John Akred: Data scientist in search of the right question
“It can look as though my life led up to this moment in what seems like a grand plan to become a data scientist. The truth is I just got lucky that the world decided to call what I do data science.” So says John Akred, co-founder and CTO at Silicon Valley Data Science (SVDS), a data science consultancy he started four years ago and which pulls together the many strands in his career to date.
Akred, who is a keynote speaker at DataFest on 23rd March, took a degree in Economics because, “it seemed to explain most about the way the world worked.” From there, he became interested in international development issues and how data could be used to make an impact. An internship at the US Agency for National Development, part of the State Department, involved him in considering whether aid payments were being diverted into buying defence technologies like tanks. “It got interested in answering questions with data very early on,” he recalled in an interview with DataIQ recently.
A switch to capital markets in Chicago found Akred developing algorithms for banking regulation. “I could spend lots of time looking at massive volumes of data from social networks on top of individual trading streams which was then very hard to do, but was a good way to start my career because I was developing models, building regressions to predict the optimal outcome and identify outliers,” said Akred.
Spotting insider trading or fraud may have been a fascinating technical challenge, but Akred discovered it was no way to build a career. “People appreciate innovative ways to make money more than they do finding crooks,” he noted. So he switched to creating a volatility index using mergers and acquisitions data. “If data science had come along at that time, that’s what I would have called myself - and asked for a raise!”
Having decided to take a Masters in Computer Science, specialisting in distributed systems, he moved into the technology industry as a sales engineer and solutions consultant for SPSS. He explained: “I was helping customers to use the software to do advanced analytics and data mining, talking to people doing things with neural networks. Half of our use cases were for marketing, like who to send emails to, but the other half were around fraud detection and process optimisation.” Some more pieces of the puzzle that would resolve itself as data science were falling into place, not least because of the rapid expansion which data and analytics was undergoing at that time. “I wouldn’t have got that insight from any other perspective,” Akred said.
His next move was into a R&D role at Accenture and a different culture that was not as single-minded about data. “I had to figure out how to find hard problems to go after and build up a data science and predictive analytics practice,” he recalled. Fortunately, his cohort at the time included Rayid Ghani, who went on to serve as chief scientist on Barack Obama’s second presidential election campaign, and Andy Fano, who created the AI lab at Accenture. Seven years at the management consultancy gave him a broad client base and set of problems around which to innovate.
Then four years ago he decided that driving projects for $30 billion-plus organisations was restricting and that there was a group of businesses who could also benefit from these skills and techniques. “Organisations who are looking to grow and want to innovate, but who don’t have data engineers, architects and scientists, want help to solve theirproblems. That is a very different type of project,” he said.
Akred’s vision for his start-up was managing data science projects and teams to solve business problems and drive growth in the tech development sector. “It is an important issue that people don’t talk about. Most data science teams are in-house and work for one company - they don’t see the value in explaining how they work. Our clients like the fact that they learn from us when we work with them,” he explained.
A key dimension to what SVDS does is to overcome some of the human issues which can be a barrier to getting value from data science. An essentially introverted, expert community is being managed by an extravert, generalist executive.”That is the hard part. There are a lot of unhappy data scientists because the people who manage them don’t understand how to do it.”
One way this clash can express itself is in the way projects progress. Typically, a consultancy programme will deliver in steady increments of 20, 40, 60, 80 per cent until the project is complete. Data science can often stick at the 10 per cent mark before making a big leap to 90 per cent. “There is a lot of banging your head against the wall,” explaied Akred. “So it is about managing expectations and managing stakeholders to realise that is how it works. It is a huge mindset shift to do that.”
His years as in sales and management consultancy built his own skills as a relationship manager. “I realised that expectations in the organisation have a huge impact on the success of a data science team. Out of that realisation came this business.”
The rush to hire data scientists has led to a lot of disillusionment about their impact. Organisations get excited about their new recruit, but three months later have yet to see any of the magical data-driven innovations they were expecting. Data scientists can also be disruptive of traditional data management and IT because of their requests for ever more data, which may not be available within a timescale or under the right data governance to be productive. “Then there is the problem of finding valuable questions to go after. Understanding what those are and how they can be implemented in a business is not something that is taught in maths school,” said Akred.
Akred knows about this issue from personal experience. While at Accenture working on a project for USPS, he developed an alogorithm to optimise machine sorting of products that could have led to hundreds of millions of dollars in savings across the supply chain. Unfortunately, that organisation did not have the capability to change its sortation process more than once a month.
SVDS has since developed a methodology based on agile techniques which translates business problems into technical questions while building into a project an understanding among stakeholders about how they will handle the answers when they arrive. The result should avoid those culture clashes which have delayed the return on value for many. As Akred pointed out: “If you want to upset a data scientist, tell them you want to do something for engineering.”