Speakers:
Thomas Ebermann (Liip)
Daniel Meister (Datahouse)
Data Science
On Wednesday 27 February the 10th Digital-Shift event took place at Hinderling Volkart, on the topic of Data Science. All too often decisions are made according to the HiPPO model (highest paid person's opinion). Data Science, by contrast, offers a way to ground decisions in scientific foundations. Since ever larger volumes of data can be used, and methods such as machine learning can make increasingly precise statements, profound changes are taking place in our decision-making culture.
Thomas Ebermann of Liip introduced the topic. He explored the question of what a company needs in order to be able to make data-driven decisions. Within a company, a great many preconditions must first be met before methods such as machine learning can be applied. It may be that you have insufficient data to answer questions like "What do customers want?". It may also be that you have plenty of data but work in a hierarchy where others are the decision-makers; or in a culture where a task "has always been solved that way". And finally, a company may simply find no time for Data Science. A data scientist combines the competences of a statistician, a computer scientist and the domain-specific knowledge of an expert in a single person. The Data Scientist stack consists of:
- Communicating Data Science well
- Automation & experiments
- Useful reporting of the data
- A stable production pipeline
This is how the data scientist provides a company with foundations for making decisions and helps plan for the future. The most important skill is generating insights on the basis of problems and data — for example, why a product is not catching on in a given market. If Data Science is used only to gain efficiency, you give away its most important potential: with Data Science you can decide in the first place whether you should be doing something at all. The dumbest thing that can happen in a company is to do something very efficiently that you should not be doing at all. That is why the principle "Practice over Theory" sums up concisely the mindset of a company that practises Data Science. Employees can "just give it a go", because doing so generates the data on which decisions can be based, and you do not have to dig out the dusty theory books. For Thomas Ebermann there is also a culture stack:
- Achieving data maturity
- Openness, transparency, ethics, principles…
- Learning to experiment
- Learning to act on the basis of data and starting with the question
- Believing in the value of data
However, data can quickly suggest a gut feeling to you that is not necessarily correct if you no longer dig deeper into the topic. Another stumbling block can arise when people are the data points. Here the Heisenberg principle applies: that is, when you observe a data point, it changes. People often learn very quickly to game a measuring system.
Daniel Meister of Datahouse gave us insights into why, within an organisation, Data Science projects fail. A Google search for "data science project failure rate" returns roughly 396,000,000 results, many of them citing a Gartner study which states that 85% of these projects fail.
Even in the first phase, the business analysis, much can go wrong, because the focus is often in the wrong place. The question must always be perfectly clear: "Where can our (core) business actually still be optimised? What additional information is needed for that? Is there a realistic chance of obtaining this data?" The project threatens to fail at the very task definition if the question tends more in the following direction: "Facebook makes millions in profit with matching ads next to holiday photos. Do we also have images we can process with deep learning?"
To minimise the probability of a Data Science project failing, the following should be in place:
- Easy access to complete, correct and relevant data
- A good mix of experts and decision-makers from different areas
- A clear question / objective and a constant critical exchange about it
Data Science combines statistics, computer science and domain-specific knowledge, so in every field there are sources of error. Common sources of error are, in addition, wrong or faulty data, a missing mix of expertise/competences in the team, or no clear questions to be answered. In IT there are currently a great many easy-to-use tools for Data Science, but it is a challenge to choose the suitable tool for the question at hand. Another well-known problem in software development is implementing more complexity than strictly necessary. In statistics you can summarise too much, so that the result allows a false interpretation. With domain-specific expert knowledge you may simply start from an inappropriate question. These stumbling blocks apply in every discipline, and if even one of them is hit, the foundations for a decision are missing.
In summary it becomes clear: to use Data Science correctly, a company must be able to build a bridge between the goal (e.g. the corporate strategy) and the possibilities of the tool (Data Science). Only then can Data Science answer business-critical questions from which strategies or actions can be derived. There are not a few examples where companies spend enormous sums on Data Science without the understanding to deploy these resources effectively.
We see that a quiet revolution is under way, because in the end the company culture plays an important role. Are you willing to believe in the meaningfulness of data, and do you therefore build a transparent corporate culture that makes actions analysable and success and failure traceable? Or do you prefer opaque processes where hierarchy and politics tip the balance?
Explore more topics on how technology affects business, society and culture.
Don't want to miss our next event? Subscribe to the newsletter.