplest decision making is, always or even as a rule, the best decision making. In 1934 Albert Einstein noted that "the supreme goal of all theory is to make the irreducible basic elements as simple and as few as possible without having to surrender the adequate representation of a single datum of experience." This has often been paraphrased "Make things as simple as possible, but not simpler."
Smart heuristics achieve robustness through their simplicity, and achieve their simplicity by exploiting humankind's evolved perceptual capabilities. But not every heuristic can predict the future well, and the problem for contemporary researchers has become understanding which heuristic is best suited to which problem.
Simplicity creates not only robustness, but also transparency. One danger is that the complex methods so often used by managers in areas such as finance, economics, marketing and strategic planning can become an end in themselves, a ritual to impress others yet one that may not offer even an iota of additional marketplace punch.
Besides wanting to impress others, leaders sometimes choose to advertise complexity instead of transparency because they believe it has a calming effect on stakeholders, letting them know that state-of-the-art machinery is hard at work on the important problems. But failures to forecast, such as the present economic downturn, become evident over time. It takes but a few extreme events to teach people that it is safer to trust the transparent forecast (with its transparent limits), than the complex forecast (with its incomprehensible limits), when the predictive quality is the same.
At the beginning of this discussion, I cited another article with similar interests about fast and frugal decision making in the medical profession. Positing the prospect of a patient with heart troubles just entering a hospital, the authors of that article queried, "Should the patient be assigned to the coronary care unit or to a regular nursing bed for monitoring? How do doctors make such decisions? And how should they?" Though the decisions made by business leaders each day may not have the immediate drama of such a scenario, I suggest that you or other managers in your firm may soon be making life-and-death decisions about the future health of your enterprise during an era of economic uncertainty and expanded competitive attacks. Can you answer these two questions: How do managers make such decisions? And how should they?
Resources
Daniel G. Goldstein and Gerd Gigerenzer, 'Fast and frugal forecasting', International Journal of Forecasting (2009), doi:10.1016/j.ijforecast.2009.05.010.
Odette Wegwarth, Wolfgang Gaissmaier and Gerd Gigerenzer, 'Smart strategies for doctors and doctors-in-training: heuristics in medicine', Medical Education 43, no. 7, published online July 2009.
Markus Wübben and Florian V. Wangenheim, 'Instant customer base analysis: Managerial heuristics often get it right', Journal of Marketing 72, 2008.
Daniel G. Goldstein (dgoldstein@london.edu) is Assistant Professor of Marketing at London Business School.
For further information about Professor Goldstein and his research go to his personal website or his descision science news website.
Watch Daniel's podcast recorded at the recent Global Leadership Summit
Find out more about London Business School's Global Leadership Summit