How to improve decision-making with useful insights
In our previous article we talked about how to start collecting dynamic and static data as well as how to ensure your data is coherent. In this article we’ll show how to improve decision-making with useful insights. You will learn ways to generate useful insights about your customers’ behavior or your product’s performance.
TYPES OF INSIGHTS YOU CAN GENERATE FROM YOUR DATA
1-) Segmentation: Start by segmenting your customers/users into homogeneous groups. Here you can play with your dynamic and static data and create clusters of customers with a combination of demographic and behavioral data. Could it be interesting to know that the subscribers of a French newspaper with the highest customer lifetime value are aged between 40–45 years old? Or that the subscribers from in and around Marseille are churning more than others
2-) Prediction: With enough data and the right tools, you can create prediction models that will anticipate future behaviors or interactions. Using the example above, you could identify a leading indicator that predicts the likelihood of churn. For example, it may be possible to determine that if a subscriber views fewer than 5 articles a week, he or she is 95% more likely to churn relative to all other subscribers.
3-) Data Visualization: in order for your data to be useful, it has to be easily consumable, so you can act upon it and make better decisions. As Edward Tufte defined in his 1983 classic The Visual Display of Quantitative Information, “Excellence in statistical graphics consists of complex ideas communicated with clarity, precision and efficiency.” There are several ways to visualise data, from tables and charts, to geographical maps and heatmaps, etc. Nevertheless, there are simple principles you should be careful to respect. As Tufte puts it, data visualisation has to:
- induce the viewer to think about the substance rather than about methodology, graphic design, the technology of graphic production or something else
- avoid distorting what the data has to say
- present many numbers in a small space
- make large data sets coherent
- encourage the eye to compare different pieces of data
- reveal the data at several levels of detail, from a broad overview to the fine structure
- serve a reasonably clear purpose: description, exploration, tabulation or decoration
- be closely integrated with the statistical and verbal descriptions of a data set.
Using our example of churn, by analyzing the data of our newspaper’s articles and the churn of each individual user, we might discover that readers from cities that are not covered by local news are churning much more than those that are covered by local news. We could visualize this data by using a chart with bars of two different colors, the first color would display the average churn for regions with both local and national news coverage, while the other would display churn for regions with only national news coverage.
We could also display a chart with each region represented as a bubble, where the size of a bubble would reflect its relative churn rate. The bubbles would be plotted on a two-dimensional chart, with the Y-axis measuring the churn rate and the X-axis showing the number of subscribers. We could even go further and use bubbles indicating the number of subscribers of the newspaper each city is losing every month. The larger the bubble, the higher the loss. Such a method would allow us to identify regions with a high number of subscribers and high churn rate, with the assumption that these regions have a higher financial impact for the newspaper. If we plotted these cities on a map, we would discover that they are all located in the South of France.
4-) Recommendation: now imagine if we could go one step further and take the same clusters described in our segmentation example above and identify which types of articles each of these clusters prefers. With this information, we could create a recommendation model based on readers’ past behavior. For example, our model may identify that readers who read the theater reviews every week, may more likely appreciate recommendations for new restaurants. As a result, our recommendation system could be applied to recommending new restaurants either on our newspaper’s website or via automated email campaigns to readers of theater reviews.
5-) Automation: imagine you are the Editor in Chief of our newspaper. You are in possession of the information described above. You know that users who read fewer than 5 articles per week are very likely to churn; you also know that the cities without local news coverage have higher churn rates and they are located in the South of France, a region for which you don’t have any dedicated news team. Additionally, you are aware that there is a strong correlation between certain types of content such as theater and restaurant reviews. You decide along with the newspaper’s management team to increase the local news coverage in the South of France by opening an office in Marseille. However, as you must ensure that the readers that are more likely to churn continue to view more than 5 articles per week, you begin implementing an automated process. In order to re-engage subscribers, each time a reader views less than 5 articles per week for 2 consecutive weeks, he or she receives a personalised series of local news articles via email or on your newspaper’s website. The decision to send the recommendation when the 5 article rule is triggered is thus completely automated. As time passes, and data accumulates from interactions with email recommendations, our model gets increasingly robust and the recommendations potentially more personalized, with the goal of increasing sales and decreasing churn.
IMPLEMENTING AUTOMATION AND RECOMMENDATION SYSTEMS
Although seemingly simple, developing and maintaining these automation and recommendation systems can prove to be a complex process. For this reason, at metriq we have developed a methodology we teach internally and apply to all of our projects and that help companies develop their transformation through data, without disruption.
1-) It all starts with an educational course to understand what you want to do with your data. This program emphasizes a methodology to reduce risk, while clarifying, quantifying and tracking objectives.
2-) Once your objectives have been defined, we identify how much data you have and how useful it is.
3-) This data audit enables us to define and implement a strategy adapted to your organization’s stage:
- If you have no data, we work with you to define what data you need and how to collect it.
- If you have data, but it’s not useful, we identify ways of collecting new data or enriching your existing data to make it useful.
- If you have data and it’s useful, we structure it for some application using a model or algorithm.
4-) When you have enough data, we start building models to test the impact of decisions on your objectives.
- If you require decision-making tools, we use business intelligence software and data visualization techniques to display KPIs and metrics related to your objectives.
- We create prediction and recommendation systems to estimate what could plausibly happen in the future, based on past observable experience.
- And finally we automate repetitive processes and workflows, in order to reduce errors, with the goal of increasing efficiency and reducing cost.
If you are interested in learning how metriq can support your company in making data more useful, don’t hesitate to schedule a call with one of our experts.
Originally published at www.metriq.io on July 26, 2017.