Data scientists need to absorb the KISS principle: keep it simple, stupid.
That was the advice of panelists at this week’s Computing Big Data and IoT Summit 2017. Not that anyone would accuse data scientists of being stupid, of course. After all, they are generously rewarded precisely because of their brain power. The problem is, they know it, said Bob Tulloch, director European hospital products at the Decision Resources Group.
“It is very arrogant of data scientists to think that they are the ones who are going to produce the report that everyone wants, because they don’t. Let everybody get to the data they need through a very easy to use interface,” he said.
Tulloch’s words were echoed by Tom Dalglish, head of technical services, applied innovation at HSBC, who said some data scientists behave as if they are still doing a PhD.
“I read a report the other day and it was beautiful, with graphs and diagrams about why this particular machine learning model wouldn’t work, and everyone from the practical side said: ‘And now what? What do we do now?’ There was a conclusion but no recommendation about what we should try next.”
“You’ve got to keep it simple,” urged Tristran Isles, solution development manager at Honda. “Part of your job is to find things out and break it down into language other people can understand, and sell your findings. If you can’t do that, you need to work on those skills.”
Kaspar Gerling, data scientist at TransferWise urged his counterparts to focus on things that will gain business attention. “Show the profits or the losses,” he said. “That’s what the business wants to see”.
There is an obvious dilemma here. Most data scientists relish the challenge of unearthing ‘unknown unknowns’, which are likely to be in pretty niche areas, whereas those discoveries of most interest to the business are likely to be more prosaic.
Experimentation is all well and good, but it must be disciplined, said Dalglish.
“Deliver, deliver, deliver. Keep it simple, experiment, produce results, then move on.”
Gerling said there was a need for data scientists “who are more like data engineers and less like analysts”, those able to put the pieces together in a production environment, while Isles suggested data scientists need to get their hands dirty by acquainting themselves with real business problems “at the sharp end.”
Meanwhile, both Tulloch and Isles emphasised the need to develop self-explanatory front ends to foster a self-service model for information sharing of.
“If you’re a data scientist try to make yourself redundant. Try to democratise the data so anyone can use it, Tulloch said. “Stop being the guy in the white coat.”
Save this article