Finance based large scale company is collecting millions of form which are handwritten. They want to extract this information in digital records, So that the required analytics can be done.

• Poor Handwriting

• Variety of forms to extract information

• Overlapping data in different fields

• Document scan quality

• Accuracy

With properly trained for specific document formats, Deep Neural Network technology – including categories like Convolutional Networks, Recurrent Networks, Attention Networks, Differentiable Networks, etc. – we can reach up to 99% accuracy at field level

• Information digitization

• Human intervention minimized, and resource costs saved

• Self learning solution

• High accuracy and reliability

A global travel retailer sells product lines from various big fashion brands, and wants to select a small but high potential customer base for running targeted marketing campaigns for specific product lines. Thus, it becomes a problem of searching for a needle in a haystack to find the right customer for the different brands and product lines.

• Huge customer base of 5-6 million.

• Marketing constraints of 40k emails allow for only a fraction of recipient selection.

• After accounting for BAU and control group, we can only select about 10k users for a fashion brand campaign.

• The customer base is also quite niche.

The solution used an ensemble of temporal buying behaviour and deep neural network models to model the propensity of the entire customer base to buy the product line/ brand. Top 10k users with highest propensity were selected for campaign.

• Replicable campaign solution was created for the client

• Top 10k users showed a higher lift and gain across validation and live tests

• A/B test confirmed improvement over BAU methods of mass marketing

• 40% improvement in engagement over our control set was observed over multiple campaigns