Most hospitality organizations were affected by the pandemic last year. Their bookings had drastically fallen and there was hardly any respite. A well-known hospitality chain of Europe planned to strategically restart the business using the feedback from their loyal customers.
Additionally, the client also wanted to employ a better method to gather feedback data from their customers. The chain usually sends a feedback form to every customer after their experience at the hotel / somewhere else. They usually had a Net Promoter Score of eight out of ten, but the bookings did not match the positive score.
Most of their bookings mainly happened on their website that displayed ratings and reviews for each of their suites and properties. The client also had a good social media presence on leading sites, but there were inefficiencies in handling complaints or feedback obtained from these sites.
The client wanted to tackle the above problems using their industry experience as well as technological expertise that could help them utilize current data, gather more relevant data from their customers, and translate that into effective business strategies.
To summarize, the following were the objectives of the client:
- Employ better methods to collect relevant actionable data from customers
- Utilize the collected data to devise a model that could help them recoup their business
Sentiment Analysis Using Text Cognition
We observed that the client had a good amount of sources to collect and gather data from customers. They specifically received customer feedback from three main sources:
- Survey forms sent out to customers after their stay
- Website reviews obtained
- Brand mentions on social media
The surveys initially being used only asked the customer to score a feature/offering /experience on a scale of 1 to 10. There was no scope for extrapolating more detailed feedback on every aspect. We added some more detailed questions to dig deep to understand why a customer gave a particular score.
rom three main sources:
Secondly, though the client had a good amount of website reviews, there was no strategic implementation of this feedback. Customers also shared their experiences and complaints on social media, but the handling of these was not up to the mark.
With such a thorough analysis, we proposed to build a sentiment analysis model that could help the client understand the net overall experience of the customers as well as to detect what customers are saying about their brand on social media on an hourly/daily basis.
We went about building this model with two distinct phases:
- Phase 1: To analyze hourly website reviews and customer feedback obtained from surveys, and use it to improve offerings
- Phase 2: A real-time sentiment analysis model that could detect what people are saying about the brand, classify the issue, and rate it as a net positive or negative experience.
We built the sentiment analysis model using TensorFlow because of its rapid iteration and rich library resources. To add to that, the client was familiar with the tool and was already using it to build other applications.
Analyzing Website Reviews and Surveys
The model analyzed the feedback obtained from website reviews and customer surveys and detected what were the words that were used the most by the customers and in which sense. The model classified the issue/feature and assigned a customer experience score between -1 to +1 to every feedback obtained.
From a given review, the model could identify what was the underlying issue/feature. We had trained the model to classify the following features:
- Website experience
- Stay Experience
- Staff behavior
- Appliance issue
A score of -1 indicated extremely negative feedback, a score of +1 indicated extremely positive feedback with 0 acting as a neutral experience.
This analysis showed the most used words in the feedback with the associated customer experience score. The combined score gave a quantifiable idea to the client to decide which features needed to be improved, what offerings should be added, and how to improve the website experience.
Real-Time Sentiment Analysis on Social Media
The model continuously analyzed every textual post/tweet to detect where the brand was being mentioned in what context. The same features/issues were used as in the review analyzer, but the experience score was tweaked a little.
The model was mainly used to track customer complaints on social media and assign a support score to it. Every support score contained the underlying sentiment of the feedback and the priority of the issue. The score ranged from 0 to +1, with 0 being the least priority and +1 being the highest priority.
The score helped the support team to quickly identify the major complaints and service them. This feature helped the client revamp their customer support experience by organizing the complaints and service them in a well-defined manner.
Upon successful completion of the project, we delivered our flagship UX workshop to the client’s engineers so that they could independently manage and govern the model. We also offered them our remote MLOps support for any high-priority issue that could affect the model.