By: Lara Hoyem, SVP, Data Insights and Rachel Boim, VP, Data Strategy & Analytics
Do you have the resources to read every comment from every customer from every delivery service provider for every off-premise brand at every one of your locations?
Vast amounts of raw data from customer feedback can be overwhelming and hard to organize. It’s tempting to simply ignore this data, however research reveals that over half of internet users inside the US leave an online review at least monthly. This data represents a vast treasure trove of data about your brand, specifics about what customers like (or don’t like), and the overall customer experience.
While customer ratings provide general sentiment, customer comments are an incredible resource to direct action. That’s why the Nextbite data science team created a Natural Language Processing (NLP) tool to wade through the clamor of online reviews for over 3,500 concept locations. I am excited to share two main features of this model: topics and red flags.
Topics:
Our Data Engineering and Data Strategy & Insights teams combined forces to map the disparate reviews and rating data from numerous Delivery Service Provider (DSP) sources into one consistent automatically refreshed data set. The data science team’s NLP model scans this pool of comments and identifies recurring topics. The model is trained at a brand level and then further validated at a location level. Location topics can then be compared to brand topics to highlight location-specific issues versus overall brand issues. (For locations with limited data that would prevent location-level modeling, comments are scanned for the brand topics.) Topics are ranked and occurrences are tracked by various groupings (time, brand, region, franchise, location, item, etc.). For instance, a high occurrence of the word “soggy” for a chicken sandwich led to testing different types of buns. “Missing” for sauces prompted process and training material updates. “No meat” and “tiny wings” identified locations that were not following ingredient quality standards. Topic frequency is tracked over time to test the efficacy of solutions and ensure location-level compliance with processes.
Red Flags:
The maxim “Any publicity is good publicity” does not apply to food. A viral tweet about food poisoning from a spoiled burger or a TikTok video exposing hair in a mac n’ cheese bite can kill a brand. You want to identify these red flag comments and address them immediately. Comments are continually scanned for red flag phrases and our customer service team is immediately alerted, enabling prompt and proactive damage control. Red flag analysis should include sentiment analysis for brand-specific tweaking. As we discovered, “sick” could be a red flag or a huge compliment: “Never ordering again – this burger made me sick!” versus “This burger is sick – I’m on my third order this week!”
Interested in learning how to use platform, proprietary and third-party data to maximize your off- and on-premises revenue?
Join us at our inaugural Data Science Summit for Digital First Brands this March in Denver, where we’ll share more of our team’s great work using NLP and other machine learning models.
Lara Hoyem, SVP, Data Insights
With nearly 25 years experience working at celebrated online consumer brands, Lara has helped grow digital businesses from start-ups to publicly traded companies. As Nextbite’s SVP, Data Insights, she is leading the company in its next phase of growth in the exploding delivery-optimized virtual restaurant category. Prior to joining Nextbite, she held key marketing and product leadership roles spanning ecommerce, subscription services, and ad-supported content publishing for companies including Study.com and Shutterfly, where she was most recently Vice President and General Manager, Photo Books, Calendars and Print. A consumer brand expert, Lara is passionate about working cross-functionally, using data for insights, developing sustainable differentiation for brands, and building deep customer relationships.