Of all the technological advances improving daily life, machine learning has perhaps flown under the radar a little – outside of technology circles at least. Most people remain blissfully unaware that sophisticated algorithms are analysing patterns to learn and create data predictions behind the scenes of many consumer technologies that most people know and use every day.
Yet it’s everywhere. Amazon recently announced that it’s expanding its already established machine learning work in predicting product prices, by rolling out its platform for European developers and hiring scientists to help predict sales and fraud.
Then we also have spam filtering in Microsoft Outlook; news ranking, grouping of notifications, and face recognition in Facebook; and, educated guessing for filling out blanks in spreadsheets in Google Docs… these are all examples of machine learning we’ve probably made use of.
> See also: Is machine learning about to go mainstream?
Some firms have made a little more noise about machine learning. Famously, Netflix offered a $1M prize in 2009 for anyone who could improve their recommendation algorithm using ratings and viewing history.
The winning team used a mix of machine learning and conventional algorithms and beat Netflix’ then-current algorithm by 10%. With recommendations driving around 75% of all Netflix video views, the impact was massive.
There’s huge potential for machine learning to have a major impact in a B2B environment through the creation of new revenue streams and cost savings through improving process efficiency.
Last year, Russian search giant Yandex unveiled a new product – Yandex Data Factory – based on the machine learning that it has developed internally for its own services. The tool uses algorithms to help businesses turn large volumes of passive data in to useful business information.
Tradeshift has also invested in machine learning in order to improve supply chain processes. Tradeshift CloudScan is the first product on the platform to use machine learning to create automatic mappings from image files and PDFs into a structured format such as UBL that is suitable for zero-touch processing and the digital supply chain.
It ensures that the learnings of any one company’s mapping of an image into structured format will benefit all companies in the network. With hundreds of thousands of companies in the network, CloudScan grows continuously more accurate with previously unseen data.
Machine learning potential extends much further. For example, it can offer decision support and automation in workflows that were previously formalised and implemented by specialised workflow engineers.
And it can be used to provide real-time insight and prediction from streams of real-time data. The current enterprise approach of extracting reports based on consolidated information in ERP systems has its use cases, but is not scalable across processes and data domains, or in areas where change is common.
Connected processes, straight-through processing, and other P2P applications for machine learning have the potential of completely changing the connectedness landscape of B2B, along with the form and shape of supply chains altogether.
Sourced from Gert Slyvest, CTO and co-founder, Tradeshift