In uncertain times, smart leaders question their instincts and place value on data. And the smartest know how to use it.
Organizations are rapidly repivoting as they deal with a very different world. Decision makers face a lot of very difficult decisions, which they must take quickly, from adjusting supply chains, to deciding which product lines to focus on, to managing hospital capacity, to what vaccine research to pursue.
Much of the intuition that got them where they are no longer holds. Those wise enough to recognise their own limitations will now turn to data to help them glean insights into the best course of action.
This is the right approach, but that does not mean it is an easy solution. Data, too, is a reflection of the past. A trend predicted by data, does not hold if the driving factors of that data has changed. Data is only useful if you know exactly where, and how, to look.
Data has a reputation – unfairly – of representing an absolute truth. There is a risk that decision makers in urgent need of answers will leap on any pattern in the data, potentially leading to even worse decisions.
Decision makers should follow these three principles to avoid pitfalls, and make better data-driven decisions.
1. Maintain focus and align data to goals
All good decisions start with clear objectives. This is no time to be collating data and seeing what it throws up.
What are your objectives? If you run a manufacturing facility, your objective may be restarting operations whilst keeping employees safe. Parts of life sciences now have extremely clear objectives, developing therapeutics and vaccines for COVID-19.
Once you have clear objectives, and the decisions you need to take to reach them, then you should set out to gather data. Such decisions may involve examining existing data and gathering new data.
To reopen a shop floor, you may need to look at historical machine operating data, to review what aspects of operation and maintenance can be automated, or operated remotely, to reduce human interaction. To allow some people back to work, you may need to combine your own floorplans with airflow and infection models to redesign layouts.
Pharma must take decisions about which research avenues to pursue and how to speed the R&D process. This may mean gathering libraries of molecules and building AIs that can search them for candidates for COVID-19 drugs, or deploying new wearables to gather improved data on patient responses during clinical trials.
Tessella has long cautioned against gathering data for the sake of it. When critical decisions need to be taken quickly, a laser focus on finding the right data is critical.
2. Bring all the experts into one room
McKinsey points out that, at times of crisis, leaders should reject comfortable hierarchical models and “involve many more stakeholders and encourage different views and debate. This approach can lead to smarter decisions without sacrificing speed”. This is all the more true for data-driven decisions.
Leaders need evidence, but may not understand the nuances, biases and limitations of data and the models that make use of it. Determination to find quick answers may lead them to jump on promising-sounding outcomes before they have been properly validated, as has happened in situations as diverse as recruitment tools and premature birth predictions.
Technical people may be able to find the answers but not understand the problem, or the data. Business leaders need to frame the objective. Then domain experts need to decode the real world meaning of the data (eg what a machine readout means), and identify potential sources of bias that could lead to bad decisions (eg if a clinical trial misrepresents the wider population). Translators may be needed to help these different groups communicate effectively.
The best decisions bring together business, technical and domain experts to clearly define what the challenges are and understand what the possibilities data brings for solving them.
3. Keep things as simple as possible
There are times when it’s worth throwing lots of AI firepower at a problem. But bigger doesn’t always equal better, and complexity comes with an accuracy and explainability trade-off. Right now, we want speed and clarity.
If you have the right data (as opposed to just lots of data), then simpler models can often be used to find clearer and more accurate answers.
Where problems are brand new, we are unlikely to have the vast datasets needed to build a trustworthy neural network. Statistical techniques – such as Bayesian statistics, a modelling technique which involves assigning uncertainties to data – can be a very effective way of getting good predictions with very limited data. This technique has been used successfully, for example, in understanding aspects of risk associated with SARS-COV-2, such as inferring chains of transmission, which helps inform important decisions.
Start with a simple model. Iterate rapidly in short sprints aligned to clear goals. Quickly abandon projects that are not providing useful answers, and redouble efforts on those that are. Add complexity if it improves decision making, but never for the sake of it. Present and explain results clearly.
Beware AI salespeople promising the most advanced system available, or asking for access to all your data. A good data-based decision involves selecting the right data and the right model for that decision.
A company’s response to a crisis defines its long-term success
There is no doubt that COVID-19 will have an impact upon businesses and individuals. But it may also force companies to take a more targeted approach to data gathering and data-driven decision making.
Rather than focussing on amassing data, those with processes and mindsets that allow them to move rapidly - from identifying a problem, to targeted data acquisition, to outcome – will find themselves in a good position to make decisions smartly and quickly. Those that get this right will not only be best placed to weather the current uncertainty, but will come out the other side ready to thrive in the world beyond it.