The purpose of human decision making is to frame the right questions and validate the assumptions.
Analytics is the great tool to find the right Answer at the operational level, but human wisdom is essential to form the right questions at the strategic level. Generally speaking, the good distinction between strategic vs. operational decisions are defined by time horizons. Analytics and optimization require an objective function. These are hard to find when many people (executives, stakeholders, and experts) are involved. Analytics and Big Data are better suited to operational decisions which tend to be shorter-term and cyclical where high correlations between decisions and outcomes are more likely to be observed. Optimization works great for many operational decisions as well. A collaborative design process is better for strategic decisions. Most strategic decisions involve many people with different objectives. Optimization models can help but they don't quite get up to wisdom. That takes experience, courage, and leadership.
Machine computation vs. human computation: The difference between analytics and decision-making is mere of the boundary which being put between machine computation (formalization) and human "computation"/(judgment and expertise) as well as the scope and complexity of the problem one is tackling, and hence the degree to which a problem scope/boundary can be feasibly "encoded" into formalization versus left to expertise. Analytics is when the machine already suggests the best way, whereas in decision-making, one considers options with different utilities assigned (at least inherently). It's a bit like the distinction between two types of knowledge-based systems: decision support systems output "what-if" evaluation and analysis devices with expert systems output "if-then" production rules. The difference is that even though both have knowledge bases embedded; only the second also has reasoning/semantic apparatus embedded.
Analytics reflects the amazement of machine intelligence, and decision making requires the wonders of human wisdom. While information gleaned from historical data can be factored into any decision, the "wisdom" in the decision process incorporates the decision maker's tolerance for risk. Analytics can't tell a decision-maker how much risk to accept under uncertainty--you need a human to express that preference. Finding true wise human beings is not an easy task. Wisest end up far from societal conventionalisms and what is commonly recognized as wisdom. Human wisdom combined with machine intelligence generated by human beings programming the machines -and so not emotionally involved in the decision - is a good combination.
Human decision making becomes important and essential when there are multiple criteria; as in multi-objective optimization, and the decision changes according to which criteria have priority. There is fuzziness in the decision because there is fuzziness in conflicting criteria. With analytics, the machine is left with all the job of embracing all relevant criteria for decision-making. With decision analysis, it is either impossible or infeasible resulting in the human decision-maker taking the ultimate choice. That's why the decision analysis models are often in fact intended more for communication of the structure of the problem and collaboration on that to arrive at consensual decisions rather than coming up with single "point" estimate solutions.
Analytics is historically backward-looking, and decisions are forward looking. Although there is nothing in analytics that prevents it from being used on forward-looking projections, scenarios, and uncertainties. Bayesian Probability, predictive/prescriptive analytics are highly analytical, yet forward looking. The key is to orient the Analysis around decision variables tied to forecasted outcomes with uncertainty. Set the goals by shaping the questions: Do you want to maximize the mean expected result? Do you want to maximize the probability of making your goals? Do you want to minimize regret? Another issue is those analytics are no better than the metrics they're designed to measure (either maximize or minimize). Even where you can quantify things, people are increasingly coming to question the validity of some measures of value NPV has, at its core, an assumption is that there is always more stuff out there to get. Is this a good assumption in an increasingly resource-constrained world? If analytics are trying to maximize these parameters, will that lead to better, wiser decisions?
Analytics and decision making are interwoven as key business capability, applied to different levels of management - transactional, operational, tactical and strategic. With transactional/ operational level, it is more about analytics or optimization and at the strategic end of the continuum; it is predominantly about multi-criteria decision analysis with quantitative analysis support. Human wisdom is essential to frame the right questions, as wisdom requires a deep knowledge of values, which means finding acceptable tradeoffs between these competing objectives; which at the very least is a sensitivity analysis of the optimization parameters. Visualization approaches explore the trade-offs between different objectives.
Hence, how to strike the right balance between analytics and intuition, thinking slow and thinking fast; human wisdom and machine intelligence is extremely crucial, as under uncertainty, it will still be up to the decision-maker and his/her wisdom to assign probabilities and his/her willingness to take action to move the business forward and upward.
Hence, how to strike the right balance between analytics and intuition, thinking slow and thinking fast; human wisdom and machine intelligence is extremely crucial, as under uncertainty, it will still be up to the decision-maker and his/her wisdom to assign probabilities and his/her willingness to take action to move the business forward and upward.
0 comments:
Post a Comment