A Customer 360 Solution for Insights and Predictions

Smartbridge is a Customer 360 Platform Certified Salesforce Partner

Objectives

Key Challenges

First, Artificial Intelligence and Machine Learning (AI/ML) capabilities need to be added to the data flow. Typical data pipelines and database tools can perform data transformations, but do not have AI capabilities.

Second, a well-performing machine learning model needs to be generated in a short amount of time. In a traditional machine learning model training approach, one must carry out multiple rounds of data pre-processing, hyperparameter tuning, and optimization in order to identify the best-performing model. This requires time and compute resources which we did not have in a hackathon setting.

Third, is the maintainability of the machine learning model. After the ML model has been trained and deployed, it needs to be ran regularly to predict customer behaviors. As more recent data is collected, the model also needs to be retrained on a regular basis to accommodate changes. Thus, a fully automated machine learning model training and scoring system is needed.

Solution/Smartbridge Methodology

Smartbridge started with a relevant business opportunity for our real estate clients which would provide a simple, scalable solution for them to implement. The talented team of data engineers, architects, scientists, and analysts built an end-to-end solution that involved some critical steps. By using Azure Synapse Analytics, Azure Data Lake Storage, Azure ML, and Power BI services, all within the Microsoft Azure ecosystem, we were able to integrate data from multiple sources (such as Salesforce), data transformation, AI/ML predictive analysis, and end-user visualization and interaction under one umbrella.

The following architecture was used to build the solution.

After identifying features that can be useful for predicting a customer’s likelihood to renew (demographic information, work order history, survey responses, renew history, customer lifetime, etc), automatic data pipelines in SQL were set up in Azure Synapse which perform data crunching then sends the training/predicting-ready data to the Azure Data Lake to be read and used by Azure ML.

Azure Automated Machine Learning (AutoML) is used to quickly generate a high-performing, predictive machine learning model. AutoML automatically iterates through multiple pre-processed techniques and model-training algorithms in parallel, while supporting the computing need with cloud computing. It saves time and resources by automating algorithm selection and hyperparameter tuning. To create an AutoML job, one just needs to set up a few configurations. Then, an optimal machine learning model with explanations will be returned within a couple of hours of time.

AutoML-training regression model (SDK v1) – Azure Machine Learning | Microsoft Docs

The pipeline steps in Azure ML allow data scientists to add additional data processing steps before training/scoring, which was not implemented in Azure Synapse. The ML pipeline can also be orchestrated to run on a regular basis to keep the model and the prediction results regularly updated.

Results of ML Driven Tenant Lease Prediction

The trained ML model provides predictive intelligence to help solve this business challenge, including evaluating what-if scenarios. For example, when changing certain tenants’ satisfaction levels from a lower value to a higher value, the model can predict the tenants’ inclination to renew after satisfaction improvement. This what-if analysis enabled through ML will help stakeholders identify revenue opportunities and focus efforts on the most promising parts. An example outcome of the what-if analysis is shown below:

The actionable insights are visualized using Power BI. As seen below, the tenants with a high likelihood of renewal are highlighted automatically and this can be used by the administrative staff for renewals and prioritizing them as necessary.

Extending the Solution

This project shows the possibilities of using similar approaches in identifying the customers with revenue-generating potential and prioritizing those tasks based on the data present across various enterprise systems.

What are some business use-cases this solution can be implemented for?

How can this solution be used in the future?

This solution can be packaged and rolled to any service-oriented clients working with Smartbridge as a quick way to get business value from their customer data.

Are there any enhancements that could be made post-hackathon?

More capabilities can be added to the existing projects, including integrations with other external data sources (e.g. Zillow, Social Networks) and unstructured data sources in this customer 360 view.

This solution and approach can be used by any business trying to develop actionable insights from their customer data. For instance, any service-oriented business (where the current customer has a choice to switch their service provider) can prioritize the customers that need attention based on the likelihood of renewal. This applies to customers in utilities, streaming video, food delivery, restaurants, e-commerce, and many more.