In 2010, 636 million litres of oil gushed into the Gulf of Mexico when the Deep Water Horizon drilling platform exploded and sank. It took 87 days for the BP oil company to plug the seabed, but in the meantime the company had tried to find solutions to the problem, among other things through what we now call open innovation; via crowdsourcing with a digital mailbox for solution suggestions. In a very short time, this generated 144,000 ideas for the company from people all over the world.
The challenge was - and still is - that it is difficult to manage and process so many ideas, and this encouraged me and my colleagues to think about whether it was possible to identify the best ideas more quickly and easily," says Lars Frederiksen.
Lars is a professor at the Department of Management at Aarhus BSS, and he has co-authored the article "In Search of New Product Ideas: Identifying Ideas in Online Communities by Machine Learning and Text Mining”. Researchers are developing an automation tool to optimise identification of new, useful ideas in online communities - ideas that could have a great impact on product development and innovation at companies. However, these ideas are normally cumbersome and cost-intensive to track down, because people with experience and expertise have to trawl through everything users enter on the company's online forums to identify the tiny part of the content that is actually the original idea.
But by using machine learning, whereby a test data set can teach a computer to improve its own ability to find and recognise patterns, and text mining, which automatically maps the use of specific words, links between words and sentence structures in a collection of online texts, Lars Frederiksen and his colleagues have developed an algorithm that, with around 90% probability, is able to identify the information in the vast ocean of information that actually represents ideas.
"The project shows that human selection can be replaced reasonably well with an algorithm. But this doesn’t mean that we don’t need people any longer. When the algorithm has identified the ideas, we still need human judgment to qualify and activate the ideas, so they become an innovation benefit for the company. The advantage is, however, that they can do this on the basis of a very much smaller amount of information than if the algorithm hadn’t filtered the information and found the small fraction that represents new and useful ideas," says Lars Frederiksen.
According to Las Frederiksen, the perspectives are particularly bright for small and medium-sized enterprises, for example in the service industry, or perhaps for manufacturers of fast-moving consumer goods such as toys, shoes or ice cream, where you need ideas for new flavours, shapes, consistency, etc. In this context, it is relatively rare for users to create a new ice cream or a new bike, but companies can get ideas and preferences. And the algorithm is a faster way to search through ideas for product development - and not least to make a valid selection from these.
"If you want to find additional information for product development, this is an opportunity to do so at low cost, although it won't be free. You have to go through a number of difficult steps before you reach your goal: First, you have to create a community with users and customers, then train the algorithm, implement it, etc., etc. So there’s still a number two selection step, but this is now with more qualified ideas than without the algorithm," explains Lars Frederiksen, and he goes on:
"It’ll be cheaper and faster to find the gold in your online community, so perhaps this is a brand new opportunity for SMEs and they can start experimenting to become more digitally prepared. Too many small companies seem to be having difficulties starting their digital transformation".
"Data will be necessary for virtually all companies in order to predict a large part of their development - even companies which change tyres and sell sausages."
Lars Frederiksen - professor, Department of Management, Aarhus BSS
When companies experiment with new technology, a number of new challenges will almost always emerge as well.
Firstly, you must have employees who can code robots, do text mining etc., or you need to recruit them. Companies frequently lack the necessary skills in this area. In addition, the article "The Barriers to Recruiting and Employing Digital Talent", which was recently published in the Harvard Business Review, shows that one of the biggest challenges facing companies today is to find digital talent. Companies will therefore have to be creative in their recruitment and employee development, but many of the ways in which companies are trying to increase their digital skills have innate challenges with regard to integrating the digital talent into the core business, exploiting the digital skills throughout the organisation, and implementing bottom-up initiatives.
Secondly, it is often impossible for small companies to build a comprehensive digital setup from day one, so perhaps they will start at a more modest level to become familiar with digital technologies.
"Perhaps a company could invest in a camera with facial recognition so you can open doors without keys, or maybe get a parking sensor that can register when people arrive and leave the workplace. Slowly, slowly they can find out what creates value for precisely their own specific business," says Lars Frederiksen.
This leads to the third challenge: to clarify where the technology should be used. Is it to get an idea of demand, complete market analyses or something completely different?
"In this context, companies must first overcome their reluctance to experiment with digital options. They can send employees on courses, look at the literature, or make a minor investment to help them on their way," says Lars Frederiksen, and he stresses that, above all this, is a fourth and much more fundamental challenge: The vast majority of companies have to start thinking about collecting and processing data.
"Data will be necessary for virtually all companies in order to predict a large part of their development - even companies which change tyres and sell sausages."
In addition to the empirical challenges of where and when companies should use different digital technologies, Lars Frederiksen's article also asks new questions, purely in a research perspective.
The article has analysed 3,000 entries from the LUGNET online community, in which Lego fans share their experiences and information, and the article introduces the question of how the algorithm does what it does, and why it does it. For example, by finding out what stable and meaningful semantic representations the algorithm can represent across contributions. Or in other words, what characterises what we human beings perceive as an idea - is it the actual sentence structure, is it the use of verbs, or is it the composition or the complexity of the words or perhaps adjectives that mean we see something as an (good) idea?
Moreover, it could also be interesting to investigate what to use a human assessment for, and what to use a robot-assessment for. Perhaps the robot is good at finding complexity and originality, but not at assessing whether the ideas are useful in the real world. Of course, an idea is usually only a good idea if it can be implemented. So how should the employees in company R & D departments find the right division of labour with robots, so that robots do the part of the work where they create the greatest value and where the human assessment is used in the right places?
If we can find some of the answers to these questions, perhaps in the future it won’t take 87 days to plug a hole in the seabed.
FACTS: What the researchers did
|
Sources:
Christensen K, Nørskov S, Frederiksen L, Scholderer J. In Search of New Product Ideas: Identifying Ideas in Online Communities by Machine Learning and Text Mining. Creativity and Innovation Management, Volume 26, Issue 1, March 2017, Pages 17-30
Christensen K, Scholderer J, Hersleth SA, et al. How good are ideas identified by an automatic idea detection system? Creat Innov Manag. 2018;27:23-31
Dahlander, L, Wallin, M. The Barriers to Recruiting and Employing Digital Talent. Harvard Business Review, July 09 2018