The first thing to do to make sure that Big Data will actually improve performance rather than wreak havoc in a company is to separate myths about artificial intelligence from the real achievements in this area.
Machine vision and image analysis
What is it? Collecting visual data using cameras and analyzing it is one of the fastest evolving technologies. Its obvious advantage is that cameras don’t make mistakes people do and are able to catch what people cannot, such as infrared radiation.
Industries and application. The amount of information you can get from sensors and cameras in machine vision projects is billions of times greater than that in classical projects. For example, analyzing data from cameras in real time using machine vision helps increase production in the mining industry by giving recommendations on changing the direction of drilling. Cameras are indispensable in safety and security in many contexts, tracking movements and issuing warning signals when threats arise.
Problem areas. In insurance business, clients are still required to fill out forms and dial the call center to make a claim, although, unlike a person, a photo or video of the accident from a surveillance camera cannot lie. Still, full transition to digital operation in the insurance business will take time. Other problems have to do with personnel rather than technology. Specialists who know how to use image analysis for the benefit of your business are still in short supply. But this deficiency is gradually shrinking.
What is it? Today’s excitement about machine learning is too high, which is unnerving. However, the concept is quite simple. Machine learning is pure statistics: when algorithms solve a problem, they predict the outcome. Machine learning is matching the prediction against what actually happened and adjusting the algorithms, thereby increasing prediction accuracy.
Industries and application. One of the obvious applications for machine learning might be banking, as banks have tо answer the highly sensitive question about the probability that the client will not pay their mortgage every day.
Problem areas. Machine learning is helpful only when you combine it with human decision-making. The machine can hardly learn anything alone. So it would be kind of premature to expect an unmanned taxi to pick you up tomorrow. The barriers to machine learning are in the minds of people who believe that a machine is capable of learning all it needs to know all by itself, sooner or later, and will make decisions on its own. Machine learning is a tandem of man and machine, not just a machine.
Virtual digital assistants
What is it? A computer program that responds to incoming information will be able to help you and advise you in various situations through various channels – by telephone, video, internet, etc.
Industries and application. It is highly relevant for businesses where making appointments is part of the process. For example, Google has recently made a demonstration of a virtual assistant signing a customer up for a service. The solution can be applied in hairdressing salons, beauty shops or clinics.
Problem areas. Despite all the hype, virtual assistants are still very primitive. But with the improvement of computing resources, software and related infrastructure, they will inevitably grow more sophisticated. When they become able to freely communicate with humans, many organizational processes at companies will change dramatically, and in many areas: from quality control in manufacturing to verifying reports at an office.
What is it? Program robots that can analyze texts, recognize images and draft letters should not be confused with the robots that have been used in car manufacturing for the past 30 years. The latter are used for performing repeated operations and have no learning ability.
Industries and application. Within the next three or four years, most call centers will switch to program robots that can correct their own actions by learning from their mistakes. Some organizations expect productivity to grow by 50% to 60%.
Problem areas. When the capabilities of the assistants that are now distributed across different types of solutions will be combined by software, this will allow significantly improving the repeated tasks performed by people – who will, however, still be in charge of assessing and selecting. This cannot be avoided.
Speech recognition and analysis
What is it? Speech analytics is another rapidly developing technology that can transform audio into text and analyze it in real time.
Industries and application. Many US and European call centers record calls and analyze transcripts. In some cases, the tone conveyed through so-called “voice print” is also recorded. As a result, if a customer calls the call center to report that he wants to complain about an invoice and it is his third call, the system will record “complain,” “invoice” and “third call.” The customer’s discontent will be noted and his call will be moved to the front of the line or transferred directly to the back office.
Problem areas. Not detected. Even today, speech analytics has thousands of applications and their number is growing rapidly. Coupled with image analysis and sensors, it is becoming part of our everyday life regardless of which market segment your company belongs to. Most certainly, speech analytics will be crucial to you both for customer interaction and internal operations.
What is it? Combining sound and image and their simultaneous use for analysis. Basically, it is the analysis of images in combination with speech analytics.
Industries and application. For example, video analytics serves as the basis for increasing the efficiency of teamwork. Video analytics can help to optimize composition of a team, cooperation within the team, interpersonal interaction and work results. Video analytics can help to maximize teams’ potential regardless of what they do, R&D or sales.
Problem areas. Not yet detected – perhaps because the technology is new.
Where to begin?
What should be the first steps of a manager who wants to understand new technologies and apply digital tools to his business?
Implementing new technology as a project must be based on the company’s goals, include task prioritizing and consecutive solving, engaging new specialists and personnel training programs. As the company accumulates successful case studies, discusses them and extends their practices, the company will eventually acquire a critical mass of skills and experience that make the process of changes irreversible. It usually takes about 18 months.
Self-education: First and foremost. It is necessary to create a management team. Take a day or two to learn about deep analytics and its possibilities. It is most important that new methods are understood by the top officials.
The second step is to find the advantages. To outline ‘power spots’ – opportunities opened by deep analytics. The mains questions are: What is the potential of finding power spots and where can they be found? How can the productivity of the personnel be improved? How will digitalization improve the forecast of the company’s needs? How can technology optimize deliveries? To what extent can operational activity and pricing be improved?
The third step is to achieve the task from the point of view of the company staff. The main problem at the stage of the implementation of a project is the personnel’s sentiments and a lack of necessary experts in the company. The usual solution is to outsource.
Who is ahead?
The main areas of AI development are like tidal waves that come one after another.
The first wave is the use of existing data to make decisions based on pure analytics. Today a significant part of deep analytics represents this category. We are currently in the middle of the first wave.
The second wave is the combination of machine vision and image analysis, and voice analytics with machine learning. Against the backdrop of a growing data inflow, the computer becomes a full participant of the dialogue with the analyst or the user. We are in the beginning of this wave now.
The third wave is the transition of AI technology into the physical world, the appearance of robots. For instance, it is the era when robots make you coffee or carry your bags in the supermarket. This wave is only beginning to emerge.
What’s the problem?
The limitations of technology in the introduction of digital technology in business are not as important as personnel, organizational and psychological ones.
The issue that hinders the development of the new technology solutions is by 75% related to experts, mentality and organizational changes that are required to support innovations. This was shown by a poll held by McKinsey among business owners and leaders whose companies are currently in the process of digital transformation.
Technology limits do not bother top managers too much. Only one of each four respondents said there is a lack of processing capacity in solving tasks that require working with big bulks of data, for instance, in creating virtual assistants and physical robots. The capacities of large computers are especially needed in working with big data in real time.
Even fewer business leaders are concerned about the issues of security and technology that allows for creating advanced robots. At the same time, most of respondents note that technological evolutionspeeds up exponentially: tech breakthroughs happen every two-three years.
What you should not do
The most typical mistakes executives and their companies make on the way to digital transformation.
First, you should not centralize analytical skills within one advanced team and expect your organization to absorb it. Sadly, this approach does not work. Guess what is going to happen if specialists in data procession and analysis approach a manager that has a 30-year experience of working in the company and tell him: “This is the model – use it!” The possibility of the model to be used for the correct purpose is minimal. Yet, you should not go to other extremes. The practice of introduction shows that the company should not create chaos where every staff member can experiment with advanced analytics without considering the organization’s common methods and protocols. Such approach can only create further difficulties hard to handle. The formula for success is a correct balance between the order and freedom within business units on the basis of the common corporate methodic standard.
Normally, the organizations that take pride in their intra-corporate digitization projects have several advanced experience centers, where a large number of employees learn new skills. Agile laboratories, which determine variants of using analytics, should work within business areas, from an organizational point of view.
Second, you should precisely define the functions and the role of new digital specialists in the company. A specialist in data processing and analysis – what is he and what does he do? And what about data handling engineer? And data architect? And the so-called ‘translator’, or coordinator? Otherwise, you would not be able to incorporate them into the organization and its processes; there is a major risk of them leaving relatively fast unless you have a clear idea of the digital team members’ functions and their place in the corporate hierarchy. You have to provide highly demanded professionals with enough motivation – a clear plan of their career development and a system of qualification and certification advancement, so that they have an incentive. Sadly, not everyone understands this simple truth.
Third, executives should be aware that creation of value is not simply mathematical processing and analysis of data. The greatest potential lies in employees that already work in your company – those who translate the accumulated experience into information which is then modeled by technology specialists. The change drivers are 10-15% of the best employees that have to be trained in the course of work – including the skills of ‘translating’ that will allow for an expensive use of analytics for business purposes.
When Russian clients start introducing advanced analytics, they often forget about the importance of human resources and organizational changes. As our experience shows, these are the constraining factors in gaining practical usefulness from machine learning technologies. A specific choice of a mechanism for building models and their accuracy contribute no more than 10-20% to the project’s success. An organization’s ability and willingness to use new advanced tools in everyday work are much more significant.
By Alexey Belkin, partner, McKinsey & Company