Abstract:

Artificial intelligence and allied technologies are driving processes which were till now available only to humans such as learning, problem solving, pattern recognition and decision making. Curated data can be used to train models using statistical methods for solving problems related to classification of data and predicting outputs. Natural language processing, entity extraction, machine learning and deep learning are a few areas in Artificial intelligence which are making an impact in the world of technology. Futuristic scenarios such as Terminator aside, AI is already here and making definitive difference in domains such as Finance, Medicine, Science and Technology just to name a few.

The customer is a primary player in the discount card market in the UK and Europe region. The products offered by these customers include student cards which open a world of discounts for the student along with other benefits such as proof of identity, age, and place of study. The customer has been in business for a period of two years and has data on about two million users who are either free or paid members of this card. The card is also available as a Mobile application based digital card removing the need for carrying a plastic card around.
The customer had a lot of data being collected on a day-on-day basis which includes but not limited to

• Student data
• Card sales data
• Discount and Offer redemption data.
• Customer help desk data
• Customer community interaction data

The team from Arrk group did an “EmbArrk” journey with the customer to understand sources of data and the type of information which they were interested in. The Data science team at Arrk studied the various sources of data and did a cross relation of the significant data bits which were required to build a meaningful model. Date engineering and integration specialists worked with the customers to fetch this information in a Redshift Data Warehouse. Here the data was collated into meaningful views which would act as a data feeder to the Recommender Engine model.

 

 

Once the data pipeline was estabished, the Data scientists on the team built a deep learning model using Python and Tensorflow. The model takes current offer adoption data along with demographic and purchase data as an input to train the model. The model trains of hundreds of thousands of such adoption data and infers the deep relations between various features. The created Tensor flow model was validated against test data and achived 70 % accuracy during trials. The Python model was served as a library to a Lambda function which was sitting behind an API Gateway. Devices and collaborating services made direct calls to this API end point with user data to get the best possible recommendations for the data. The model keeps learning every fifteen days on the latest adoption data to ensure that it is as accurate as possible.

Replacing an older static recommender with a Intellegent recommender has improved the customer offer adoption by approximately 30% as can be observed in the monthly offer reports. The models accuracy will improve over time as we also feed it the adoption levels of the recommended data. The DataScience team at Arrk can work with you to use your collated data to build such intellegent recommenders which will give features based recommendations and help improve your sales and profitability. Please reach out to one ouf our Data experts below to help you start up on this journey.