Carlos Escapa, AWS, on Applying Machine Learning To Business Problems

4097

By Allison Proffitt, Editorial Director

It almost sounds like Carlos Escapa is in agriculture. He talks enthusiastically of low hanging fruit, and much higher-growing fruit, and how the coming harvest will change everything. He even waxes poetic about a new tractor.

He’s not, of course, a farmer (though he tells me that his family has been in the past). Carlos Escapa is the global lead for the AI/ML consulting practice at Amazon Web Services. The harvest he’s anticipating are the data coming from the sensors now embedded in everything from toothbrushes to toasters to traffic lights.

Carlos Escapa, Global AL/ML Practice Leader, AWS

“All of these data can be harvested to solve new kinds of problems. The number of connected devices, the internet of things or IOT, is just exploding,” Escapa tells AI Trends. “All manner of tools that humans use on a daily basis are going to be producing data about themselves, about their performance, about their need to be maintained and so on. And these data can be then used by companies to either improve their products or provide new solutions and create new business opportunities.”

The next step is best using the data we are generating. Escapa advocates for machine learning. It’s a different approach to solving business problems, he concedes. Generally business decisions are made by applying logic, methodologies, and ways of thinking handed down by business leaders.

“Machine learning is very different from that,” he says. “It is an alternative way to tackle problems where you take empirical data, and the data actually develop the code. It’s like applying the scientific method to business Now we can apply algorithms that can look at what has happened in the past and then find out what factors contribute to specific outcomes.”

This is easy to envision for business functions like sales. A company can collect data about what time of day sales happen, and characteristics of the customer including age, geography, and gender, Escapa explains. “Then you let the data develop the code,” that drives advertisements to the right people at the right time for future sales.

“Advertising has been the first field where machine learning has been applied at large scale, which is what let groups like Facebook and Google turn the advertising industry upside down by being able to collect massive amounts of data from customers.”

But advertising is only the low-hanging fruit. “Any business function, decision, or event for which you can collect the data can be addressed with machine learning,” Escapa argues. “There’s no particular restriction on the kinds of problems; the only restriction is that you are able to collect data that are germane.”

That could be a fairly significant caveat. Mathematically, machine learning detects correlations between data and the outcome that we want to understand or measure. Having the right data is essential.

“The value that data scientists bring to a company or to a customer is that they can look at the problem, then they inspect the data or discover the data, and they can assess whether or not those data can help to improve a business outcome,” Escapa explains.  That process could take a few weeks if the right data are available and accessible. But it is rare, he says, that a company has all the germane data in one place.

“Data scientists spend a lot of time—most of their time—collecting and curating data,” he says. Most companies do not have data lakes, at least not yet, he says. Amazon Web Services, of course, has tools to curate data, create data lakes, and build machine learning models.

Many data scientists, though, “have to go to this database over here, or that one over there, or they collect third party data. Most of the effort is actually in consolidating and bringing together all of these data and getting them into a single place, so they can apply the algorithms and build the machine learning models,” he says.

Question Bank

Doing that data curation work is well worth it if you have questions that can deliver business value, but Escapa says decision-makers are not yet sufficiently aware of the range of problems that can be addressed by machine learning. He hopes to “open their eyes just a little bit.”

For instance, the same techniques that drive advertising based on sales data can be used, “to manage a supply chain, where you want to be better at predicting how many items you need to have in the warehouses in what geographical locations so that you can prevent situations where you run out of stock,” he says. Computer vision is being used to carry out visual inspection of a production line, recognizing defects in silicon wafers as they pass. Machine learning models flag suspicious financial transactions—all with greater consistency than even the best trained humans.

And then there’s the tractor.

“Deep learning is beginning to provide a lot of help with precision agriculture,” he explains. Blue River Technologies (acquired by John Deere in 2017) built a machine learning platform within a rig connected to a tractor. The See & Spray scans fields with computer vision, recognizes weeds, and instantly applies localized herbicide.

We are still, in many ways, conditioned by this thinking around software development,” Escapa says. “When we look at the problem, we try to break it down into pieces and think about how we can apply software in order to solve the problem. With machine learning, you kind of do the opposite. You don’t try to break the problem down into pieces. You try to see if you can collect data about that problem—as much data as possible. The bigger the dataset that you have, the more chances there are that will be signals within those data so that you can build a machine learning model.”

For more information, visit the AWS blog.