Real time answers to customer queries
Instant automated responses to questions posed by potential buyers when shopping for furniture and home decor products on the website which usually took upto 8 hours
How we did it
Q&A Matching Engine
Providing semantic information of a search query
Natural Language Understanding
Mining relevant data and providing the required results
“A lot of our customers post questions related to the products we offer. Handling the sheer volume takes upto 8 hours to get them, verify details from our suppliers and then post the response. The solution MindWave gave us was fantastic, it gives instant accurate replies to over 80% of the questions and sends the rest to our agents so we have everything covered within the hour!”
The idea was to build a system to answer real time questions asked by the customers and the main challenge here was to interpret the buyer’s requirement. Another problem was to detect user intent correctly, in a single action or a series of user input actions on the basis of any background or contextual knowledge that may be available.
- Interpreting the buyer’s need regarding any product correctly no matter how it is expressed
- Using the product’s data to work out the right answer
- Integrating with existing interfaces to demonstrate a live product together to the team members
- Expressing the right answer in natural language to the buyer
- Fulfilling the buyer requirement in real-time
- Match user intent to the underlying content most effectively to surface the right recommendations solutions, from the searched content.
The solution was built systematically, based on the challenges listed. Using the core natural language generating and core natural language understanding engine helped in developing unique feature to answer all type of queries asked by the buyer.
- The solution helped to detect correctly the user intent no matter how it is expressed, in a single action or a series of user input actions on the basis of any background/contextual knowledge that may be available.
- It also helped to interpret the meaning of all the underlying searched content such as incident reports/calls no matter what form it is and how it is stored.
- The solution helped to match user intent to the underlying content effectively to surface the right insights, incidents, resolution recommendations, and products from within the searched content.
- Detect correctly the meaning of all the underlying searched content such as incident reports/calls no matter what form it is and how it is stored.
As a result, the core natural language generating and core natural language understanding engine helped in creating powerful and unique features to provide answers to the buyer based on product analysis. To further improve the solution, design and implementation of some new algorithms were also done.
Extensions of this solution were also built.
- For example, if a buyer asks ‘What is the width of the pillow case?’ and follows up with ‘And the length?’; Then the solution relates the 2nd question with the ‘pillow case’ context and returns the right answer.