All cases
Restaurant holding · BălțiMoldova · 13 venues in one city

Google Reviews Loop

Reputation LoopGoogle ReviewsAI Systems

01

Challenge

Thirteen venues of a holding and benchmark competitors in one city, including a coffee chain with five branches. Google listings were opened occasionally, negative reviews sat for days, and the owner saw ratings only when someone sent a screenshot. What was needed was not a per-location service but an owned loop: reviews, analysis and an index per venue in one chat.

02

Solution

01
One loop for Google reviews: listings polled three times a day in slots, silence at night
02
Every review gets a translation, sentiment and topics; negatives arrive as a personal card, the rest goes into the digest
03
Daily, weekly and monthly digests with a loyalty index per venue, no holding-wide average
04
The coffee chain split into branches with separate cards and indexes: otherwise a weak branch hid behind the mean
05
The loop reports its own failure: if the data source is down, the owner sees the last successful poll date, not «no new reviews»
06
Running cost after rescheduling — about €4 a month for data

03

Deep dive

Lesson: silence and failure must look different

Two months in, the paid data source fell into an unpaid invoice and returned errors for nine days. The bot honestly reported «no new reviews», which was true of its database, not of the listings. The outage was indistinguishable from silence.

Fixed in code: if a whole polling round fails, an alert goes to the administrator, and the client digest gets a health block with the date of the last successful poll, without technical detail. The backlog was loaded in one pass: four hundred reviews over three weeks, seven negatives delivered personally, the rest as a single summary so the chat would not turn into four hundred cards.

The rule we took into every loop: any poller of a paid API must be able to announce its own death. «No data» and «source unavailable» must look different in a client report.

Loop economics

The data price is driven by the number of polls, not the number of useful reviews: a listing with no new reviews costs the same as one with them. Five polls a day across fifteen venues cost about $8 a month, three polls across thirteen about $4. The owner chose three slots: morning, afternoon, evening.

That is the argument for «owned implementation plus operation»: the loop belongs to the holding, and the monthly bill is data and a server, not a licence per venue.

04

Results

0
Venues in the loop
0
Listing polls a day
0
Reviews processed in 3 weeks
≈ €0
Running cost per month