Earlier, Quibble released a pricing model incorporating computer vision. Next came a Natural Language Processing model built for hospitality, designed to interpret consumer sentiment from guest reviews and feed it into the pricing engine.
What is NLP?
Natural Language Processing enables computers to understand, interpret, and generate human text or speech — tasks like text recognition, translation, sentiment analysis, and generation. ChatGPT is its most familiar face; in marketing, it powers sentiment analysis, chatbots, and content optimization.
Why build, not buy
Off-the-shelf NLP saves time, but Quibble’s requirements were specific: the output had to feed directly into the pricing engine with high precision and consistency, so the team built and trained its own model. The hardest part wasn’t training but data — scraping review text from OTAs and then manually labeling sentiment across thousands of reviews.
How reviews impact choice
Quibble’s choice model already used review scores, review counts, and image-quality ratings. The team noticed properties with stable booking histories suddenly going quiet despite correct pricing and availability — the cause was almost always a recent negative review. Guests typically read the most recent three to five reviews, so a single recent negative one hits purchase decisions hard in the short term.
Results and limitations
Testing showed positive reviews don’t meaningfully increase bookings, so the model’s main value is mitigating the impact of negative ones. The goal isn’t to discount after a bad review gets pushed down — it’s to cushion the effect and then return rates to where they were. Focusing on the five most recent reviews also lets Quibble track competitors’ reviews and adjust when a competitor’s negative feedback opens up market share.