Just Eat's Alberto Rey Villaverde on machine learning, AI, and bridging the imagination gap (pt. 1)

Key Takeaways:

• AI allows the data to speak for itself. Alberto's application of simple neural networking algorithms to massive sets of data efficiently and consistently optimized price points saved his company money, justifying an expanded budget for AI adoption and development.
• AI + humans is better than AI alone. AI augments human abilities, but it doesn't replace them. In unprecedented times (as we're experiencing on multiple fronts in 2020), algorithms especially rely on human imagination to fill in the blanks where there's no historical data to inform an outcome.
• Three components crucial to the success of your data products and services. How Alberto sees access, model, and delivery fitting together to justify the cost of innovation necessary to implement new technologies and methodologies for handling data.

Key Quotes

I remember in 2008 I read The Singularity Is Near from Ray Kurzweil … it was talking about what will happen in 2010, 2020. … He made quite bold predictions in there. He was talking as well about the data revolution, about robotics, and things like that. I remember the first explanation he made in that book about how a neural network worked, and I took that thing and I built one — with [help from] papers on the Internet — in an Excel spreadsheet with all the different cycles per line. … I was fascinated by the whole thing, how that was the key for machine learning — was the key to [letting] data speak by itself.

How will data help organizations navigate the current economic crisis better and faster?

Some people are asking: ‘How can we make better forecasts right in the middle of this uncertainty?’ I think that this is a Black Swan event and it’s going to be very difficult if not impossible to … predict this thing. … Data cannot make magic, and I think this is where the human touch comes in place. I’m an advocate of Demis Hassabis’ take on this: human plus AI is much better than AI alone.

Any data product has three components you need to get right. One is the access. Are the problems you’re trying to solve too big for the tools you have today? Use the correct infrastructure. [Two is] the model: your data scientists, your analysts, your experts with the right level of skills, talent, and passion that [work with] the data long enough so they can tell you exactly what is [happening] in there. The third bit — which is, in my opinion, the one where most people fail — is the last mile. The delivery. Sometimes we forget that the product we build needs to live in the organization we are in, and we need to be adaptable to that.

Data lakes versus data warehouses?

Data lakes are definitely really fun because they really help you to not have data silos anymore. However, data needs to have structure, right? And you still need your data models that are reliable and easy to consume. I honestly have faith in all data warehouse technology and approach, so I think there is space for both, to be honest.

The future of data:

It’s hard to say how everything will evolve. … Put yourself in 50 years time, data is going to be so pervasive and we’re going to be talking with our own AIs. But data will be so pervasive that everybody will master the domain. Between now and then there’s a big gap that we don’t know what it will happen, but I think that in the next five to 10 years what we will see is the emergence of a data function that functions on its own, that offers horizontal services: ‘I can do neural networks for you,’ for example, ‘I can provide forecasting services,’ and then a set of business partnering and analytics expertise to those functions.

[I advocate a hybrid model of data organization] because the word ‘centralized’ is very dangerous. People feel that they are going to be losing ownership of [what’s being done with their] data. I guess at the end of the day we’re building the Excel spreadsheet of the 21st century … they kind of do the job and everybody understands them.

On merging with another company:

Integration is not about so much about technology in my opinion; it doesn’t really matter which platform choices you make. More than anything, it’s about what really powers that technology, which is the skills and the talent that you have sitting around and how you make sure that all the things you did well pass into the new organization and the things you didn’t do well, you are able to understand that you didn’t do them well and you’re able to learn from the other [company].

The importance of scheduling regular hackathons and innovation days over the course of a year to generate good ideas and boost morale:

You need to have the right culture. For our tech and data teams to have those innovation days is like bread and butter. You really need them because you see a lot of super interesting ideas — ideas that end up being implemented and really help us to bring the culture that we want to promote.

Bio

Alberto Rey Villaverde is the Chief Data Officer at British online food order and delivery service Just Eat. He is a pricing and revenue management professional with extensive experience within the data science field, particularly on BI and advanced analytics, data mining, machine learning techniques, and scenario modeling.

Alberto started his career in advanced analytics as a member of the pricing and revenue management team at easyJet, working in the development of one of the most advanced pricing engines within the industry, where his team pioneered the implementation of machine learning techniques to drive pricing. He holds an MSc in data mining and an MBA from Cranfield University.