Home » Machine Learning with Adapt – The results of the first beta users

Machine Learning with Adapt – The results of the first beta users

Operational Program Competitiveness 2014-2020
Project financed by the European Fund for Regional Development

Press Release by Omniconvert

Adapt – what it stands for and its first results

Humans crave connections. This is why brick-and-mortar stores still represent a significant part of commerce: 89.1% in the US, in 2019, according to an estimation by emarketer.com. And that is also why e-commerce growth is still limited.

One way to increase the conversion rate in e-commerce is to mimic the human connection that takes place within the space of a physical store. In the online space, graphics that are displayed within the website can create an interactive experience for the customer – these are overlays.

Because planning this behavior properly beforehand is complex and guaranteed to produce suboptimal results, we, the team of Omniconvert, designed and created Adapt.

Problems that are addressed

Welcoming the customer – Similar to how a sales consultant welcomes you when entering a physical store, we created a class of overlays can do the exact same thing creating a note of personalization for the customer.

Choice reduction – Having a too broad catalog has been proven to produce a paradox of choice, and actually, prevent the customer to decide on buying. This is why we created overlays for pushing the search and the filters for the customer to easily use, by directing the customers’ attention to them.

Social proof – Everybody wants to belong to a group. By pointing out that other people are looking at or have bought a product, it increases the chances that a user decides on buying.

Fear, Uncertainty and Doubt – Most people have second thoughts before making a purchase, so reassuring the customer that it is a good deal is a great idea.

How the product works

Creating an efficient one-size-fits-all plan to display these overlays for customers is nearly impossible because every customer has its own journey.

Adapt chooses the right overlay for the customer, based on their similarity with other customers, and their track record.

The website owner still needs to perform an initial setup and create the interactions that the website needs to emulate. It is indicated that multiple content styles are used for addressing the same issue.

Technical implementation details

The problem that Adapt is resolving is an unsupervised classification problem, used to recommend overlays based on purchases.

Adapt uses clustering for determining the chance that an overlay helps the customer towards the purchase decision. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is an algorithm that is being used, due to low complexity and good performance.

Data is collected from the website in the form of events, such as page open/view/scroll, cart and wishlist view, filter and search interactions, and user attributes, such as number of sessions, time spent on site, number of orders, data regarding the last order, and more.

Training is performed daily to update the model, which consists of the actual clusters of users, against which the prediction is being performed, with a separate model for each overlay.

The traffic is split into 3 adjustable-by-website-owner groups of users: control (unaffected users), learning (users in random tests) and optimization (users in the prediction based on machine learning generated models).

Initial results

Currently, Adapt is still in its initial phase, of Learning and building the models.

Results so far include:

  • 4 tested websites
  • 1.3M+ users
  • 2.2M+ sessions
  • 90M+ events collected

The initial conversion rate spans from 0.9% to 1.3%.

After enough data has been collected, it will be switched into the Optimization phase and the impact on Conversion Rate will be closely monitored.

The Adapt software was developed with the financial support from the European Fund for Regional Development through the Operational Program Competitiveness 2014-2020, project MySMis code 2017+115806, having a total value of 3,557,577.52 RON, from which 1.855.368,68 RON is the non refundable financial assistance from the FEDR budget and 463.842,17 RON the non refundable financial assistance from the Romanian national budget.


Omniconvert SRL
Str. Vasile Stroescu nr 14, etaj 1, sector 2, Bucuresti
Web: www.omniconvert.com
Email: contact@omniconvert.com

About the author

Atish Ranjan

Atish Ranjan is an established and independent voice dedicated to providing you with unique, well-researched and original information from the field of technology, SEO, social media, and blogging. He has in-depth knowledge of computers and tech as he pursued computer science.

Add Comment

Click here to post a comment

All the data shown above will be stored by Techtricksworld.com on https://www.techtricksworld.com. At any point of time, you can contact us and select the data you wish to anonymise or delete so it cannot be linked to your email address any longer. When your data is anonymised or deleted, you will receive an email confirmation. We also use cookies and/or similar technologies to analyse customer behaviour, administer the website, track users' movements, and to collect information about users. This is done in order to personalise and enhance your experience with us.

Pin It on Pinterest