The statistical project was developed and lives in Snowflake

Objective: Provide a way to identifying more customers that contact us that my have not been identified by the IVR. Additionally correct situations where the IVR might have been wrong.

Various sources were used in the project, all of which were housed in the data warehouse. The data included customer calls and chat data, payments, work orders, salesforce case data etc.

The data was in the data warehouse Snowflake; this is where the logic was written. It was written in SQL. Then a Power BI dashboard was created for analysis. DAX is used for the Power BI Dashboard

Methods and Techniques: I employed various data analysis techniques, Multi-Criteria Decision Analysis (MCDA) or Multi-Criteria Decision-making (MCDM) is at the heart of how the best guess is derived. The method used is the Analytic Hierarchy Process (AHP).

Key Deliverables: The project delivered a model that is stored and maintained in a Data Warehouse, Snowflake. The Model allows users to search with an interaction ID for a call or chat and it will provide the best possible match of the account number tied to the call. Additionally, a dashboard was developed to facilitate ongoing monitoring and analysis.

Account Matching monitoring Power BI Dashboard

Project Impact: The implementation of this model improved our ability to identify customers from an average of 59% to 85% of all interactions. This allows for the implementation of a first call resolution model that could notify the business of frequent callers and the agents taking those calls or causing them. Additionally, this model was able to enhance the QA and Coaching processes as it was now easier for leaders to get specific action-based interactions.

Client: Shaw Communications Inc.

As the lead Data Analyst, my responsibilities included planning the project and personnel involved in the project. I formulated the AHP method and wrote the MCDA/MCDM logic in the Data Warehouse. I worked closely with source data teams to ensure data accuracy. At the end of the development cycle, I validated the model and built the report to monitor performance.

”Being able to clearly define the objective is the first problem, solve that well and the rest unfolds.”

How the core logic operates.

This page has the daily match rate along the number of actions identified each day. It is clear to see that on days when there are more action the match rate tends to be higher.