The Polish startup Plastream, the daughter company of Mediacap S.A., has built an analytical platform for the recognition of content in the photo and video. Based on the algorithms its share with clients its Content-to-Commerce platform. The platform uses artificial intelligence and neural networks to search for similarities, automates C2C processes for Publishers and e-commerce stores.
In the very first “episode” of our #PlastreamCoffee, we are talking with Rafał Wyderka, COO in Plastream, who explains how working on the product and their C2C approach looks.
Rafał, the Content-to-Commerce approach means that Content really is Commerce and why?
Monetizing online content is not a new concept. Many publishers manually combine the published content with the advertised products. The challenge is to automate this process, which will allow to scale it to many sites and terabytes of data. Publishers are looking for a way to adequately respond to social mechanisms – the desire to personalise the advertising experience and to look for such methods of monetisation that are not used by competitors yet or are used incompletely. This is standard practice, not only in e-commerce – you can either chase your competitors or overtake them.
One of them is the approach of Content To Commerce automation, which bases its operating mechanics on the use of the ability to create exciting content and combine it with e-commerce products. To improve this process – both concerning speed and accuracy of product’s match – the use of neural networks and machine learning is becoming more and more popular, which drives analysis and search of similar products in online stores that can be presented in the advertisement. We use this mechanism in Plastream – the latest version allows you to generate personalised widgets with products from online stores, and embed them on your pages. Recommendations for widgets can be assigned automatically, and the precision of matches is very high.
What challenges await for those for whom advertising is one of the primary sources of sales revenue?
Recipients receive thousands of advertising messages every day, which are difficult for them in acquiring and even noticing. Additionally, their use of AdBlock software effectively blocks unwanted marketing content. The lack of matching content and reckless ads don’t cause shopping impulses and result in the lack of purchase decisions. In the “banner blindness” era, it’s crucial to publishers who are struggling for customers as well as for e-commerce entities that try to encourage Internet users to purchase. Both sides of the transaction can’t stand still, so they continuously strive to increase the quality of user experience, and one of the most effective ways is to provide the user with the content they need, which they want to engage in, which they may desire in a given context.
However the term “desired adverts” does exist?
Recipients will more often and more willingly engage in interaction with the brand that offers them solutions tailored to their needs. Example: I want to buy a product, which I see in the film or picture. This is a specific value for the client that goes beyond the traditional ads. At the same time, it will be easier for them to click on the ad in a contextual match to what they’ve just seen and what may have attracted their attention. That’s why the context plays the role of a specific, natural intermediary between the content and e-commerce activities. Every day, however, billions of new content are created, and manual adjustment of contextual advertising becomes very difficult or requires a lot of work, time and people. New technologies come to the rescue – for example, deep learning, which is used in Plastream to improve the model recognising objects in photo and video regularly. Certainly, the car manufacturer wouldn’t want to display ad next to the photo showing a car accident, despite the proper recognition of the brand in the picture or the film. This is just one example where the context can significantly affect the reception of the ad.
How does it work in practice?
Our tests proved that ads strongly related to the content of an article or video image offer a measurable benefit in the form of the possibility of immediate purchase of items shown in the material record above-average CTR – from 1% to even 20% in case the product is ideally suited to the content.
Imagine a couple walking on the beach. Such a scene in a TV series, commercial or movie perfectly matches the advertising of dresses, airlines, summer clothing or accessories. Showing these types of products or services in the context of a selected scene increases the likelihood of higher conversion and better CTR.
Another example, a photo of any celebrity in a beautiful, red dress. Thanks to the recommendations, it’s possible to match the picture to the same or very similar product available in the selected online shop. The chances of going to the purchase page increase thanks to the context: showing the ads in a non-invasive and non-intrusive way exactly where people interested in buying could look for answers.
There are many possibilities of using such mechanisms – although we’re currently focusing on the fashion and retail industries, shortly we can see the Plastream application for many others. We’re constantly developing the product.
How does it technically work?
Platform mechanisms such as Plastream, thanks to a few-element analysis – both image/video and text (text search allows you to focus only on the analysis of the described element, not all objects in the content), can automatically identify the most precise matches and generate results that can be personalized as part of widgets on interesting videos or photos. Models are constantly “learning” thanks to machine/deep learning techniques – the longer and more they can recognise, the better the match becomes. This is an interesting statement that gives us many new results and inspirations every day.
In Plastream, the products in the video/photo are identified and classified, based on each content detection. This process is supported by the earlier textual analysis, which so to say ensures the results generated by the engine and tags them. Next, the game involves mechanisms looking for similarities – a technological machine combining content (photo/video) with commerce (products): searching for matches in online stores. The publisher can see the effect of it in the panel.
So Plastream is a typical Content to Commerce platform?
Yes and no. In addition to the widget panel, available to publishers, and opportunities to cooperate with advertisers (they can submit their online store free of charge if they operate in the widely understood fashion industry), Plastream also works with some companies, offering its technology to be implemented in internal organisation systems. We already cooperate with Allegro and TVN. Recently, our company has been incorporated into the nVidia Inception Program, collecting the most innovative companies in the world using GPU technology and processes, which are necessary to train models and process vast amounts of data in the cloud.
It’s important to use new technologies in improving the processes and scaling the business of Publishers and e-Commerce networks, who can automate what they often already do: manually or to a limited extent. Advertisers get the opportunity to gain valuable traffic that can translate into higher sales. The potential of contextual advertising and precise adjustments is appreciated, as well as the power of the message of Advertisers, who have a source of quality traffic with high chances for conversion. This is another step towards optimising the business.
I believe that the key essential here is the end customer, who chooses by buying what he sees in the picture or the video material. On demand.