Predicting supply chain risk just got a whole lot more social
Start-up Prewave wants to turn the tsunami of social and open-source data washing over the globe into meaningful alerts on supply chain risks and resilience
If you’re one of the world’s largest manufacturers, then your supply chain is a huge and complex patchwork quilt of suppliers that blankets the globe. Volkswagen, for example, notes more than 40,000 suppliers. Even smaller companies will likely still have hundreds of suppliers and this creates a headache for any planner: How can you assess potential risks, bad actors and points of weakness in order to prevent upstream failures severely impacting your downstream operations?
Clearly, auditing them one-by-one is a herculean task, but what if you could source the wisdom of the crowd and localised knowledge to give you a head start on what is happening on the ground in your supply chain?
Start-up Prewave believes that it can do that by using machine learning to process open-source intelligence from across the web, including local news outlets and social media, and distilling this down into alerts on companies in any of the 50 languages it currently processes information in.
Crowdsourcing supply chain visibility
“The idea and the concept from the very outset was why don't we use social media and news media as the source to try to capture those risks and those incidents on a global scale?” explains Harald Nitschinger, Co-Founder & Managing Director of Prewave.
We have quickly found that there is a vast sea of information and, in many cases … that is sort of hidden from us in the West
“We have quickly found that there is a vast sea of information and, in many cases … that is sort of hidden from us in the West, because the information resides in local language [and] in local media and social media channels.” For example, there could be “Citizens in Indonesia simply tweeting or complaining in Indonesian that this or that factory is polluting, but how do you capture that? And how do you reach that?”
We predicted a port worker strike 18 days in advance
The answer was to build a system that would capture this information, scraping it from public social media accounts, local news services and government agencies, before crunching that through a machine learning system. Their approach then translates and breaks down the language used and then assign relevance to that information from a supply chain perspective. Finally, it looks at the volume of information and how the elements link together to understand whether to elevate it to a major risk that a company would need to be aware of via an alert.
“We focus on the very localised local language messages,” explains Nitschinger “and oftentimes, those give you a big head start in following [an] event through its early development. So, for instance, we have a case study … where we predicted a port worker strike 18 days in advance. Prediction in that sense simply means that we picked up on the very first stages of that of that event.
“This could be oftentimes months or weeks before that is then put on a larger stage by an NGO, or by some sort of news platform,” claims Nitschinger. By catching the potential for a risk to open into a shutdown, supply chain planners have one more tool to give them the heads-up that they need to prepare back-ups to mitigate the risk, or that they should be exploring alternative suppliers.
Furthermore, Nitschinger points out that their “Approach of using open-source intelligence and publicly available data is a permission-less approach. You don't need to ask your supplier for permission. You don't need to cooperate…. You can roll out that that visibility and transparency along your entire supply chain, without asking anybody for permission.”
The close relation between sustainability and supplier risk
Nitschinger is excited about the potential for the platform to provide a foundation for understanding supply chain sustainability and really linking that to resilience and performance in a concrete way.
There is a strong and significant link between sustainability and resilience of supply chains
“There is a strong and significant link between sustainability and resilience of supply chains,” he says “because a supplier that is not adhering to certain labour standards, is much more prone to labour unrest and labour strikes and disruptions. The same holds true for legal issues like corruption. These are often precursors to regulatory shutdowns a few weeks or months down the line,” as are issues around negligence and polluting local environments, which can then go “all the way to potential insolvencies and financial issues. So, it's all interrelated.”
Not looking at incidents that are classified typically as sustainability-related, things like pollution and working conditions, is a big mistake
He thinks that “from a resilience standpoint, not looking at incidents that are classified typically as sustainability-related, things like pollution and working conditions, is a big mistake,” for supply chain planners. “I think most companies have realised that and care very much about the compliance … but also the sustainability of their suppliers and for good reasons.”
“This is also has a very practical background, because in Germany, and on the EU level, there is now increasing regulation of requiring companies to actually conduct stringent due diligence within their supply chains,” he points out. Most prominent of these new laws have been in Germany and the UK, with the former planning to make firms liable for human rights and environmental abuses by suppliers and the latter strengthening requirements around due diligence for slavery among suppliers and looking at adding in laws to reduce deforestation.
A challenge of scale, a solution of relevance
“It's all about capturing data, and then filtering it down in a way that's that reduces it to a few actionable data points that actually are relevant to our customers,” says Nitschinger. “Well, how does this work? And what does this mean? So … we use machine learning models that allow us to really understand, in an automated fashion, whether a text, and it might be a tweet, it might be a news snippet, is actually relevant or not.”
“The fundamental challenge in natural language processing is simply capturing the semantic relevance of a written text,” he says, “which comes so naturally to us as humans. Reading the difference between … the air strike and the baseball strike and the labour strike … these are simple things where non-machine learning, artificial intelligence-based approaches fail very quickly. So, the answer is and has been in those cases to not explicitly program the machine to understand certain keywords, but to rather train an algorithm based on examples to understand patterns.”
As long as you show enough examples to the algorithm, that the algorithm figures out the patterns that are actually relevant
The way they have done this is to take large volumes of data to give the machine-learning programme examples and then “You need to take a text, you need to split it up into what's called tokens, which are basically the words that the text consists of. That is different for Chinese sentence than it is for English language sentence, but at the end of the day, most of those questions are actually solved by simply creating an algorithm with examples, and not explicitly defining what's relevant, but rather letting the algorithm, in a sense, figure that out for itself.” He explains that for their machine learning approach “As long as you show enough examples to the algorithm, that the algorithm figures out the patterns that are actually relevant, and differentiates relevant text from a non-relevant text.”
For a labour strike at a factory or a seaport, you typically have thousands, or even hundreds of sources across different mediums: Social media, news media posts, YouTube videos
Then it is a case of deciding relevance and relationship between the different text items that have been contextualised. “For a labour strike at a factory or a seaport, you typically have thousands, or even hundreds of sources across different mediums: Social media, news media posts, YouTube videos, and so on. Then what we do, and this is actually the second important element, is we actually cluster all of that together into a single risk alert. So we understand that there's a labour strike happening - who's affected, when's it happening? How big and important is it?”
For the Austrian start-up, the main challenge after they germinated the idea in university was “Mainly in scaling it across the 50 languages we currently cover because, of course, the resources and the processing differs from language to language, from language class to the language cluster. So that's, of course been a huge challenge,” but now their algorithm has got its AI-brain around the many weird and wonderful aspects of the human language, they hope to make supply chains more open around the world.