I have never seriously thought about it before.
This time, I want to walk outside the data science garden and look hard at decision making by itself.
What’s covered in the following:Analyze four types of decision making to inform your data science workUse a N:N model to help you navigate the complex dynamics of today’s decision making processesDecision Makings Have Distinct CharacteristicsAlthough we can universally define decision making as the action to select one option from alternatives, the process to agree on a selection differs greatly from one decision making to another.
Using the number of distinct stakeholders, we can classify decision making into individual decision making and corporate decision making.
Individual decision making is familiar to most people because it is a part of everyday life.
Individual decision making has one or a small number of stakeholders, usually decision maker himself or herself.
Individual decision making is focused, cost efficient, and has a clear accountability.
Individual decision making takes place both in personal life and workplace; you can decide on your college major and you can decide on the best way to present at work.
Individual decision making usually has limited consequences.
“Decision-making involves the selection of a course of action from among two or more possible alternatives in order to arrive at a solution for a given problem”.
— Trewatha & NewportWhen the number of stakeholders is large, we have corporate decision making.
Corporate decision making usually takes place in various levels of organizations, and has earned a bad reputation of being slow, costly, political and lacking accountability, despite the fact that more rigorous evaluations and multiple perspectives are considered in corporate decision making.
In large organizations, a small decision could end up with rounds of meetings and planning.
In data science exercises, we usually focus on individual decision making where stakeholder management presents no issue at all.
However, in real world applications, we often apply data science to decision making in organizational context with many stakeholders, hence we primarily work on corporate decision making.
Because dealing with stakeholders is never a thing among data scientists, those data scientists who have a channel interest in modeling techniques are likely to feel shocked by the complexity of corporate decision making.
In corporate decision making, stakeholders have competing perspectives and interests, rely on both quantitative and qualitative information for decision making.
Data science results may not appear sexy but rather powerless in the face of bureaucracy.
When a data scientists believe data should predominately drive decision making (data-driven decision making ideology), this data scientist may end up saying, “the result is clear, they just don’t listen”.
We can further classify decisions into operation decisions and strategy decisions.
Operation decisions are characterized as recurring and structured; such decisions concern day to day running of the business and are made by middle managers and front line employees.
Strategy decisions are characterized as non-routine and unstructured; such decisions concern organizational policies and strategies, and are made by top management and executives.
Successful data science applications today are mostly in operation decision making, because operation decisions are more quantifiable, have large available data for modeling and have less stakeholders.
Think about house price estimation, loan application approval model and user retention prediction.
These successful data science applications are operationalized as code and have limited stakeholders intervention.
When we try to extrapolate these successes to other categories of decision making, we need to understand decision makings have distinct characteristics.
Build Influence to Convince a VillageToday’s decision making landscape is very challenging.
Leadership structure is flat, projects are ran by cross functional teams, therefore we often do not have a single authority for decision making.
To support decentralized decision making, data scientists have a village to convince.
As we talked about corporate decision making earlier, decision making complexity increases greatly when the number of people increases.
In flat leadership structure, a single decision maker or power center is less likely to retain big enough authority, therefore the final decision may not get honored completely by the workforce, which leads to half heart execution and poor decision making outcome.
In a simplified world, a data scientist need to convince one decision maker; in today’s corporate, a data scientists is just one of many to convince many decision makers.
We need go from a 1:1 model to a N:N model.
In N:N model, from the perspective of a data scientist, the data scientist need deliver data stories to multiple team leaders from multiple lens.
In my work, a sign of success is not the applause at the end my presentation, instead a sign of success is the referral I get to present to another business leader.
One more presentation sounds alright, but it is not easy to tell persuasive data stories to N stakeholders, because the data science work has to be robust enough against N dimensions of scrutiny.
In the communication with N stakeholders, data scientists should not only know their different perspectives but also their different decision making preferences.
Some want others to make decisions for them, some frame questions and interpret results, some can talk about database systems and modeling, and some just want data to prove existing belief.
Data scientists need to be prepared to encounter N decision makers out there.
In N:N model, from the perspective of a decision maker, the decision maker takes input from multiple sources to form an opinion.
Decision makers are consistently fed with information by internal teams and external consultants, through emails, PowerPoint and spreadsheets.
In large corporate where teams of data scientists are employed, due to data sources variation and methodology differences, a decision maker may hear opposite decision suggestions from different data scientists, therefore a data scientist competes with other data scientists and other professionals to give the best insight for decision making.
Decision makers never rely on one input, hence do not take it personally if your suggestion got ignored.
Billionaire Ray Dalio perfectly summarizes this dynamics.
“The best decision are made by an idea meritocracy with believability-weighted decision making.
It is far better to weight the opinions of more capable decision makers more heavily than those of less capable decision makers.
This is what we mean by Believability Weighting.
” — Ray Dalio, PrinciplesLeadership is EverythingDespite the progress in data science, the ability to make informed decisions is a leadership trait.
Bold leaders can hear divergent views and still make great decisions.
The job of data scientists is to bring rigorous insights to decision makers, but data scientists can not change the leadership trait of decision makers.
Technology helps, but leadership is everything.
Decision making is more than quantitative problem solving.
Data scientists should not limit themselves to problem solving skills, but to build leadership in persuasion, influence and decision making.