This is why old industries will be the first winners in the AI revolution

Posted: 21 March 2017 | By Darcie Thompson-Fields

Like all technologies going through their “hype cycle”, machine learning and AI expectations are higher than ever. Yet, many believe these technologies have nothing to do with their business.

Indeed, media conversations have focused increasingly on the prospect of new business models arising through the shared economy, AI-powered assistants and chatbots, or of course self-driving cars flooding the streets, leaving the perception that all the disruption will happen within the “new” economy.

They couldn’t be more wrong. The real AI and machine learning potential of today is not only very different, it’s a lot more attainable.

The change

While it is certain that machine learning and AI have the potential to revolutionise our lives, it is also clear that it will take decades for innovations such as self-driving cars to become a part of our day-to-day world. What’s more, it will require billions of investment, and the profits will only be seen in the long-term.

In the meantime, however, there exist a whole other set of AI applications that do not disrupt the world as we know it, but rather improve what already exists, making processes more precise, efficient, or personalized. For example, a smart algorithm can individually choose the offers for an upsell campaign, adjust production parameters to lower energy consumption, or forecast demand by taking into account more factors than ever before. In these instances, it is all about optimisation.

Old and mature industries, such as manufacturing, retail or logistics, will be the first to take advantage of these opportunities, with many already doing so, possessing all the prerequisites to succeed.

First, these industries have established processes and subsequently, abundant historical data to learn from. Ultimately, this is what is needed for AI-based optimisation: a known process with specific KPIs and constraints, and past data on how it went for the last thousands of times. And while it is often the impression that older industries are slow to embrace change and new technologies, when it comes to machine learning and AI, the wealth of historical data available will in fact bolster their machine learning and AI projects and put these industries in an advantageous position.

Second, old industries value optimisation, and have already used all other available options. For mature, capital-intensive industries, getting an additional 2-3% percent optimisation often requires introducing new equipment, or upgrading technologies – a costly, time-consuming effort with no return on investment for several years. Machine learning on the other hand brings a revolutionary opportunity to achieve these extra efficiencies in just a matter of months and with no capital investment.

Preparation over hesitation

When compared to the futuristic and almost “outlandish” use cases, achieving extra percents somewhere at production level is a relatively dull prospect. Nevertheless, these real-life applications are exactly what impact the bottom line today.

If businesses are to embrace these opportunities, they must quickly take practical steps to lay the foundations for success. To start, they must discover the processes where the technology can bring the greatest value right away.

As a rule of a thumb, it makes sense to investigate the processes that, on the one hand, incur significant costs, and, on the other hand, have already been tuned and optimised with the known means. Especially if these processes include human or statistical judgement to predict future events. That is where AI is best placed to deliver the extra efficiencies. For example, in retail this may be a replenishment process, which, if not precise enough, leads to overstock and understock losses. For businesses that hire thousands of employees per year, AI may help in the form of candidate screening, where it could help to dodge recruiting costs. On production lines, AI could be real-time tuning of process parameters to improve output at the same cost, and so on.

Organisations should set up pilot projects, in order to test and prove the value of the technology in practice. This would help to guide further data collection strategies, and pave the larger transformation projects based on actual results and not some theoretic AI roadmap born in Powerpoint slides.

Machine learning in action

As mentioned previously, one such industry benefiting from machine learning is manufacturing. In 2016, one of the world’s largest steel producers, Magnitogorsk Iron & Steel Works (MMK), did just that. As with all manufacturers, MMK must meet and adhere to international quality standards and certifications. But striking the perfect balance between cost-efficiency and quality is not always a straightforward task – and that is where machine learning comes in.

A machine learning recommendation system was created to predict the optimal amount of ferroalloys needed during smelting, to produce high quality steel at the lowest possible cost. In this case, AI delivered additional accuracy on top of physics-based models by analysing the patterns in seven years of steelmaking data. The estimated reduction in use of ferroalloys was 5%, equating to an annual saving of more than $4 million in production costs, whilst maintaining the same quality of steel output and the same process. The ability to reach similar direct, measurable results without capital investments and long return period explains why for traditional industries new technologies bring nothing short of a revolution.

While we all envision almost far-fetched ideals about what our work life could be like in five or ten years, the enhancements machine learning can bring to industries today are significant, albeit somewhat more realistic. And the “old and boring” industries, not often usually associated with change, are now in an advantageous position to harness these new technologies and stay ahead of the curve.

Jane Zavalishina is CEO of Yandex Data Factory 

Related industries

Related functions

Related organisations