From Hunches
to Data-Driven
Decisions
Beware the true, but useless
Big Data has been a hot topic for years. In the age of analytics, we’ve been told that we must have big data and can’t compete without it. Many companies have responded. Every customer interaction is tracked. Every purchase. Every contact point. All fueling sales data, forecasts, and market share. Advanced organizations also consider outside data points like weather, economic trends, and social media information.
However for many organizations, this data lives in a series of dark caves where only the bravest data scientists are plumbing its depths.
This may be because few know how to access this data, why they might want to, or because important information is difficult to find behind complicated interfaces. Sometimes employees with little data literacy are given raw data with hopes of making sense of it. These issues can lead to false correlations and a focus on validating hunches, rather than on making data-driven decisions based on actionable insights.
But there is hope! Whether you feel overwhelmed by the sheer amount of data your company has collected or you feel left out of the analytics game, a few helpful concepts can start you down the path of data empowerment.
The first concept is to beware the true, but useless, originally coined by Martha Cotton, the Group Design Research Director of Fjord North America. While true in many areas of life, this concept is essential in the world of analytics.
No matter how many predictive models or machine-learning algorithms we create, it is up to us to decide what is actually meaningful.
One of the cardinal sins of big data is collecting so much irrelevant information that false correlations are found everywhere.
For example, a strong positive correlation between pool drownings and Nicholas Cage movies released each year1 may exist, but there is clearly no causal relationship there. At least, we hope not.
When moving from hunch-based to analytics-based decision-making, it is important to draw on the experience of your people rather than following analytics blindly. Until artificial intelligence can match common sense, it is still up to humans to help know what is critical, actionable, and impactful information, versus what is true but useless.
The second concept to keep in mind is to have clearly defined goals before you interact with data. A research professor in grad school drilled into my head that “if you feel like you’re on a fishing trip just casting your lure into the ocean of the world’s information, then you’ve completely wasted your time.”
If you hope to learn something by just throwing data at a problem, you will most likely be led astray. You may know exactly what you want.
You may know you have a problem, but not know the root cause. Or you may know the problem and cause, but are unsure of what action to take. Each of these starting points likely needs a balance of three areas: report, explore, predict. These three areas may use separate tools, but ideally will use a single unified system.
A reporting engine takes known questions with known inputs, and automates the communication of these key metrics. This might be a static report for a less data-savvy employee or simply someone who needs information in a preprocessed, efficient vehicle. Oftentimes organizations are creating these reports manually. Transforming them into automated processes can help free up the creators to actually begin exploring root causes.
If the questions or drivers of the questions are vague or complex, this is the time for a more discovery-based tool for exploration. Here, too, we want to avoid a raw dump of information and instead present a summary of critical information. From there, the user can decide if they want to dive further or explore complex relationships.
The final area, prediction, will usually be done by data scientists and advanced users. These groups are forward-facing in their approach, because they ask hard questions that go beyond the typical user’s need. In a modern organization, the information gleaned is not kept within the data science community, but rather turned into predictive models or fed into insights engines that help less data-knowledgeable users make forecasts and data-driven decisions.
Most large companies could use all three of these types of systems, as leaders will always need to report performance and set strategic visions. With a powerful and needs-based toolset, they can move away from hunches about performance. When they have questions about root causes or need to understand more granular data, they can use a discovery tool to explore programmatically generated insights, rather than raw data, with the option to dive in deeper. But in order to create these engines, you need data scientists who can get in the weeds, tweak models, and push the organizations knowledge forward.
At Fjord Chicago, our data and design practice pairs designers, who make information and services accessible, with data scientists, who find actionable insights and use innovative methods to express these back to users. We have the knowledge and experience to help craft a new culture with the best tools so that your organization is empowered to make smart data-based decisions.
Footnotes
1. Spurious Correlations, Tyler Vigen, 2018, www.tylervigen.com/spurious-correlations.
