Skip to main content

Q&A: What machine learning can bring to your business

CEO of Yandex Data Factory, Jane Zavalishina says that to be a truly data-driven company you must be willing to move from big data to an age of machines and algorithms. With business experts warning that, while valuable, data is fast becoming a commodity - and with more and more companies looking to invest in machine learning technologies – she may have a point.

One such company is Russian mobile operator Beeline – a subsidiary of VimpelCom – who has recently partnered with Yandex Data Factory to create a portfolio of new services based on analysis of its big data. The two companies will work together to enable Beeline to better predict mobile subscribers’ churn, forecast demand for mobile Internet services, improve call-centre salesforce efficiency and optimise its advertising spend through personalised targeting – using the technologies born out of Russia’s largest search engine, Yandex.

Here Jane explains why all industries will eventually have to follow this approach to big data analytics.

Please give a general overview of Yandex Data Factory – who you are, what you do and who you currently work with?

Yandex Data Factory was born out of Yandex – Russia's leading search engine and one of the largest Internet businesses in Europe. As a search engine, Yandex has invested in data processing, analytics and machine learning for almost two decades – long before the term “big data” was a buzzword.

The idea behind Yandex Data Factory is to use all these assets to bring the power of machine learning and predictive analytics to other industries: from retail, finance or telecommunications to manufacturing and logistics. When working with customers, the focus for us is not on generating knowledge and insights from the data, but in applying machine learning to automate decision-making. This means we develop algorithms that are directly embedded into business processes to optimise costs and help revenue growth.

For example, we are working with AstraZeneca, a global, science-led biopharmaceutical business; Magnitogorsk Iron and Steel Works (MMK) and VimpelComs’ Beeline, a telecommunications brand.

What is machine learning and why should CIOs start to focus on it over other trends in big data?

Big data is the new oil – everyone wants it and expects to gain large revenues from it, but in itself it can bring more issues than value. Why? Many companies jump into big data ventures without a clear understanding of how to measure ROI – leaving them disappointed about the investment. When thinking about big data, many managers expect to gain new knowledge about their customers, understand their behaviour and obtain insight into how to optimise business processes. Yet while this decision support is important for strategy building, the value is hard to measure.

To get measurable and direct ROI out of big data, businesses must focus on machine learning – smart algorithms that can make predictions and decisions at scale and continuously learn and improve from new data. When a human looks into every particular case to make a decision, our abilities are amazing. But when you need to make a decision about every customer among hundreds of thousands of them – for example, to provide the best recommendation for what they may want to buy – machines do the job better.

The ability to deliver this level of automation in many areas of the business is why machine learning should be front of mind for every C-level manager now, not only the CIO.

How does machine learning differ from general big data analysis, and what benefits does it bring to business?

Machine learning moves beyond traditional descriptive analytics that looks into the past, describes what happened, and serves as food for thought for humans; to predictive and prescriptive analytics. This means the machine recommends best actions based on probabilistic models of future events and expected reactions. Therefore, instead of aiding human analytics, machine learning replaces it – well, in certain well-defined cases. This is: when we know what we want to optimise and have enough data to teach the algorithm and freedom to experiment.

On top of this, once an algorithm is put in place you are guaranteed measurable results that can be verified, compared against existing solutions, and expressed in terms of revenue generated or costs saved. For example, how many more products customers are purchasing after they receive a personalised recommendation in their e-mail.

What challenges have your customers faced when applying machine learning solutions to everyday business?

There’s still a way to go until these technologies are perceived as part of normal business practice.

It is often hard for companies from traditional industries to understand the scope of technologies and to define use cases – a lot of our work with clients includes brainstorming together on defining the best applications that will have direct impact on their business.

Also, many underestimate the data they have, often wanting to build their infrastructure and clean the data first. Yet in fact, they may have everything in place to have the first pilots, define the business cases for machine learning and prove the effect. Even when the data is “dirty” and unstructured.

Bringing such innovations into business also means there is a need to change the culture. The whole idea of working with and trusting decisions made by “robots” is often challenging for traditional companies. It requires establishing the culture of experimentation and learning to run tests as part of a normal work process.

How will Beeline be using machine learning technology from Yandex Data Factory?

In partnership with Beeline, we will create a set of new services for them based on big data analytics. One of which is churn prediction – defining and pinpointing the customers who may leave the service, so that pre-emptive actions can be taken.

We will also help increase sales of mobile Internet services by finding the customers in the existing base who are likely to start using data services, provide personalised ad targeting, and work on improving call-centre sales force efficiency.

How advanced is the telecoms industry in their adoption of big data analytics and machine learning?

The telecoms industry was among the first to follow the trend. Largely because it is a “big data” industry, acquiring large volumes daily - customer profiles, billing and payments data, history of communications, geolocation data, network equipment telemetry, etc. – which is perfect for applying big data analytics technologies.

On top of this, the industry is very focused on customer service. This makes the possibility of using machine learning to optimise and improve customer experience very interesting to telecoms players.

I would say of all of the consumer-facing industries telecoms comes in second place, after e-commerce, for its growing adoption of data analytics technologies.

And how does this differ with other industries?

The telecoms industry knows its data, understands its value, and is rather advanced in IT. That is very different to, for example agriculture, where the business is not used to thinking about data in a commercial way and the cycles are long. However, this doesn’t make the potential less – precision agriculture and remote monitoring with computer vision are just a few of the possible applications for this industry.

Finally, how do you think machine learning will evolve during the next few years?

It is clear that we are facing a new industrial revolution imminently, as technological advancements make it possible to replace humans with intelligent algorithms. The scale of the change happening makes machine learning a priority not only for IT companies, but for all businesses. It will be changing the workforce, new professions will develop, and some will become out-dated. On the consumer side, personalisation of everything will soon become a norm, not only in commerce, but also in other areas like medicine.

Image Credit: Sarah Holmlund / Shutterstock