It’s hard to escape the hype surrounding machine learning in 2018. Everyone is talking about the technology as a conference topic, content of newspaper articles or trend in product development.
With the often speculative reporting (robot apocalypse, secret computer languages, artificial brains and ” machine feelings” have all been conjured up this year) and the ever-same examples, it is not always easy to find out what potential machine learning really has for companies and in which areas it pays to use it. Especially for companies whose business model is not originally digital, it is important to soberly evaluate meaningful application areas for the technology and not to rely on blanket promises of progress.
Significant improvements of business processes possible
The changes that result from the correct use of machine learning in companies are often considerable. Especially in the automation of business processes new dimensions of optimization are reached. Amazon, for example, can offer delivery times of less than one hour for some areas and products through predictive logistics. Google has achieved a cost reduction of 40% by using deep learning in data center cooling. More traditional companies are also taking advantage of the opportunities offered by machine learning to make significant improvements. One prominent example is the Zurich Insurance Group, which has been able to reduce the processing time for relevant claims through machine learning from an average of one hour to five seconds. An apt comparison is therefore that of machine learning with relational databases. Just as in the 1970s database technology became unavoidable for all companies that wanted to keep up with the competition, in a few years there will probably be only a few competitive companies that do not use machine learning methods in any area.
Most companies still have low hanging fruits
The possibilities for a profitable use of the technology are already numerous in most companies today. As a result of increasing digitalization, increasing internet usage and the Internet of Things, more and more data is becoming available. The majority of this data treasure is not used. This is not surprising, as most companies do not yet have the necessary skills for advanced use. In a recent study conducted by technology publisher O’Reilly, 50% of participants said they had no significant experience with machine learning. (Interestingly, the difference between the US and Western Europe in these statistics was very small – so the idea that Europe is generally inferior to the US in the use of artificial intelligence is not necessarily correct.)
Almost every business model offers an opportunity to create tangible improvements with small, clearly defined machine learning projects. “Low hanging fruits”, which are unused in many companies, are for example
- A more targeted sales management through lead scoring / predictive sales analytics
- Better pricing through machine learning based pricing models
- Lower maintenance costs through predictive maintenance
- Efficient and better customer service through prioritization of incoming e-mails according to their urgency
- Predictive logistics through machine learning based prediction of demand quantities
In these examples, as in many machine learning applications in today’s business processes, human work is not completely replaced, but complemented. Just as the rise of Microsoft Excel has not led to the abolition of controlling departments, but has made employees in that department faster and more effective, hardly any machine learning solution takes over an entire business process.
Interestingly, finding innovative machine learning solutions is not necessarily reserved for the big technology companies. For example, the Cologne-based startup DeepL, with just 20 employees, has managed to produce better machine translations than Google, Microsoft or Facebook (Link) – in fact, this article has been translated by DeepL with minimal adjustment by the author. And in the renowned DAWNBench competition for artificial intelligence at Stanford University, a team of four data scientists from the startup fast.ai won this year in the field of fast image classification and beat the submissions from Google and others (Link).
“Walk before you run” – creating acceptance in the organisation with quick and visible improvements
The large-scale application of machine learning in a company requires a number of prerequisites in order to be successful. The data science team must have the necessary experience in model development and statistical knowledge. The required data must be available and an architecture for suitable data pipelines must exist. The skills to correctly integrate a model into the relevant production systems, e.g. via APIs, must be available. And last but not least, the data science team must be so well integrated with the business departments that the resulting solution is not only a technological showcase, but also supports the actual business processes. Small projects, limited to one business area, which achieve a visible improvement within a few weeks, create acceptance in the organisation and are the first step towards a large-scale use of machine learning in the company.
Fintu Data Science implements customized data science and machine learning solutions to automate and optimize processes. We combine in-depth machine learning expertise with the necessary strategy and business experience to successfully apply the technologies in the corporate environment. Our clients include medium-sized companies, start-ups and NGOs from Germany and Europe.