What Are Review Predictions?
Written By Chad McGuire (Sparrow Intel)
Review Predictions are one of Sparrow Intel's most distinctive features. Before a guest ever leaves a public review, Sparrow uses Chirp AI to predict whether that guest is likely to leave you a good or bad review — and lets you trigger different automations based on that prediction.
This guide covers what predictions are, when they're generated, what data they contain, and how to use them to protect your reputation and proactively address issues before guests leave you a review.
Table of Contents
What is a Review Prediction?
A Review Prediction is an AI-generated forecast of how a guest is likely to feel about their stay - generated at the time of check-out before they leave a public review.
Behind the scenes, Sparrow Intel runs an AI model fine-tuned on the question:
"Is this guest likely to leave a good or bad review?"
To answer that, the model looks at the full context of the stay:
Reservation details from your PMS — property, dates, guest info, channel, etc.
The complete conversation history between you and the guest — this is the most important signal, because messages are where guests actually tell you what happened during the stay (issues raised, requests made, how they were resolved)
Operational context where available — turnover cleans, inspections, and tasks linked to the stay
The output is a prediction (positive, neutral, or negative) plus structured details about what drove that prediction — top positives, top negatives, and any cleaning or maintenance issues surfaced during the stay.
When Do Predictions Occur?
Predictions are generated automatically at the time of a guest’s check-out.
In practical terms, this means a prediction is available before the typical "please leave us a 5-star review" message would normally go out. That's the entire point: you want to know how the stay went before you ask the guest to broadcast it publicly.
What's in a Prediction
Each prediction is a structured record, not just a thumbs up or thumbs down. You can see all of this in the platform, and you can use any of it as conditions in your rules.
When a public review eventually arrives, Sparrow Intel ties it back to the prediction so you can see how often the prediction matched reality.
Why Predictions Exist
Most property managers send a templated "please leave us a 5-star review" message at the end of every stay — usually fired from their PMS or a messaging tool. The problem is well known:
The guest dumps a complaint into messages, and before anyone has time to respond, an automated "Hey! Go leave us a 5-star review!" fires off. Now you've made things worse.
The standard workaround is to remember to manually turn that automation off for problem stays. In practice, no one remembers — and at scale, no one can remember.
Review Predictions exist to make this decision automatically, with full awareness of what actually happened during the stay. Instead of one templated message going to every guest, you can branch:
Positive prediction → send the review-request message
Negative prediction → do something different (or do nothing at all)
That branching is what protects your public rating.
Common Use Cases
Predictions are an entity in Sparrow's rule engine, so anything you can do with rules — send messages, create tasks, send webhooks, notify a teammate — you can trigger off a prediction. A few of the most common patterns:
1. Only ask happy guests for reviews
The most common configuration. A rule along the lines of:
When the prediction sentiment is positive, send a templated message say 24 hours after checkout asking the guest to leave a review.
The negative-prediction guests simply don't get asked. Your review request volume goes down slightly; the percentage of those requests that turn into 5-star reviews goes up substantially.
2. Privately re-engage unhappy guests
Pair the rule above with its inverse. When a prediction is negative, instead of asking for a public review, give the guest somewhere private to vent. Common implementations:
Send a different message acknowledging that things may not have gone perfectly and inviting feedback directly
Create a task in your PMS for a teammate to do a personalized outreach
Send an email to the guest experience manager so they can reach out personally
3. Book a call for high-value or seriously unhappy guests
One of our customers does this and it's worth highlighting. When a prediction comes back strongly negative, they send the guest a message that acknowledges the stay didn't run up to par and includes a Calendly link to book a 15-minute call to share feedback.
It sounds aggressive on paper. In practice, guests who feel heard rarely leave a public review at all — and the ones who do often soften their tone.
4. Proactively triage cleaning and maintenance issues
Predictions don't just predict sentiment — they extract concrete cleaning and maintenance issues from the stay context. You can route these straight into your task management system or a webhook before the next guest checks in.
For example: a guest mentions in messages that the towel bar is loose. Sparrow surfaces that as a maintenance issue on the prediction, and a rule can turn it into a Breezeway work order automatically. The next guest never sees it.
See the Task Creation Guide for the full mechanics of how issue-based tasks get created.
FAQ
How accurate are predictions? Accuracy depends heavily on how much guest interaction your stays actually have. Stays with rich conversation history (issues raised, requests made, back-and-forth) produce the most accurate predictions. Stays with no messages at all are harder to predict because there's less signal to work with.
Will the guest ever see the prediction? No. Predictions are internal to your team. Guests only ever see the messages, tasks, or replies that your rules generate as a result of the prediction.
Can I review predictions manually before any automation fires? Yes. You can choose not to attach any Rules to predictions and instead use them purely as a queue for your team to work through. Many customers start this way before turning on automations.