Part 9: Seven Management Initiatives to Benefit from Artificial Intelligence

From a data-driven to an evidence-based company

The triumph of artificial intelligence (AI) has already begun. Automation, increased efficiency and the raising of cost effects are already contributing to added value in many companies. In the future, AI will increasingly not only generate cost advantages, but also become a strategic factor that encompasses all parts of the value chain. In many cases, the data-driven company is still regarded as the ideal of an agile and adaptable system. The strategic use of AI, which combines artificial and human intelligence to the right extent, can lead to the next stage of business development, evidence-based management. Together, humans and machines form a highly adaptive and rapidly learning unit that shapes the future with foresight and effectiveness. Big data is the raw material, analytics and AI are the tools, but only the skilful handling of both and taking action lead to strategic advantages for companies.

 

What can managers do today to benefit from developments in Artificial Intelligence? The conditions for a start are excellent. AI systems have been developed to such an extent that they can be used with manageable effort and at reasonable costs, and development has not yet come to the point where every field of application would be covered by AI. Those who start now may no longer be among the pioneers, but they do not need to cope with all the technological problems of early development stages and still have a good chance of bringing groundbreaking solutions to market in a particular field of application. These seven initiatives can help to build the business of tomorrow today.

1. Create space for experiments

In general, AI is used via web services and not via locally installed systems. This has many advantages: minimized initial investments, no system maintenance required, no interference with internal IT. On this basis, each company can start its own experiments with AI. The best thing is to set up a small team. In the long term, this can lead to the creation of internal service departments that evaluate the use of AI with small applications for specialist departments and, if the results of the experiments are positive, pave the way to a productive solution.

2. Identify routine activities and test AI

Routine activities are often unpopular and many do not even contribute to value creation. Replacing them with AI routines is a huge efficiency gain. This development has been going on for years in the field of production automation. However, AI offers many more possibilities, namely the relief of routine activities in highly qualified areas, which have often not been regarded as automatable up to now, for example the answering of customer inquiries, the evaluation of legal texts, assessment of diagnostic imaging techniques, etc. First of all, the art is to identify which activities are suitable for this leap. Good cooperation between the AI experiment team and the specialist departments can lead to quick and often astonishing solutions. This does not eliminate any jobs and clearly enhances the role of the human being: from worker to expert and creative.

3. Migrate software development to AI

It may sound brutal to seasoned developers, but the direction is clear: in future, system development based on algorithms will be an exception to the norm. In most cases it will be better to set up and train an AI system via machine learning. This is considerably faster, costs less, delivers better results - at least if sufficient training data sets in good quality are available - and adapts more easily to future developments. It is the next step from agile to evidence-based development.

4. Establish training, assessment and decision-making competence

The advantage of AI is that large and complex amounts of data can be analyzed very quickly, patterns can be recognized and similarities found. As we have seen in the experiments on image analysis and text comprehension, a drastic reduction process precedes which, in the worst case, leads to misjudgements by AI. This can have dramatic consequences, as accidents with Tesla autopilots have shown. Therefore, each company should establish a set of rules which define the following points:

• application and exclusion scenarios for AI
  
  
• quality criteria for training and training data for AI applications
  
  
• differentiation in which cases AI takes over which activities: analysis only; analysis and evaluation; analysis, evaluation and decision; analysis, evaluation, decision and action
  

In addition, there should be training opportunities for using and handling AI, not only for employees in operations, but also for managers right up to the top of the company. Only those who understand what AI can achieve and which limitations and sources of failure exist can use them consistently for the benefit of the company.

5. Examine the business model for opportunities for predictive approaches

AI creates the basis for predicting events and reacting to them proactively or at least at an early stage. Predictive maintenance is currently the most discussed concept. In addition, there are many other starting points, such as recognizing and reacting to market trends, production monitoring and much more. It is exciting to review one's own business model to see to what extent analytics and foresight can make their own value contribution. In some cases, there is data available that can be recycled and sold as a separate service. Sometimes the procedures are in place but the data is not or not sufficiently available. Then partnering with data can help.

6. Develop new business models

AI offers an enormous opportunity to develop completely new business models. Wherever they fit into a company's business model portfolio, it is worth testing them. This is possible with manageable effort. Here are a few suggestions. The list can be continued at will

• Personal assistants for specific tasks
  Alexa, Siri and Co are just the forerunners. Over the next few years, there will be a whole series of new assistants with very different skill sets.
  
  
• Cyber behaviour design
  How humanoid may or must a system be? The more people interact with intelligent machines, the more important it is to design the behavior of the machines. Anyone who has played with the Eliza Bot will have noticed that the Bots reaction in fractions of a second feels kind of unnatural. It's not human. On the other hand, human behaviour does not always make sense. It can be scary. There will be specialists to design that.
  
  
• Data trust center / data broker
  Analyzing your own data is good. Analyzing a lot of data from many sources brings significantly more benefits, such as Connected Machine Learning. However, the willingness to share data is - for good reason - limited. However, in fields such as health care it is crucial for success. Data exchange and collaboration can only succeed if Data Trust Centers and Data Brokers with a very high level of trust are established that control access, anonymization levels and the use of data.
  
  
• Smart autonomous systems provider
  The market is far from saturated. Whoever develops a service in the field of complex (predictive) analytics, cognitive computing, affective computing, decision support, personalized user experiences or similar, which provides users with a real benefit, will find good market opportunities. A market with similar dynamics to that of mobile apps can emerge.

7. Continuous testing, measuring and evaluation

All these points are experiments that need to be evaluated. Fast cycles of experiments, evaluations and adaptations of the idea should become the standard in the company where AI testing is performed. The task of continuously checking results can be assigned to the experimental team mentioned under point 1.

Limits and challenges: lack of transparency and trust

Of course the development is not finished yet and there are still many hurdles to be overcome and obstacles to be removed until AI works smoothly in every application. I would like to highlight two current challenges, which are not of a technical nature, but are decisive for the success of AI.

The first is the lack of transparency in artificial intelligence. It's a standard scene in many detective stories. The ingenious investigator convicts the murderer and once again subtly explains how she managed to convict him by skilfully combining the individual clues, profiling and drawing conclusions. AI can't do that. Currently it works as a black box. It is trained with data and then delivers results that are often amazing. But it cannot say anything about how and why it came to the results. AI can't offer traceability these days. However, the more strategic AI is used, the greater the need to be able to justify decisions or decision recommendations. This calls for additional work by AI developers.

The second critical factor is a lack of trust. Recent scandals of data abuse have further undermined confidence in data processing and in particular data collectors and processors. It certainly won't be the last scandals. It is not only important for citizens and customers to trust data processing companies, but also for companies to have a basic relationship of trust so that they can share and use data. Companies must work on this themselves and develop measures that make them trustworthy business partners. Professionalization and externalization of the trust function, as shown above in the business model example "Data Trust Center / Data Broker", can help.

 

Conclusion

The future will be shaped by artificial intelligence and it has only just begun. Using AI is not magic and doesn't cost a fortune anymore. It is time to start experiments, learn quickly and grow with the trend.

 

---

 

Read more

 

⬅ previous blog entry     

published: December 18, 2018, © Uwe Weinreich

new comment

nick name / alias  

privacy