Reviews Across Ten Locations: Why Manual Monitoring Doesn't Scale

The three-location threshold

Feedback is the one process in a chain that usually has no owner. The kitchen has a chef, sales have a manager, finance has an accountant. Reviews have nobody: the assumption is that «we keep an eye on it».

In practice, «we keep an eye on it» means whoever remembered is looking. Some public listings have not been opened in months. A one-star review sits for a week and works as advertising against you the entire time.

The second gap is distortion. The owner hears about the complaint through the manager's retelling, and the manager has every incentive to soften it. The primary source — the guest's own text — never reaches the top.

The cost of silence

A location's public listing is the storefront a guest sees before deciding. The rating and the latest reviews act as the only reference about you that a person reads before visiting or ordering.

Unanswered negative feedback costs twice. Once when the next guest reads it and picks the place next door. Again when the author receives no reaction and locks in the impression: nobody here cares.

A separate line of loss is delivery platform ratings. A storefront with a slipping score loses position in the aggregator's listing, and that shows up in revenue before anyone notices the rating itself.

What should happen to a review: the chain of events

A working system is not a «new review» notification — it is the full chain from the rating appearing to a management conclusion.

The review appears on the location's listing. The system pulls it with its original text and language, resolves the location and country, and filters duplicates. A model labels sentiment, topic and urgency, and translates the text into the owner's language.

Then a draft reply is prepared — in the guest's language, following the brand's approved scripts, anchored in the specifics of the review. The owner sees a card in the messenger and either publishes with one button or corrects it in a single message.

And the last link, the one most often skipped: the guest is moved into a private channel, while topics and ratings accumulate into per-location statistics. Without that link the system is an autoresponder, not a management tool.

Reply language is not a detail

For chains operating across countries, a review reply is a public statement by the brand in the region's language. A reply in mixed languages or machine translation reads as carelessness, and for a premium brand that is worse than silence.

Second rule: reply in the language of the review, not the language of the owner. The guest wrote in Romanian — the reply is in Romanian. The owner should see the text in their own language, together with the translation and the topic.

Third: reply quality is not where you economize on the model. The reply reads as the face of the brand, so that step deserves a strong model, while utility work — filtering and classification — stays on a cheap one.

The private channel: resolve before publication

Most complaints are solved in one message: the wrong item, a slow delivery, missing cutlery. The problem is that the guest has no address other than the public listing — which is how a small thing turns into a rating.

A private channel flips the order: every reply carries an invitation to write directly, and this guest's next complaint arrives as a message. The owner answers, and the answer reaches the guest through the same channel.

A practical detail: the channel has to remind you about itself. If a message sits unanswered for more than a couple of hours, the system should nudge — otherwise the private channel becomes one more place nobody opens.

How to measure: a loyalty index per location

The chain's average rating is a useless metric: it averages a strong location with a failing one and shows no trend. Management meaning appears at the location level.

A workable version is an NPS-style index: the share of top ratings minus the share of low ones, scaled to 0–100, with a month-over-month delta. When ratings are few, «not enough data» is more honest than false precision.

The second layer is topics. Once every review is labeled by topic (kitchen, speed, service, delivery, price, cleanliness), you can tell a systemic issue from one bad evening.

Seven questions before buying a tool

1. Which platforms are covered besides Google — and do they include the local delivery aggregators in your market?

2. What language is the reply written in: the review's language or the template's?

3. Does the system know your scripts, per-location phone numbers and forbidden topics — or does it answer in generalities?

4. Where does the notification land: in the messenger you already live in, or in a dashboard you have to log into?

5. Is there a private channel with the guest — or does all communication stay public?

6. How does cost grow when a new location opens: linearly per subscription, or inside one implementation?

7. Who owns the data and the conversations if you end the relationship tomorrow?

If most answers are «no» or «depends on the plan», you are buying notifications, not a system. How ours is built: mrktr.pro/en/reputation-loop/