By: Mariano Allevato
Most often companies think they are data driven, however, being data driven is not just collecting data that ends up being mostly unused. Furthermore, putting together dashboards, or creating a “fancy” ML/AI model, does not mean you are data driven unless all of these are used in the appropriate manner: You have got to take actions and experiment.
What follows is a high level road-map on how your company can leverage data, combined with machine learning and AI, together with dashboards and experimentation to have a real impact on the business.
We have already talked about the importance of data collection in a previous article. However, it’s worth touching on it again: Collecting data is essential (you are most likely collecting it and you do not know it). Without it you can’t measure what’s going on with your customers, and without measuring you can’t understand their behaviors, create hypotheses or run experiments. And without all of these, your company’s performance will suffer. Moreover, you are missing good opportunities to increase your customers’ engagement with your brand.
For example, let’s say you want to understand how retention and churn rate are affecting your business, you would need the data collected from any platform that interacts with your customers – website, mobile app, POS system, marketing systems, surveys, media partners, etc.
Your data should be collected onto a Data Lake to make it not only more complete, but also more efficiently accessed.
Analysis and Machine Learning
Once you have data you have to analyze it. This initial analysis does not need to be too complex, after all, you need to start to understand the “basic elements” of your business from a data POV and your customers’ behaviors. The analysis can be as simple as “how many customers are purchasing through time”, “how many customers are engaged with the brand and how many are not engaged”, “how many customers visit your website”.
To continue with our previous example, you would want to go a step further and not only understand what is happening with your churn rate, but why customers churn and how to prevent them from churning! It is easier and less expensive to retain a customer than to acquire a new one.
You should create a machine learning model to predict which of your customers are at risk of leaving your brand. This model would take as input behavioral data from your customers and in turn, it would provide a list of customers which have a high probability of churning. This process should be automated as a robust system that refreshes results weekly.
KPIs, reports, results, etc. need to be visualized. That’s where dashboards come into play. Dashboards are not only for analyzing basic metrics such as sales, orders, active customers, etc. They also used to express and make readily available the results of Machine Learning models. You would visualize how your retention strategies, campaigns and experiments are performing over time as well as how churn is being affected, and also how good the machine learning model is in predicting churners, and adjust it over time.
Dashboards are not the final product, but a way to measure and define new hypotheses to test. Stand alone they provide limited value.
Now that you have data, you have analyzed it, have a churn model and got a list of highly probable churners and are able to visualize them in dashboards; What do you do? You have to create actions focused on retaining those customers in the form of experiments.
It is paramount to note that these experiments are and should be to the benefit of the business, not a “science experiment”.
For example, target the list of customers produced by the churn model with an email retention campaign just for them. However, what’s the message? What’s the offer? Which subject should you use? What time should it be sent?
These are questions that get answered over time, but as a starter, get two randomized samples of customers and send a different message to each group. Measure it, see if there is any statistically significant result, and you get a starting point “winner” to continue your campaigns and experiments.
This cycle of collecting, analyzing, hypothesis creation, test set-up, running the test, and looking at the results is a process that should be continuos and ongoing.
Still not sure about this process?
Here is another example going from data collection to execution.
Problem: You want to understand what people are saying about your Brand, and precisely what they are reviewing on your products on the Amazon platform.
Collection: Create a system to extract data from reviews in Amazon or any other platform to get to the data.
Analysis: Create a text and sentiment analysis model to understand what the reviews are about, what they are saying about your brand, which words are most often used, which are negative and which ones positive. Again, data extraction and analysis would be automated.
Dashboards: Create a visualization that showcases the sentiment analysis, with a clear representation of negative and positive comments on the brand.
Experimentation: You are now able to concisely understand what customers are saying about your brand, what they do and don’t like. Let’s say that you are looking at customers that are reviewing one of your products poorly but with the analysis you have done you detect that their comments reflect they are using the product wrong. You can try to perform marketing actions to educate customers on the right way of using them. Create some hypotheses and test them to improve customer satisfaction.
Again, going from the observation to the execution is the meaningful step to deliver value.
In conclusion, there are many aspects of businesses that can benefit from experimental data science and machine learning, but in order to do so, a culture of experimentation within the company is needed to execute ideas and turn data into actions.