The story of machine learning since around the millenium has been one of exponential growth fueled by the coming together of multiple factors: a massive expansion in the quantities of data being produced, a commensurate surge in the power of the machines called upon to process that data, and the continual improvement, by a growing global community of developers, of the algorithms that enable their work.
The real-world effect of all this has been dramatic: now, for the first time, AI, powered by machine learning, is in the mainstream.
To understand the rise of machine learning, it’s essential to be aware of the vastness of the quantities of data now being produced by human beings – a situation powered by our unprecedented access to the Internet via smartphones and wearable devices.
On just one of the Internet’s most popular sites, YouTube, more than 300 hours of video is now uploaded every minute. Meanwhile, Google processes some 1.2 trillion searches worldwide per year. And while that may sound like a lot, most of this activity still comes from a relatively small proportion of the world’s population – meaning that as technology and prosperity spread, exponential growth is likely to continue.
Then there’s the Internet of Things (IoT), which is already well established in many industrial settings. From a consumer point of view, it’s just kicking off – with smart lighting systems, refrigerators, and cars making their appearance in the affluent homes mostly of Europe, North America, and East Asia. But as costs fall, uptake is set to surge.
The total data accumulated on our planet reached its first zettabyte last year. That’s a unit of measurement too large for most people to comprehend. And yet, by the end of this decade, that total is set to have tripled, fueled by the rollout of 5G mobile data networks.
Machine learning can be incredibly useful to virtually any organization – but exactly how best to harness it will depend on how you want to move forward. We’ll take a closer look at the needs of B2B marketplace owners in a moment, but first, let’s quickly survey the wider landscape of possibilities.
A Recurrent Neural Network (RNN) refers to any app that makes use of ‘temporal’ data such as an audio recording. RNN can facilitate translation from one language to another, or speech recognition – among many other things.
Standard Neural Networks (SNNs) depend on structured data sources such as databases. They then use the information these contain to make complex sets of predictions. For example, if you have an online marketplace for renting out office space, you could use an SNN to give landlords guidance about how much rent to charge for a given space in a given location.
Convolutional Neural Networks (CNNs) are used to provide facilities such as image-based searching. Google already offers this, and for owners of B2B marketplaces, it can be an important aid when a user has a visual reference with no words to accompany it.
Finally, a Custom Neural Network combines an SNN with the powers of a Convolutional Neural Network to produce complex outputs from multiple visual sources. One of the most topical examples of this is in self-driving cars now undergoing testing around the world.
The quick answer is ‘a lot.’ Here are just some possibilities:
- User verification. One of the biggest challenges faced by owners of online marketplaces is in ensuring that users feel able to trust the people at the other end of the transaction. With a B2B marketplace, this is even more important – because that transaction is likely to be large in value, and other parts of your business process will depend upon it. You can use machine learning to process multiple sources of data about your users, so as to reach a reliable understanding of their trustworthiness.
- Product and service recommendations. Every marketplace owner wants to capitalise on users’ existing purchase history to encourage them to make further transactions, but this has to be done in a smart way if it isn’t to be annoying and off-putting. Neural networks can help us to make more accurate predictions about what people really need next.
- Customer care. Customers’ interactions with your company produce a complex array of data, and this can and should be used to adjust the service you provide to their needs. Machine learning can help you provide the sort of tailored service one would traditionally expect from a small business, even when you’re dealing with a large number of users.
- Logistics management. For marketplaces that facilitate the sale of physical products, moving these around the world is a major cost and a major challenge. The best partner is likely to vary from case to case, and using machine learning, it’s possible to select from the many different options available in a consistent, evidence-based, and time-effective manner.
- Borrowing. This goes back a little to our first use category, but deserves special attention. If your marketplace offers any sort of credit facility, you’ll obviously be wanting to make sure users are fully vetted before letting them make purchases on account. Neural networks enable you to do this with maximum objectivity – not only making for greater reliability, but also giving the process the greatest possible feeling of professionalism.
Back when B2B online marketplaces first appeared at the end of the last century, AI seemed like an interesting idea, but one with little immediate scope for usability in most marketplaces. Since then the growth in data, processing power, and the sophistication of algorithms for interpreting data has lead to a dramatic sea change, where neural networks can be used to drive new projects forward right from the outset.
The technical challenges to implementing machine learning solutions still means that they need to be backed up by solid expertise. Increasingly, though, the biggest challenges are creative ones: what new things can we do with machine learning, and how can we apply it to the growing diversity of online B2B marketplaces to help them serve their users in a way that is at once more efficient, and more personal.
