Data Science Intern at Explore Data Science Academy (2019 – 2019)
This was a 1 year full time course with sponsorship from a corporate company. The course involved learning Python, SQL, as well as working with GitHub, PowerBI as well as AWS. We were afforded the opportunity to work on corporate research projects involving sentiment analysis, voice analytics, topic modelling and a range of projects that utilized regression models.
Data Scientist Trainee at African Bank (2020-01 – 2021-01)
I've worked on a number of projects across the business mainly utilizing SAS and SQL.
- Analytics to help de-risk information from credit bureaus.
- A separate scoring model was developed for a cohort of users that utilized less features of the old model but included more information rich features. This was launched and ran alongside the original scoring model to AB test the outcomes of the new model. After 6 months the cohorts were compared and indicated an improved risk profile.
- Identifying areas of opportunity within the business where data science could have an impact
- A project to increase the high-value customer base by leveraging data science and machine learning techniques.
- A model was developed to predict a customer's lifetime value, and using this to offer an incentive to high valued customers to switch their bank. This was implemented initially in a pilot phase where users were tested against a control group to determine their likelihood of joining and to test the effectiveness of the lifetime value prediction. The result after 8 months when comparing the control group with the active group showed that offering the incentive increased the likelihood of joining and subsequently the increased revenue from the customer. A separate test was done to determine the effectiveness of the lifetime value. With all users in the test being offered the incentive, those that were predicted to bring in higher revenue, did so consistently and within 5% of the prediction.
Senior Manager: Data Science at African Bank (2021-01 – 2022-08)
My duties include working closely with stakeholders in order to derive insights from data and then implement projects and processes that take advantage of those insights. Duties also included project managing initiatives to increase high-value customer base, developing and maintaining machine learning models, monitoring and reporting of key metrics as well as ad-hoc support in the operations space.
- Leading a team to implement OCR to scan bank statements to identify debit orders
- Optimising debit order switching process
- Initial analytics and testing around implementing a rewards program. This included analysis that looked at the causal influence of having a customer main-banked and how that affects their uptake of credit and engagement with the bank.
- Another study was done on which mechanism would work best in convincing users to switch banks. This was done across channels such as calls, emails and in app prompts.
- Controlling for factors such as credit profile, income and usage, we tested prompting the user to switch vs no prompts. We also tested for those that received prompting, different channels. The end result proved that the prompting provided a significant lift in uptake and that depending on the users profile, specific channels were the best communication channel.
Data Scientist (Operations Intelligence) at Discovery Health (2022-08 – 2023-11)
Duties are to support operations by making use of analytics as well as developing machine learning models to improve on processes, reduce service load and drive efficiency. The role also provides exposure to risk intelligence within the insurance industry. Some of the tools used are Python, PySpark, Hive and Cloudera.
- Monitoring and maintaining a model that predicts whether a claim will be reworked including productionalizng and implementing a model update.
- Setting up an A/B testing framework for a new app. This involved setting up key monitoring for factors that we were testing such as engagement rate, time spent on certain screens as well as monitoring of indirect affects such as incoming queries to the call center and an increase in claims reworks, with the main aim of the app to provide additional self help features. Different aspects were tested such as positioning and timing of messaging to the user. Usage of features and the impact of removing features.
- A major area of testing as well was on the technical side of things where time to load and time for development of additional features were monitored.
- Working closely with app development team to ensure that correct data tracking is in place with the goal of having a recommender system built and implemented
- Analytics around service load in the operations space
- Project involving causal inference modelling to predict heterogeneous treatment effects. Here a model was built using a clinical treatment process where the model simulated the affects of certain 'treatments' or in this case activities without the effect of other activities and determining the direct impact on their health outcomes measured in hospital costs. This put into context allowed us to determine the next best action for a user in terms of their health care program. This was implemented such that 20% of users were given a Next best action where the rest of the users were allowed to determine their own next action. After implementation we saw an improvement of 30% lower hospital costs for those exposed to the next best action. This was done on users that were normalized across their health profiles and demographic data.
- As part of this an additional tool was created to match users based off their demographic data, health records and financial information. This was then used to determine cohorts for general A/B testing across the board.
Data Scientist at HearX Group (2023-11 – Present)
The role utilizes tools such as Python, R, SQL, AWS, the use of G-Suite as well as Tableau.
- Developing predictive models for preventative interventions into customer support. This involved creating a model to predict likelihood of return. The results of which were implemented by splitting those with a high likelihood into a control group and a test group normalizing the two across the distribution of their data allowing for comparative splits. The interventions were then tested and ran for 6 months with the result showing a absolute decrease in return rate of 8%. This was also compared against those with a low likelihood of returning and it also showed a decrease in return rate of 3%.
- Development of a Next Best Action model to assist customers and drive desired behaviour. This involved determining the causal effects of different nudges and the impact on behaviour. This causal effect was done by analysis and careful testing across users to ensure that when we nudged users that the only differentiating factor would be the nudge. The resulting behaviours were determined and used to build a platform that nudged users to behaviours that led to increased satisfaction and engagement.
- Natural Language Processing to gain insights into customer experience
- Developing machine learning models and analytics for marketing purposes. This model was based off historical data to determine the effect of marketing spend on revenue. The data had to be structured in a way that allowed the model to determine the indirect effects and direct effects from spending on a specific channel. The data was engineered in a way to ensure isolated spends were more apparent thus allowing the model to better determine the effect of spend on revenue.
- Development of a chat bot
- Overall analytics into several processes across the business
- Project managing any initiatives started from the above responsibilities including, implementation, monitoring and reporting
