On the 25th of January the first edition of our new event 'IMPACT live' took place. During this event thought leaders in Data Science and leaders in business came together and shared their vision on - and discussed about their experiences with - Data Science. In this article we will review the topics discussed and share practical insights on how to accelerate your business with Data Science.
The speakers of this event were: Noud van Alem (Partner RevelX, former Director Marketing EMEA UBER), Patrick van Buuren (Brand Director Sanoma), Mark Boelhouwer (Partner RevelX, former CEO Ricoh Nederland) and David de Jong (former Head of Innovation Achmea Non-life).
Many organizations talk about Data Science, and more and more organizations are experimenting with – and start implementing - Data Science solutions. But Data Science in business hasn’t lived up to its full potential yet: Data Science is revolutionizing traditional propositions, customer relationships, personalizing the customer journey and opening doors to explosive growth. Most of the well-known and biggest successes lie within big tech companies, but haven’t found their way into more traditional corporates and mid-market businesses… yet!
Often organizations are struggling with where they should start with Data Science, let alone implement it in an effective way. The cause of this might be the almost unlimited amount of possibilities in Data Science which make it hard to decide where to start and which route to take. Another factor why it is hard to start is the scarcity of experienced data scientists on the labor market.
To diminish the complexity of implementing Data Science in business we will discuss four important insights, based on the discussions and presentations during IMPACT live, that will help you to get started and implement Data Science solutions more effectively.
To effectively solve a problem, you always have to start with finding the root cause of a problem. For instance, take a turnover deficit, a sudden drop in repeat purchases or low customer satisfaction. These are all serious issues, but what causes these problems?
Before we start collecting and analyzing data, we should take a step back, take it slow, and find out what causes these problems. Although quickly gathering a lot of data may sound promising, often it only leads to confusion. We simply don’t know where to look and what we are exactly looking for in this enormous piles of data.
So after establishing a problem, there are two things we can do:
This might surprise some of you, but unfortunately not all answers can be found in the data – at least not right away. In these cases, qualitative research, also known as traditionally talking to your customers, can provide a solution.
Noud van Alem provided an excellent example within UBER to illustrate this insight. At UBER they noticed that a lot of people only used their service once. The root cause couldn’t be extracted from existing data, because users simply stopped using their service, not leaving behind any feedback to work with. After analyzing their data without success, UBER started talking to their customers, and found out there was a lot of uncertainty about the price and duration of an UBER ride.
Now the cause of the problem was found it was time to solve it. For this matter, in contrary to the initial qualitative customer research, the data was very important. Based on the data, the data scientists at UBER were able to make models that made accurate predictions on the prices of a ride (which is based on several factors such as for example; available UBERs, time-of-the-day, traffic delays) and show these price predictions real-time to their users in their app. This took away an important part of the uncertainty which make them use UBER more often.
This proves the most important growth hack is to know your customers better than your competitors know theirs. This can be established by analyzing data. but also by doing qualitive research.
TARGETED DATA COLLECTION: QUALITY OVER QUANTITY
After finding the root cause of the problem, targeted data collection is the next step in the process . The goal shouldn't be to obtain as much data as possible, but only the right data needed for the particular matter. This is data that adds value to your models and is really needed to solve the problem at hand. You can test this by asking yourself 'what am I going to do with this data? every time you’re gathering data.
This means that we have to research what variables influence the process in which the problem occurs. If the problem is low customer satisfaction for example, relevant insights can be found in the product range, customer service, or price data. Price may in turn depend on for example demographics, geography, availability, etc. After mapping out these variables, we can do a targeted data collection (or do a targeted search in data we already own). The result is maximized effectiveness of our data.
And that’s not the only upside. During IMPACT live many questions evolved around privacy concerns and how far we should go as an organization in collecting data of our customers? Targeted data collection minimizes the intrusion of the privacy of customers. We don’t have to know each detail of the customers day-to-day life to be of better service to our customers, just the relevant things!
DEVELOP, APPLY AND TEST THE RIGHT ALGORITHMS
After collecting the right data, the data scientist can start with selecting and applying the right Data Science algorithms to provide actionable insights. First, the algorithms will be tested and executed in a test environment. When the proposed models have proven to be successful the solutions can be implemented on larger scale and ultimately implemented throughout the whole organisation. So, start small and scale up gradually. Or in Mark Boelhouwer's words: think, tests, learn.
BE CRITICAL ABOUT THE OUTCOMES
We now know the root cause of the problem, we collected the right data and developed the right algorithms. This means we’re ready for the last step in solving the problem, which is interpreting the results, and being critical while doing this.
This means not blindly accepting the outcomes to be true, because in every step taken, little mistakes can slip into the process. For instance, data entries can be:
In the end, data errors lead to incorrect interpretations and possibly poor business decisions. Aside from these possible slips in the data itself, it is also possible that the algorithm isn’t programmed right or should be updated from time-to-time because of environmental changes. So be critical on the data you use and keep track of what the algorithms are doing. Try to find and work with qualified people, who can discover and rectify these errors.
Mark Boelhouwer experienced the latter himself during his career. An important lesson that we can learn from his experience: when data presents information that is counter intuitive, don’t blindly trust the information provided, but use your intuition and investigate what is really going on.
At IMPACT live we have learned that not all problems can be found in the data and we sometimes need to talk to our customers to find the root cause. But when we get a clear picture of what is really happening, data science can be of great added value in problem solving. Starting small helps you to get started and to get results as soon as possible! When the outcomes are good, you can start scaling up within the organization. And last but not least, make sure that business and analytics departments work together closely so that the limited resources are used to solve real world problems!