Meet Rosina Norton: Angaza Data Scientist

Angaza’s mission is to create the technology that allows businesses to offer life-changing products, to anyone, anywhere. In service of that mission, we prioritize our customers’ trust by ensuring the security and confidentiality of customer data. The Angaza Data Science team is an integral part of ensuring that data improves our customers’ operational performance and adds maximum value to their Angaza experience.

Learn more about Angaza’s data initiatives from Angaza’s lead Data Scientist: Rosina Norton. In this Q&A, she discusses her interdisciplinary role at Angaza, the data science methodologies that guide her day-to-day work, and what’s in store for future data science initiatives at Angaza.  


Hello, Rosina! Tell us about yourself.

[Rosina Norton] I’m Rosina Norton, Angaza’s Data Scientist. I have a Master’s in physics and started my professional life as a wind energy analyst. Simultaneously, I was involved in several energy access initiatives as a volunteer, and then decided to merge that with my professional life and ended up moving to Kenya to work with a solar consultancy. Since then, I’ve been working in technical modeling roles in East Africa, varying from modeling energy consumption in off-grid health centers to looking at commercial and industrial facilities. And that’s how I came to join Angaza as a data scientist in the San Francisco office.

 

What does it mean to be a data scientist at Angaza?

[Rosina Norton] Angaza is really interesting in that we have such a rich and unique data set, which covers product distributors, their clients, and clients’ repayment histories. And as a data scientist my job is to work with that data and think about how we can leverage it to provide additional value to our customers. Once we’ve defined a specific customer need, I’m responsible for building the model that lies at the heart of addressing that need. I dig deep into the data, look at specific trends and patterns, decide which algorithms are relevant for that particular problem, and then build out the model pipeline such that we can go from raw data all the way through to prediction, and provide the results to our customers.

 

As Angaza’s lead data scientist, what are some methodologies from the data science field that guide your day-to-day work in the context of Angaza’s customers?

[Rosina Norton] I think that one of the most unique challenges that I’ve come across compared to my previous work is the diversity of our customers. Working across 30 different countries, we have an extremely diverse set of customers who, individually, all have their own unique operations and different products they’re selling. This means that when we’re building data features to serve a wide customer base we need to be continuously taking that into consideration. Our models have to be extremely flexible in their data input, because some customers might collect some kind of data, and others might not. When evaluating our expected performance, we are actually doing that on a customer-by-customer basis. Consequently, we have a pretty high bar for performance. We want to make sure that our products are working well for as many of our customers as possible and that we’re really confident it’s bringing them extra value.

 

What are some of the initiatives that you’re working on now?

[Rosina Norton] I’m working on a new initiative based around GSM connected solar home systems. Over the last year we have more and more GSM units in the field, which are collecting detailed sets of data about system operation: solar panel power and battery voltage, for example. This is really interesting physical data and our aim is to use it to help our customers provide a better service to their clients. Part of the challenge is simply to identify the best ways to provide value to our customers, given the number of available directions.

 

One of our most recent data feature releases on the platform is the Repayment Risk feature, indicating a client’s likelihood of non payment. Can you share a little bit more about the process that you took to identify this as a customer need? What were some  approaches you took to creating the feature?

[Rosina Norton] That feature was identified as one of the most immediate ways we could have a positive impact on our customer’s operations. Providing them with a specific risk metric at an account level helps them make targeted decisions that improve their repayment rates. From the data science point-of-view, the first step was what we call an exploratory data analysis, looking at characteristics or behavior associated with an account that potentially impact repayment. Once we have a handful of those characteristics, which we are sure have some kind of predictive power, I start building a model. Each step I’m adding more data and optimizing the model by maximizing the expected predictive performance. So, it’s a very iterative process. Once we’re happy at the level of performance, the model is implemented and deployed to provide predictions which are then fed to our customers via the Hub.

I think it’s worth noting that over time a customer’s behavior is going to change, their client base is going to change, so this isn’t a static process. We are operationally monitoring the performance over time. At any point if there is a regression, or there is potential for an improvement, we continue updating and improving the model—especially as we get feedback from the Hub and our customers.

 

What insights have you gained about the unique needs of  last-mile distributors or their clients, since you joined Angaza?

[Rosina Norton] I’ve been really impressed with the creativity in the field. Our customers come up with new ideas of how to sell products and how to make these products more accessible to their client base. For example, Angaza recently built-out the Remetering capability, which is the ability to relock an unlocked metered device and to use that to sell and enforce an unmetered product. Seeing all the clever ways in which our customers are trying to make these products more accessible to their clients is really, really impressive. The other thing is that, while East Africa is well known for its high uptake of pay-as-you-go products, it’s super exciting to see other distributors popping up all over the world and this kind of technology taking off in other areas.

 

What are some trends or challenges that you see in your field?

[Rosina Norton] One trend which we see in customer-facing data science is this desire for more tailored and intelligent insights built into the platform, without requiring a customer to run their own analyses using external tools. This seems to be a trend in the sales and energy industries as well. Angaza is investing resources and expertise into providing these capabilities for our customers. That’s definitely one trend that we’re spearheading in the pay-as-you-go industry.

A technical challenge is the growing size of these data sets. Angaza customers have grown incredibly over the last couple of years. As that trend of growth continues, we’ll be moving from more traditional data science techniques, executed on a single machine, towards more distributed architectures.

The other challenge and trend in the whole data science field is a focus on “ethical artificial intelligence.” There is genuine concern about what power we’re putting into the hands of our algorithms. I think at Angaza that’s of particular significance. Our business customers work with an underserved demographic, and we need to make sure that data is being used not only in an ethical way, but such that it’s a benefit to individual consumers.

At Angaza we put together a Data Constitution, a binding document for all of us who work with data, to ensure that we avoid breaking any of our values or our mission, and that the way that we use data is in line with our morals.

 

What is next for Angaza’s data science initiatives?

[Rosina Norton] That’s a really good question. It’s also a tricky one because our initiatives are definitely influenced by the evolving needs and requests of our customers. We’re constantly listening to their feedback and using that to decide how best to focus our resources. One particular area that we are interested in, driven by customer conversations, is pre-sales risk management. These tools will help our distributors manage their portfolio risk, which in turn helps them provide additional financial services and improved access to life-changing products for their clients.

 

What do you enjoy most about being a data scientist at Angaza?

[Rosina Norton] One particularly great aspect about being a data scientist, especially at Angaza, is the fact that you get to touch so many parts of the process. It’s partly the nature of being in a smaller company, but it’s also the fact that Angaza’s management really encourages transparency and interdisciplinary collaboration between different teams. For example, a standard data science approach in some other companies would have the data scientist strictly sitting in between the data coming in and the model going out to production. At Angaza you are part of that whole process, as well as the process of working with the customers to figure out what they want on a more strategic level. For someone who loves learning, and loves knowing exactly how all the different parts of the puzzle fit together, that’s a really satisfying part of working at Angaza.