#11 - Integrative Thinking, Defence vs Offence Data Strategy, Effective ML in business, and Rolls-Royce's use of Big Data
How to be creative in thinking, choose the right data strategy, be effective in doing machine learning, and use big data to gain an edge like Rolls-Royce.
This Week’s Principle
Integrative Thinking in the eyes of Roger Martin (link)
Consider A - who graduated from Harvard, and B - who didn’t go through college. They are both very successful, but have opposite records of academic achievements. What does that make you think of the condition for success?
This question is an example of Integrative Thinking - a thinking process championed by Roger Martin - which can help you generate an innovative course of actions to achieve results.
There are four key stages to the Integrative Thinking process.
Articulate the models: identify 2 extreme and opposing answers to the problem.
Examine the models: compare the 2 models, then think about your current problem and define which elements of each models do you want to keep in the new (integrated) model.
Explore new possibilities: Create new integrated models, they are prototyped models so ideas should be created freely first.
Assess the prototypes: Test your prototyped models. This step is about small trials and errors. Identify the conditions that have to hold true for the models to work.
Going through this Integrative Thinking process, one will make better and more creative decisions. Nodding his head, Peter Drucker once stated:
“The understanding that underlies the right decision grows out of the clash of conflicting opinions and out of serious consideration of the competing alternatives.”
This Week’s Strategy
Defence vs Offence Data Strategy (link)
Data Strategies can be categorised into two types - defence and offence:
“Data defence is about minimising downside risk: ensuring compliance with regulations, using analytics to detect and limit fraud, and building systems to prevent theft. Data offence focuses on supporting business objectives such as increasing revenue, profitability, and customer satisfaction.” - Thomas H. Davenport, Hbr.org
Companies need both defence and offence to succeed, however, they may lean toward one depends on their industry, business objectives and regulatory landscape, as illustrated in the diagram below:
Companies can also shift between these two positions, depending on the changing landscapes and data maturity. When choosing to shift, you can also consider what Sun Tzu said on the Art of War:
“The general who is skilled in defence hides in the most secret recesses of the earth; he who is skilled in attack flashes forth from the topmost heights of heaven. Thus on the one hand we have the ability to protect ourselves; on the other, a victory that is complete.”
This Week’s Operation
Machine Learning in Business - things a data science course wont teach you (link)
As a data scientist, you train and test your models multiple times. The accuracy rate is 99,9%. Great! Your job is done here?
Not so much, according to Guillaume Colley. To be effective as a data scientist, there are some key practices to follow:
Think twice about selecting the target: the target variable must represent a behaviour, not a data point (e.g. click, buy, etc).
Testing should be done on untreated dataset: test dataset should be real and untreated, never re-balance the dataset before splitting into training and test data.
Choosing the right cut-off level, depending on whether the priority is on high precision, or high recall.
Frame the metrics in a business-friendly manner. Instead of saying “The model has 15% precision for an audience of a thousand”, say, “The model’s audience of 1000 customer is expected to yield 150 purchases, compared to only 5 purchases if the audience is randomly selected”.
Editor’s note: I know the newsletter has become quite long recently, but I encourage you to read on to the last section here. I often find This Week’s Impact the most inspiring segment. The thought of data being used skilfully - generating the right insights for the right actions to be taken - has kept me motivated. Either that, or my morning coffee …
This Week’s Impact
Rolls-Royce’s Use of Big Data in its Engine Manufacturing business [1]
For Rolls-Royce, a company operating in the engine manufacturing business, failures and mistakes can be costly. It can be billions of dollars, it can even be human lives.It is therefore crucial for the company to monitor the health of their products to spot potential problems.
To tackle that challenge, Rolls-Royce uses big data in three areas: design, manufacture and after-sale support
Design: Terabytes of data are generated on each jet engine simulation. These data are then transformed and visualised to study potential extreme behaviours, which are used to make design decisions.
Manufacture: IoT applications in the manufacturing process generate lots of data for Rolls-Royce (e.g. 0.5TB of data on each individual fan blade). These data are fed into an automated quality control scheme to monitor manufacturing outliers and anomalies.
After-sale support: Sensors are attached to engines to record every tiny details about their operations. Operation data are received via wireless transmissions from the aircrafts, containing a mixture of engine performance, engine power, and interesting events during the flight. This informs Rolls-Royce of any factors and conditions under which engines may need maintenance.
With the right focus on advanced data analytics, demonstrated through research partnership with university and investment on intuitive visualisation, Rolls-Royce has gained a lot from big data:
Improve the design process: decrease product development time and improve the quality of their products.
Optimise production process: allow faults to be eliminated from future products.
Improve after-sale support: maintenance actions are identified days or weeks in advance, increase scheduled maintenance without flight disruptions.
Summarised insights from:
[1] Bernard Marr, Big Data In Practice, Chapter 4
That’s it for this week! If you enjoy or get puzzled by the content, please leave a comment so we can continue the discussion. Throw in a like as or share as well if you know of someone who may enjoy this newsletter. Thanks!