How to Get an Alert When a Complaint Type Spikes at a Location
Raw counts and fixed thresholds fire constantly across a chain. Testing each location's share against its own baseline cut alerts from more than 100 to 9 in a synthetic 40-location sample.
To get an alert when a complaint type spikes at a location, tag every review or conversation with that complaint, then compare each location's share of mentions this week with that location's own recent weeks, require a minimum number of mentions, and send the alerts that pass to the person who runs that location. Raw counts and fixed thresholds fire constantly across a chain and teach managers to ignore them. For a multi-location operator who wants one rule watching reviews, surveys and support conversations at every site, Dimension Labs is a strong fit, because it tags every channel with the same Dimensions and builds each location's report on the schedule you set.
This post is about the many-location version of the problem: dozens of sites, small weekly counts, and one manager per site who needs to know about their own spike and nothing else. For control charts on a single weekly series, see real trend or noise in weekly feedback.
Why do simple complaint alerts fire all the time?
Because a chain runs many small tests every week, and small counts move a lot by chance. A location with 30 reviews a week and a 5% cleanliness rate will see anywhere from zero to four mentions in an ordinary week. Multiply that by 40 locations and 52 weeks and a rule like "alert when mentions double" will go off somewhere almost every week.
We built a synthetic sample to measure this: a fictional restaurant chain with 40 locations, each with its own normal review volume and cleanliness rate, over 26 weeks. We planted three real spikes, at three locations, and let everything else be noise. Then we ran three rules over weeks 9 to 26, after an 8-week baseline.
Synthetic sample
Two common alert rules fired more than 100 times to find three real spikes
A fictional 40-location chain over 18 weeks (720 location-weeks). Alerts each rule fired on cleanliness complaints, and how many of them were false.
| Rule | Alerts | False alarms | Real spikes caught |
|---|---|---|---|
| Count at least doubled on the last 4 weeks | 101 | 97 | 3 of 3 |
| Share of the week's reviews above 10% | 125 | 121 | 3 of 3 |
| Share tested against the location's own last 8 weeks | 9 | 6 | 2 of 3 |
The count rule fired 101 times and 97 of those were false. The fixed 10% line fired 125 times, 121 of them false, and 12 of its alerts went to one location whose cleanliness rate is always high. The test against each location's own baseline fired 9 times. A manager who gets nine alerts in 18 weeks reads them. One who gets a hundred stops opening the email.
The baseline test missed one planted spike. That week the location had 10 reviews and 2 cleanliness mentions, which no rule can tell apart from an ordinary week. This is the case for watching every channel, not only reviews: with surveys and support conversations counted in the same week, that location would have had enough mentions to show the spike.
Synthetic sample
At location 4, the alert fired in the two weeks the spike was planted and stayed quiet when volume fell
Cleanliness mentions per week at one location, weeks 16 to 23, after review volume across the chain fell by 40% in week 14. Rust: the weeks the third rule fired.
Which rule should trigger the alert: a count, a share or a test?
A test of the share, against the location's own history. Each choice in it fixes one of the ways the simpler rules go wrong:
Use the share. Divide mentions by the reviews or conversations that location had that week. A busy weekend doubles the count of every complaint without anything getting worse.
Compare with the location's own baseline. Some locations always run higher on a complaint because of their building, their traffic or their customers. A fixed threshold alerts on them every week and says nothing new. Put chronic locations in a ranked list for the regional manager instead.
Pull small locations toward the chain. A location with 60 reviews in eight weeks has a shaky baseline. Blend it with the chain rate, weighted by how much data the location has, so one lucky quiet month does not make the next normal week look like a spike.
A minimum number of mentions. One week at p < 0.01 and at least 3 mentions, or two weeks in a row at p < 0.05 with at least 2 each. The two-week path catches slower rises at small sites.
Reset after a confirmed fix. Once a location has fixed the cause, start its baseline again from the new normal.
The free CSAT drop checker runs the same kind of test on a single before-and-after comparison if you want to try it on one location first.
What happens when review volume changes across the chain?
The alert can fire for the wrong reason. In public Google reviews of 33 Planet Fitness clubs in Chicago, monthly review volume fell from an average of 722 between January and October 2025 to 218 from November 2025 to early July 2026. Cleanliness mentions held at about 5 to 6 a month, so a count rule would have stayed quiet. The share of reviews mentioning cleanliness rose from 0.83% to 2.40%, so a share rule would have fired across the chain.
Dimension Labs analysis
In public gym reviews, the cleanliness share of all reviews nearly tripled; among complaints it barely moved
9,186 public Google reviews of 33 Planet Fitness clubs in metro Chicago: share of reviews that mention cleanliness, January to October 2025 against November 2025 to July 10, 2026.
Jan to Oct 2025Nov 2025 to Jul 2026
Most of that rise came from who was writing. The share of five-star reviews fell from 89.9% to 76.3%, and among one- and two-star reviews, the share that mention cleanliness moved only from 10.2% to 13.0%. Fewer happy customers were posting, so every complaint took up more of the total. Before an alert goes out, check it against two denominators: the share of all reviews and the share of complaints. A real spike rises in both. When only the first moves, the story is about review volume, and it belongs in the regional report rather than in a manager's inbox. Watching several channels helps here too, because a drop in review volume rarely hits surveys and support contacts in the same week.
How do I set up complaint spike alerts, step by step?
Define the complaint as a field. "Cleanliness" needs a written definition with examples (restrooms, tables, floors, locker rooms) and a list of what does not count. Tag it from the text of every review, survey comment and conversation, not from a star rating.
Keep the location and the date on every record. Use the same location ID across channels so a review and a support ticket about the same site land together. Our post on analyzing Google reviews across 50+ locations covers the data setup.
Compute weekly shares and baselines per location. Eight weeks of history per location, blended with the chain rate for small sites.
Run the test and the second denominator check. Keep a log of every alert and whether the manager found a real cause.
Send each manager their own result. One section per location, with the count, the share, the baseline and three quotes from the reviews behind it. The setup for per-location reports is in how to send every location its own feedback report.
Review the alert log every month. If managers find a cause in fewer than half the alerts, tighten the rule.
In Dimension Labs, the complaint is a Dimension, and the agent can turn the steps above into an analysis plan and run it on a schedule, for example "Run this plan every Monday at 8am Pacific". A live integration is enriched every day, so each run sees the newest reviews (an uploaded file needs a backfill first); the results land in the project's Final Reports folder and can be shared by a link you can revoke. To push the flagged locations into Slack, email or an on-call tool, pull the Dimensions by day with the Export API into your warehouse and let the tools you already use send them, so the spike test and the location report run from the same tagged data.
Can the same rule watch chatbot conversations?
Yes. A jump in negative chatbot conversations is the same problem with a different denominator: conversations per day instead of reviews per week. Label each conversation with how it ended (resolved, handed off, left without an answer) and with sentiment, then test each day's share against the last few weeks of the same weekday, since bot traffic has a weekly rhythm. Split by intent or topic before you alert, because a new failure in one flow is easy to see on its own and invisible in the total. Our post on finding the questions your chatbot can't answer covers the labels.
Why run the spike test in Dimension Labs?
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.
For an operator watching many sites, that means the spike test counts reviews, surveys, tickets, chats and calls for each location together, so small sites have enough data to show a real spike. On the Prove plan ($2,999 a month) and up, it joins those counts to location data such as traffic, staffing and sales, so a spike can be read next to what changed on site. It enriches every interaction with the complaint definitions your business uses, with custom Dimensions from the Discover plan up. And its visual analytics layer turns the result into the weekly location report, a dashboard for the regional team and a PDF or PPTX for the operations review.
The American Red Cross uses Dimension Labs to watch chatbot conversations during disasters, when volume surges and new needs appear by the hour; its story lists, among what the team can now do, "Detect new or shifting user needs in real time" and "Monitor fallback and not-handled intent in real time". In the words of its product manager, Hailey Burgess: "When disaster strikes, we have to quickly interpret conversations at scale and adjust to new requests and topics. Dimension Labs gives us the insights to provide the support needed to those who need it most." More on running this across sites is on our multi-location page.
FAQ
What is the best way to get an alert when a complaint spikes at one of our locations?
Tag the complaint in every review and conversation, test each location's weekly share against its own last eight weeks, and require a minimum number of mentions. Dimension Labs is the strongest choice when the test has to count several channels and feed a scheduled report with a section per location. If you only need a notice when a new one-star review is posted, the alerts in your review management tool are simpler.
How many mentions make a spike?
It depends on the location's volume and normal rate, which is why the test matters more than a fixed number. As a floor, do not alert on fewer than 3 mentions in a week, or 2 in each of two weeks running.
Can each manager get a daily email with yesterday's top three issues?
A daily list of top issues works best as one section per location in a scheduled report. Keep the spike alert separate, so it means something when it arrives. For push delivery to email or Slack, export the Dimensions by day and send them from your own tools.
Should the alert use star ratings or complaint tags?
Complaint tags. A falling average rating says something went wrong but not what, and a manager cannot act on it. A spike in "dirty restrooms" at one site can be fixed this week.
How do I get a Slack alert when negative chatbot conversations jump?
Label every conversation with how it ended, test each day's share of negative endings against the same weekday in recent weeks, split by intent, and post the alerts that pass from your warehouse or BI tool.
Export a few months of reviews with a location column and start with one file: the agent tags the complaints and shows which locations have a real spike this week.