Reclame AQUI is one of the world’s leading peer review websites. Visitors can leave comments about businesses that they have experience with or they can review products that they have purchased from specific companies.
Businesses are also able to have a presence on the website, allowing them to respond to both comments about experiences as well as the products that they sell. Every month, between 18 and 20 million people sign in to leave reviews, while many others use the site simply to collect information on products and services.
Having this many users visit the site presents two significant issues for Reclame AQUI that both lead back to performance.
The business wished to have a better path to helping users of the site to either leave a review, make a purchase, or read comparison reviews. Any of those actions is counted as a conversion. In order to do this, they needed to be able to predict what a user intended to do based on their initial actions. Having the ability to do this at an early point in the interaction was vital to determining when to intervene with a customized experience that would result in a specific conversion.
In addition, not being able to guide users down the correct path was having a significant effect on the amount of load placed on their system.
The business sought a means of predicting user intention in order to better serve their clients and reduce the load on their servers, which led them to begin a conversation about Kepler with Stradigi AI.
It did not take long for the data analysts at Reclame AQUI to understand the ability of Kepler to help them solve their issue. Within a week, they had assembled a data model that captured close to 10,000 visits that showed the user’s first interactions (e.g. click, scroll, and/or navigation). This data was then inputted into a Kepler workflow that was able to predict whether or not a user would leave a review at the end of their session at an accuracy of 92%.
Next, they took data from 200,000 reviews that would classify the type of review left by users. In order to anticipate a user’s action, (product review vs vendor review), and fed that data into a Kepler workflow, which was able to predict the user intention with 94% accuracy.
Within three weeks, the website was already seeing results from the initial workflow. They were able to increase their conversions by offering their visitors a customized path to conversion from the model they had built in the beginning. They were able to target ads at users that would directly advance them in a path towards conversion.
They were also able to greatly improve the user experience on the site by predicting whether a user was there to leave a product or business review and leading them down that path to a conversion. In turn, this also decreased the load on their servers, creating faster deployment and easier maintenance with improved response times that led to better site performance.
Using Kepler, the in-house team at Reclame AQUI was able to launch this project into production 60% faster than with previous methods, which allowed them to deploy more strategic business initiatives that benefit many aspects of their business.
DATA SCIENTIST
MODELS DEVELOPED IN SIX MONTHS
DAILY API CALLS
