Automating Customer Service With AI

Case Studies
November 14, 2020

Purpose is a furniture and homewares retailer with the vision of making high-end lifestyle design accessible to everyone, everywhere. They accomplish this vision through a unique business model which consists of an online-only retail model and by working directly with designers to bring high-end designs to customers without the middlemen.

It is important for MADE to get online customer service right as MADE operates a largely online first model. The project team were tasked with exploring the potential of automating parts of the customer service process at MADE. In particular, the team looked to see if the first response could be automated by categorising the inbound emails.


To address this, the team was given access to both structured and unstructured data. Unstructured data was given via access to six months of emails in the customer service database; equating to a total of 17.4 million data rows and 46K observations. Structured data was provided via access to details of the actual orders corresponding to the emails in the database. Within these emails, the team found over 100 categories of email types where no actions could be taken immediately. For these emails the team looked to automate the part of the process where a holding email could be sent, or an email telling the customer the likely next action.

The data were labelled with more than 100 categories (with the reason for contacting customer services), and grouped into relevant categories. They categories with potential to be automated were then prioritised. The team processed the content of customers’ queries/emails using various NLP techniques. In addition, they extracted other features from the text, for example the sentiment of the emails and the number of punctuations, as well as features related to order and dispatch information.

A wide range of modelling algorithms including logistic regression, naïve Bayes, random forest, support vector machine etc. were tested, and the final model chosen based on metrics such as precision, recall and interpretability. This culminated in a 2-step model that firstly predicts the ease to automate, and then classifies the content of the queries into different categories.

Social Housing Rent Arrears Prediction with AI
Iste aut non sunt. Ut cumque veniam consequatur fugiat asperiores
6:50 pm

Iste aut non sunt. Ut cumque veniam consequatur fugiat asperiores

Asperiores mollitia eligendi harum voluptatem. Soluta aut et nisi. Non est quia aut. Sed sequi maiores hic vero. Excepturi odit s


Monetary savings in terms of reduced working load for customer service agents would lead to an estimated reduction in customer waiting time of around 40%, increasing customer satisfaction.

Furthermore, there can be savings on people costs including salary, national insurance, and pension payments, as well as recruitment, training and office space costs. This work also led to benefits of speed by avoiding multiple contacts that customers make.

The combined savings associated with this is estimated between £200k to £250k per year alone in the UK. With additional benefits from rolling out across multiple countries, with compounded savings given the fast rate of growth of, this saving could grow to £1m over the next few years.

The increased customer satisfaction following from the proposed intervention would also increase sales and customer retention even further.

Bulat Yapparov

“Insight into our communication with customers created a lot of traction in the company and we are now taking steps to meaningfully improve our customer service.”

Free Consultation

Contact us today to get in touch and discover what AI can do for your business.

Hassle free
Confidential advice
No obligations

Related AI Articles

What Results Can You Expect From Our AI (Artificial Intelligence) Rent Arrears Social Housing Solution?

April 7, 2022

3 Themes For Digital Transformation In Social Housing

February 27, 2022

A New KPI For Social Housing

February 27, 2022