GitHub Link: https://github.com/AndreasWatts/Sales-and-Marketing-Analytics-Implementation-Verisure

Verisure, a global alarm and security company founded in 1988, specializes in smart alarm systems to protect the homes of their customers with a 24/7 alarm central ready to call the police or security guards, when their alarms get triggered. I joined the danish organisation to implement business intelligence for marketing and sales, and later I was put in charge of implementing my solution for Verisure Denmark across the nordic countries.

Verisure need to implement an automated reporting solution in Power BI, with data models that track the marketing performance end-to-end from click to sale. To begin, the company only had excel sheets that required daily manual data extraction from an SQL server, and then loaded into the excel file, and they could only track from lead to booking.

After implementation of the data models and Power BI, the company saved hours on manual tasks every week and they could make better decisions based on the end to end data model.

A simplified overview of the data architecture and tech stack is provided below:

The tech stack: Azure: Databricks, SQL Server, Data Factory, DevOps Data Engineering: DBT Fivetran Business Intellgence: Power BI, Tabular Editor Programming Languages: Python, Spark, SQL

Below is a small sample of some of the tables and datamodelling done for the project. The final datamodel consisted of a handful of fact tables containing information about website funnel visits, leads, bookings and sales, as well as multiple dimension tables containing demographic information, orders, product bundle etc.

Sample images of some of the dashboards & Reports:

Before implementation a lot of the decision making was based on the KPI Cost Per Booking due to not being able to track all the way down to individual sales. After the implementation the decision making for marketing improved and the overall cost per acquisition decreased. The company found cases of campaigns that performed well in terms of cost per booking but poorly for cost per acquisition and vice versa.

Certain marketing channels turned out to perform better than first anticipated based on the new data. For example, one channel performed poorly in terms of acquisition but provided great customer lifetime value.

A machine learning project utilising marketing mix modelling found that the performance could be improved without increasing the marketing budget. See below.

The company thought they were using too much money on Facebook for marketing. However, a marketing mix modelling project revealed, that this assumption was wrong, despite facebook being the largest marketing channel. In fact, the performance of the company could be improved by adding more money to facebook advertisement and decreasing the spend on other channels.