How one of Sweden's largest clothing companies improved incoming customer service and received a 4.85/5 rating.


Background

The e-commerce team at Scorett made the decision to test our chatbot solution to reduce incoming customer service cases and simplify the journey for customers to their outlet. As they previously only handled customer service via email, we opened a new channel to handle customer service for Scorett.

The goal

The goal consisted of two points:

  1. To provide fast responses to customers.

  2. Simplify the process of complicated inquiries for the customers.

Solution

1. Train Claude Haiku on data from Scorett

By training Claude Haiku (LLM from Anthropic) on data from Scorett, their chatbot became capable of broadening its knowledge and answering questions within its area of expertise.

We chose to use Claude Haiku instead of GPT-3.5 from OpenAI, because the relatively new model Haiku gives better results than GPT-3.5. We can see this in the picture on the right.

Untitled

2. Automatic language detection

Since Scorett has a large Swedish customer base, we chose to use Swedish as the default language and include active English language identification to also reach customers who only speak English.

Untitled

3. Product recommendation system

Customers who are unsure about which product to choose were given the opportunity to easily send a question through our product recommendation system.

Information we collected from the customer was:

This collected data is saved in each dashboard conversation and sent directly to Scorett's customer service team.

Through this function, visitors to the e-commerce were given the opportunity to quickly and easily submit a question about a product recommendation and then receive an answer via e-mail.

Untitled