Right Data, Wrong Decisions
Does having the right data automatically mean your organization is bound to make the right decisions? More often than not — this is not the case. This presentation from Data Leaders Quarterly Members Briefing takes this problem head on.
Pascal Clement, Chairman and CDO at Data Leaders shares his insights on how organizations can leverage good data to make equally good decisions. Specifically, he talks about:
- Understanding the concept of noise
- Explaining the way people make decisions
- How to correctly extract valuable information from data
Common Problems In Organisational Decision Making
An April survey conducted by Data Leaders sought to figure out how organizations felt about making decisions about their collected data. Surprisingly, most of them answered by saying having good data alone is sufficient for making good decisions. Particularly:
- When asked whether they believe their company is making decisions based on appropriate data, more than 60% of respondents answered that they somewhat agree.
- When asked whether these decisions based on appropriate data are the right ones, over 30% answered that they strongly agree and over 40% answered by saying they somewhat agree.
Making informed decisions is very hierarchical, first of all one starts with either having good data or bad data — these are always the primary influencers. Then you consider whether you have the right data model or the wrong data model. There is also, however, the layer that consists of bias and noise.
Noise, in this context, is different from bias in that it is based on the gap between the analysis of identical cases within an organization. Some examples of noise include weather, time, health, and stress.
The next concept is bias — which Pascal exemplifies by a study conducted by the United States Air Force in World War II on the points of vulnerability in their aircraft that were shot at. Their analysis was biased in that they only surveyed planes that survived, not ones that were shot down.
Making decisions fast is often based on gut feelings, but this can also lead to biased decision making. What automatically makes sense to the mind may not always be correct. When asked to identify the most common gender-pattern for 6 babies born in the morning, most respondents answered that it would be a healthy but uneven mix of girl and boy. The fact is, the pattern being exclusively girl, exclusively boy, or having an equal number of girl and boy is just as likely.
An example of noise in the decision making process in action was exemplified by Pascal using golf — noise in this case being a negative or positive connotation that may influence the decision.
In a game of golf, if the ball is in the same exact position on the green, would players be more motivated to putt to par (fear of failing) or putt to make a birdie (desire to win). The results showed that the negative connotation, in this case the fear of losing, was the greater driver of the decision making process.
Augmented Analytics: Where Are We?
Using tools like machine learning and deep learning can give organizations the edge when it comes to making the right decisions using their data — allowing them to have faster data preparation, improve their overall data literacy, reduce bias and noise, and reduce the time required to receive meaningful insights.
Of course a machine learning-based decision making process would take several iterations and a lot of time to get right. This takes a lot of time to measure. Not to mention that the extensive use of data would lead to a lot of security concerns, particularly because these are black-box approaches and no one exactly knows how the machine learning models process the data.
These approaches also force people to go against their intuition most of the time, and the process can be rather siloed from business analytics.
Automated Decisions, The Next Step For Every Organisation?
Automated decision making is the process of making a decision by automated means without human intervention. Industry trends that a lot of industries are starting to move towards this type of decision making because of the accuracy of the results.
This is best exemplified by the financial services industry, where the conditional approval of loans are made completely by algorithms, initially without any input from a human loan officer.
Follow Up Discussion
Following the main presentation, Pascal was joined by Decathlon’s Vice President of Data Value, Didier Mamma, to extrapolate some insights from the decision. Didier talked about some key long-term actions organizations need to make:
- For you to make good decisions based on your data, you have to properly define your problem. You cannot hope to put your data to good use if you don’t even know what the problem truly is.
- Organizations have to have a deeper understanding of the data that they are working with, beyond just using advanced algorithms in machine learning.
- Outcomes need to be measurable. If you have discrepancies, for example, you can use your metrics to prove how the outcomes were achieved. This is particularly useful when the results from the data are counterintuitive.
Pascal expounds on these points by saying that, fundamentally, there are no right and wrong decisions per-se, the main issue is the gap between decisions that are made between similar data points. Because in an organization, there needs to be a consistency with decision making with regards to similar datasets.
Finally, Didier explains that he prefers the use of these new tools not to challenge human intuition, but to augment it. Intuition and machine learning can work together instead of separately. When experienced human intuition is systematised and backed up with rich data and computing tools, better decisions will come out as the result.