Natural Language Processing for Email Automation


One of the key differentiators of value in the retail sector, especially e-commerce, involves delivering high quality customer service expereinces.


Of all incoming email traffic accurately classified as simple or complex to solve


Reduction in customer waiting times by automating email responses


Efficiency savings generated per year


Our client is one of the fastest growing online retailers of bespoke furniture and home-ware in Europe. They began operations in 2010 and have grown from a small online retailer to an international company that has consistently achieved over 49% of growth in revenues year on year, employs over 489 people, has operations in 11 European countries and generates more than £173 million in revenues per year. The senior leadership team wanted to find a cost-effective way to build capacity into their customer services operations by automating routine tasks in after-sales services activity.


Our team began by analysing the customer services database, which contained over 17 million records, 46 thousand fields and 782 billion data points. The analyses involved a variety of feature extraction and feature engineering exercises to understand workflow activity and communication patterns in the database. By using a combination of text mining, classification algorithms, decision tree analysis and business rules the team came up with a multilayered solution that could accurately classify up to 53% of all incoming emails as complex to solve or simple to solve and formulate an automated responses.

press to zoom
press to zoom

All the emails classified as simple to solve were analysed further using Natural Language Processing (NLP) techniques and matching algorithms to map customer requests for information. This information was used to create email response templates. For example, when a customer sent an email asking when their order would be delivered, our solution would classify the customer request as simple to solve then analyse the nature of the request and automatically collect the relevant information from the customer service database. The solution would then insert the date into an email template for delivery date requests.


In less than 5 weeks our team of data scientists designed and delivered a solution that identified over a 100 categories of email communications that could be automated. This helped reduce customer waiting times by 39% and generated efficiency savings of £250 thousand per year. The solution also translated to better employee experiences and greater customer satisfaction ratings because the customer services team had more time to solve complex queries.

Note: Delivered in partnership with Pivigo Limited.