When we pitch our AI solutions to businesses, they are met with mixed feelings. At the executive level, adoption looks like a no-brainer. Here, the focus is on the bottom-line, and ecosystem.Ai looks like the most attractive option. But when news trickles down to data scientists, we are met with resistance due to existing preconceptions about AI solutions.
This tension was front and center in our recent ecosystem.Ai meet-up. Our partner, Matthew Hindley, gathered data scientists from several different companies to determine what they really feel about AI Solutions.
The drag-and-drop “black box”
Data scientists often get cast as the “people who say no” to AI solutions, explained Georgina Armstrong, a data scientist and executive who has worked at Shoprite, Woolworths, and now at personalization app Zedge. Armstrong said that, while data scientists’ concerns should be seriously considered, they are instead perceived as stubborn obstacles in AI transitions.
Armstrong recalled when AWS SageMaker first hit the market: “They told us, oh we’re just going to use SageMaker now – you just drag and drop the model you want and it does the data science. That is probably the root of a lot of my deep skepticism… because of the horrors that arose from it.”
Drag-and-drop tools quickly became black boxes, said Armstrong. Logic became impenetrable in AI generated models, while the accountability for mistakes still fell on the data scientist. “The more layers of abstraction between your data scientists and how the model works, the harder it is to answer the questions when things go wrong,” Armstrong said.
The hard truth is that whenever AI tools aim to replace data scientists, businesses expose themselves to errors that can’t be traced, waste resources retraining people on short-lived platforms, and erode trust with the very professionals they rely on. Ashwinee Pandey, a data scientist from X-idian, the adoption of a new platform is always followed by a lengthy warm-up period where they are trained to use the product. “Then your skillset becomes limited to that niche scope,” he added.
Additionally, Armstrong claimed that vendors oversell their technology’s capabilities. “It’s a constant feelings of vendors always knocking on your boss’ door, telling them how easy it will be,” she said. “It’s exhausting always debunking things to your boss, always saying ‘it’s complicated’ and vendors saying ‘it’s magical’.”
Giving data scientists the freedom to choose
Every data scientist has their own threshold for what they want AI to automate. Armstrong loses trust when a tool overrides her ability to build. Tshiamo Shilowa, who led the development of Unilever’s data and AI strategy for Africa, prefers a middle ground: automation of repetitive tasks, combined with customizable tools for the harder, more rewarding problems.
Rather than replacing data scientists, the Prediction Platform gives them room to “play around”. This is according to Safaricom’s Nyadoi Odhiambo, who described the transformation after adopting the platform. Previously, campaigns were driven by batch analytics and simple A/B tests – manual, heuristic, and limited in scope. With the Prediction Platform, Safaricom now runs real-time, personalized recommendations across 50 million customers and more than six million daily interactions.
Odhiambo highlighted that the platform’s flexibility was a game-changer: “It wasn’t just locking us into what was already within the system, but giving us the ability to build our own external models, test them inside the platform, and improve them over time.”
The ecosystem.Ai Prediction Platform provides a low-code environment for data scientists to build, test and deploy models with ease. Additionally, the platform’s worker architecture enables data scientists to swap out technologies, keeping the tools they find useful.
When asked what the ideal AI transition would look like for data scientists, Armstrong said that platforms should act as “invisible utilities”, automating the mundane while empowering teams to focus on creativity. The lesson from our discussion with data scientists is clear: AI that seeks to replace data scientists breeds resistance and fragility. AI that augments their expertise, however, builds trust and long-term value. The real breakthrough comes when AI platforms become like electricity – an invisible utility powering creativity and making room for future innovation.
