Streetbees believed machine learning could unlock the value of the vast amounts of market research it was doing for clients. But the team didn’t have sector expertise to build the capability. Specifically, they lacked the domain knowledge to get the right team in place and establish an effective research and development (R&D) strategy.
At the beginning of the process, for example, it hired someone who had experience working on Netflix’s recommendation algorithms. However, it turned out that wasn’t a good fit for the large sets of open text data in need of analysis.
Co-founder Oli May said progress only really started after a funding round was closed March 2018, when key hires could be made.
The management team reached out to industry specialists Faculty.AI. Introductions to candidates and specialists helped get the team up and running. The wider company structure “changed considerably” after the data science team was hired and the head of data science is now part of the leadership team.
The team holds weekly sharing sessions where different departments talk about what they’re working on. Data science presentations are both the most interesting and challenging, according to May. It puts pressure on the data team to explain their work and the impact it has in an accessible way.
Each team member has a £500 annual training budget that they are encouraged to use, as well as an ongoing Audible subscription.
Since implementing the new strategy, the data science team have become an integral part of the company. Streetbees has transitioned from using existing natural language processing techniques to developing “groundbreaking” in-house capabilities, such as extracting meaning from unstructured text. The four year-old business now has 75 employees between its London headquarters and an office in Lisbon, and is working with corporates like Unilever, PepsiCo, BBC World Service and Vodafone.
“Do your research into what you really need before you start trying to build it and don’t be afraid of seeking expert advice. That was our initial point of failure, and it’s what we now know you need to do,” said May.