Access, model, deliver
Ever wondered how data science is done at a tech company with an already sophisticated set up? What’s the approach that you might want to think about in order to drive the best return on data? What areas do you focus on and how do you tackle them with your team?
I spoke to Alberto Rey Villaverde, the Chief Data Officer at Just Eat about the ROI of data, about how data science is a series of tools to enhance humans, and how agile needs the right conditions in order to truly deliver value.
Put simply, I’ve unpacked Alberto’s 3-step process to data science. As he puts it: “Focus on access, model and delivery.”
Have an effect and take people with you
“We’re all about creating great food moments for everybody.” Alberto says. “And my interest and my objective lies in how we make sure that data becomes a competitive advantage for Just Eat.”
The set up of data collection and analysis at Just Eat is already quite sophisticated – it is a tech company after all, and Alberto is the company’s second CDO. “A lot of groundwork has already been covered.” He adds. This enables him to build out the three key steps, which is “access to the data, the modelling of the data, and delivery.”
The challenges around this at Just Eat are mainly two-fold.
“For me it’s finding things that make a real effect.” Alberto explains.
“But critical for the success of any data function is how change management is being handled.”
“It is not just about a crazy black box AI that just does stuff. You need to bring people around.”
Tools that enhance humans
AI, machine learning, deep learning algorithms. What’s the key to cutting through the buzzwords?
“I try to move away from all this jargon and for me, when we're talking about analytics, we're just basically talking about converting a series of historical data into value for the company – for more revenue, efficiencies, cost savings, etc. It’s just tools that are there to enhance organisational capabilities and humans at the end of the day, a structure for handling rules that somebody's thought about.”
Of course, the value that data can bring to a company can be a bit abstract and therefore, can also easily be met with resistance.
“Some people even have fears that they don't even recognise.” Alberto says. “The fear is related to the fact that well, how do I know if things go south? How can I still keep control over it?
This is why I always talk about having humans in the loop and designing AI with human in the loop from the beginning. That reassures people about the fact that they always have full control and they just need to look at this as a tool.”
Finding the low hanging fruit in data
Showing what data can do as fast as possible is an effective technique to make the most risk-averse stakeholders come around.
From experience, Alberto has developed the skill of spotting quick wins with data:
“When I talk to people, normally within five to ten minutes I can spot two-three easy use cases. Something that you could see that could have an impact on the company, and would show them what data can do for them. So the first magic trick that these tools are doing is actually convincing the stakeholders that the data is useful.”
Spotting the low hanging fruit and working iteratively is often the quickest route towards creating value:
“What is the minimum viable product that you can create for your stakeholder that doesn't require a lot of mobilisation of resources on the tech side? Or even on the function that you're serving the product for?”
Dependencies and agile don’t go together
Agile can be a useful tool when it comes to data science, but Alberto points out that it needs the right conditions to thrive.
“What are the areas where we no longer need that much introduction? And how can we just set up the right environment for the data scientists, just to do pure analytics, move fast, iterate and test.”
How can CDOs isolate those use cases – assuming their organisation is mature enough to be able to do that – and reduce the amount of what Alberto calls “wasteful” interactions with other areas? He explains:
“Reduce the amount of dependencies, create the boundaries for which business stakeholders and tech stakeholders are comfortable with, and then kind of create that sandpit in which you can iterate and operate.”
As Alberto points out, this gets easier “when you have a product that has reached a point in which all the ongoing development is somehow self-contained” and you can keep involvement and communication to a minimum.
The ultimate deep learning algorithm
Beyond tangible KPIs, is there a deeper measure of success that Alberto hold himself accountable to at Just Eat?
“Yes. In simple terms, we're doing things as we were not doing them before – you know, is your life different than what it was before? How are we enhancing your life?”
Creating capabilities and enhancing humans through data doesn’t happen in isolation from other stakeholders. As Alberto puts it, he might have people on his team with PhDs, but there’s nothing quite like the knowledge of the collective brain:
“The ultimate deep learning algorithm is the one that each and every one of us has in their own head. All that thinking is embedded into all the professionals in the different functions that we have here at Just Eat. One of the critical aspects of any analyst that jumps into a new problem is how to learn all that know-how to start to deliver that impact.”
Is there a risk that one day humans will be replaced with AI?
Alberto believes that the two will naturally converge and that human interaction is an essential ingredient towards the best outcomes.
“We are here to deliver value to the company, and not in a black box scenario. We are here to deliver value, where there's humans in the loop always.”
“We are here to create capability that might look like magic, but it's capability to enhance humans at the end of the day.”
This post was originally published on the Made By Many Blog in September 2019. Alberto Rey Villaverde is the Chief Data Officer at Just Eat.