Combining Predictive Analytics with BI

Predictive analytics is a category of data analytics aimed at making predictions about future outcomes, based on historical data and analytics techniques such as statistical modeling and machine learning. The science of predictive analytics can generate future insights with a significant degree of precision. With the help of sophisticated predictive analytics tools and models, any organization can now use past and current data to reliably forecast trends and behaviors into the future.

The Scenario

As an auto body shop, the company had an interest in broadening its customer base and winning market share from competitors in the US. Optimizing customer retention is not the only goal, the shop wants to optimize other aspects such as inventory management, marketing attribution, and customer satisfaction.

The Predicament

With a large list of customers, it is simply not possible to "guess" certain behaviors such as; which of these customers might leave? What are they planning to buy next? or what are the most effective channels for advertisement? With reports, charts, and traditional statistical analyses, the best they could do was glean insight for general reasons. Without knowing customer-by-customer how those trends applied, they were unable to target effectively.

The Solution

Using machine learning to automatically determine the future behaviors of a customer who will probably churn, buy, or consume an ad. AI algorithms have distinct advantages over conventional analyses for these kinds of tasks. Unlike statistical methods, machine learning does not require any guessing at which variables will be predictive of behaviors.

Combining BI with AI

Let me clarify this first, the “intelligence” in AI refers to computer intelligence, while BI refers to the more intelligent business decision-making that data analysis and visualization can yield. Utilizing AI we move from descriptive and diagnostic to predictive and prescriptive.

Domo + Squark

Domo is one of our BI connectors.

For this scenario, I followed these simple steps to obtain my predictions using Squark:

1. Connected and pulled data from the cloud server of the Auto Body Shop and uploaded the data to Squark (it's as easy as drag and drop). This data set I considered my "training file".
2. Selected the dependent variable (Target). What do I want to predict?
3. Selected the independent variables (Predictors).
4. Uploaded my "production file" (this file contains all the customers to predict).
5. Deployed, Squark's codless solution tested hundreds of possible models and algorithms to find the most accurate one for my application using an Automated Machine Learning process. (AutoML)
6. Predictions are ready to be used in Domo!

Once I got my predictions, I built my DataViz dashboard.

Understanding my DataViz Dashboard

DataViz is the graphic representation of data. It involves producing images that communicate relationships among the represented data to viewers of the dashboards.

A dashboard is an information management tool that visually tracks, analyzes, and displays KPI's, metrics, and key data points to monitor the health of a business, department, or specific process. The big difference is that usually dashboards are fueled by business questions but when I combined with AI and predictive analytics I am fueling my dashboard by answers.

Churn Analysis.

The importance of predicting churn is not just to find which customers will stop purchasing products but, to also understand why? Witch variables of my customers had the biggest weight regarding the decision to leave my company?

• Churn Prediction: How many clients the body shop is going to lose? (binary classification) In this case, 23% of my leads.

• Variable of Importance for Churn: In this section, thanks to Machine Learning I am going to understand why. Why my customers are going to leave the body shop? What are the variables with the highest importance in my prediction? In this particular case, we can see that Customer Satisfaction is highly influential. Thanks to Squark capability for Natural Language Processing (NLP) I found that based on the reviews section of the web, most of the customers were unhappy.
• Churn by State: Where are these customers located in the US? Now that I know California and New York are the states with a higher possible churn rate, I can run specific actions to prevent it. Actions could include: Special offers/coupons in these states? Better training for my sales reps to increase customer satisfaction? Are my managers in these locations not performing well?
• Customer LTV: Since I now know how many, why, and where my customer might churn, another great question to answers is how much revenue the body shop is going to lose? Calculating Customer Lifetime Value (Regression model) is going to tell me how much each customer is worth to the organization and help me understand why I cannot let them go.

Marketing Analysis

Predicting marketing efforts is essential nowadays. In a world where businesses have unlimited needs but limited resources, the allocation of resources is critical for all companies.

• Product Demand: Predicting which product my customer is going to purchase (multi-class classification) can help organizations in multiple ways. Since you predict what X number of customers is going to buy, inventory optimization is the first aspect. How many products should I order from my supplier a year? How many units the body shop has to send to each location across the US monthly? What is my sales forecast going to look like for a determinate period of time?
• Cross-sell/up optimization is another key to effective marketing. You're 60-70% more likely to sell to an existing customer, compared to the 5-20% likelihood of selling to a new prospect. Predicting which products a customer is more likely to buy can help marketing teams. For example: If the body shop found out that X number of customers who purchase tires are also likely to buy cleaning products, now I can run email marketing campaigns with coupons/offers that encourage them to buy these products in a bundle.
• Marketing Attribution Report: Marketing attribution is a reporting strategy that allows us to see the impact that marketers made on a purchase or sale. Knowing where my clients are coming from, helps marketing departments allocate resources more effectively. In the case of the body shop, we can see that social media has the biggest portion of the marketing attribution, so I can make an informed decision to allocate more resources to social media (more ads, more content marketing in a specific channel, etc) and at the same time, I can analyze why other channels are not working, in this case why cold calling and blogging are not relevant in the purchase desition? Is my content not good enough? Should I stop spending resources on these channels since I know the low impact on my customers?

• Social Media Interactions: Finally, since the body shop understood that social media is the best channel, now we can predict how successful our efforts will be on this channel (Regression model). How many interactions are my ad's going to have? How many people are we going to reach on this channel? Is the ad is going to be successful? if no, it is worth it to do it?

Squark is a tool designed to fill the technical gap for average business users. Without statistical and coding barriers, we can democratize access to AI and Machine Learning for all types of companies and professionals. Now it is possible for companies to generate faster results without an army of data scientists!

Actions beat Insights!

Want to know more about Squark? Feel free to contact me, I'd be happy to provide you with a demo account to get you started.