TD Bank | Data Visualization Lab

I spent two years leading the data visualization efforts for an advanced analytics team in the marketing department at TD Bank. Our department consisted of 15 data scientists, business analysts, software engineers and me. While my colleagues built machine learning models to help the bank increase sales based on patterns in customer data, I designed and built data visualization tools, Tableau dashboards and short videos to help explain the output of those models.



Awards

"Grand Champion", 2017 Think Digital Tech Jam
"Best In Show", 2017 Marketing Analytics Solutions Fair
"Team of Distinction", 2017 TD's Got Talent Award



Category Corporate & Data Visualization

Date November 2016 to February 2019




Modelpedia

Our team was responsible for all the machine learning models used to determine marketing campaign lists, next best action recommendation engines, and various other use cases. Before I joined, all information about these models was stored in an Excel document. We were constantly fielding questions from stakeholders about model performance and other details. We decided to build an internal web app called Modelpedia to democratize this information.





Marketing Message Audience Estimator

This is a simple tool for marketing planners to estimate the number of customers in each demographic segment that their message will resonate with. It uses data from various sources to calculate the potential "bang for your buck" and visualize the results.





Income Volatility

With access to the income and spending data of nearly one third of all Canadians, we had the opportunity to explore the financial profiles of many different segments of the population. For this income volatility study, I designed and programmed a small web app that visualizes the flow of funds for people with especially volatile incomes using a custom Sankey diagram.





Canada Cartogram

TD has more than a thousand branches across Canada. Most of them are clustered within cities that are spread apart from each other, geographically speaking. This makes it difficult to create an overview map to compare branch performance across the whole country in a single view. The branches that are close to each other overlap and become indistinguishable. And there is a lot of wasted screen real estate in the lower populated regions. We developed a cartogram algorithm to redraw the branch map that preserves the geographical relationships while still making each branch visible. Then we built a Tableau app using this new layout to explore which branches are most effective at following up on sales leads. With this new layout, it is much easier to identify the high performers and compare them to their neighboring branches.

This is a mercator projection of all TD branches, with nodes scaled to indicate sales performance for a particular campaign.


This is an algorithmically-built cartogram of the same dataset. Note how there are no overlapping nodes even though overall spatial relationships are maintained.


This Tableau dashboard uses the cartogram layout for visualizing sales performance across the branch network.




Customer Service Chat Analtyics

A classic word tree using customer service chat transcripts. Here are all the conversations in our corpus that begin with the word, "Hello".




Financial Confidence Survey Results

Ipsos Reid did a survey asking Canadians what factors contribute to their sense of financial confidence. My team developed this web app to visualize the results.




Floor map redesign

This is kind of silly, but the floor map on the wall near the elevator was an information design nightmare. My version reduced the amount of time it took to find a meeting room by 75%.