Every Enterprise Has a Dashboard of What Happened. None Has a Model of Why.
Dashboards rank problems by how often customers mention them, and that is usually the wrong order to fix them in. At 33 gyms, the loudest complaint did the least damage.
Most leadership teams I meet run their customers off a ranked list. Every Monday a dashboard shows the ten complaints customers raised most, and the money and the meetings go to the top of it. That order feels objective, because it is a count. It is usually the wrong order to fix things in.
A count tells you how often customers mention a problem. It does not tell you what the problem does to them. Those are different quantities, and when we measured both this year in public reviews of a gym chain, they came out in nearly opposite orders.
What did 9,186 gym reviews say was worth fixing?
We read 9,186 public Google reviews of 33 Planet Fitness clubs in metro Chicago, written between January 2025 and July 2026, and turned each one into 28 structured fields. Planet Fitness is not a customer of ours; this was built entirely from public data.
Ranked the way a dashboard ranks them, broken equipment is the top complaint, named in 122 reviews. Front desk and staff problems come second, in 107.
Ranked by effect, the list turns over. A rude or absent front desk raises the chance that a review is one or two stars by 36 points. Broken equipment raises it by 5, the smallest effect of the six failures we tested. The loudest complaint is the least costly one.
From the report: Planet Fitness: a causal analysis
Both estimates hold the club, the period, the reviewer type and the review length constant. The front desk effect has a 95% range of 26 to 52 points; equipment's is 2 to 9. The bottom of the front desk range is nearly three times the top of the equipment range. So the order holds even at the edges of the ranges, and every effect was checked three ways against a false cause and held.
The report notes that equipment is what the company's public capital debate is about: the replacement cycle, the remodels, the recovery amenities. The front desk, in the report's words, is close to free to fix. A count-ranked dashboard sends the capital to the machines.
Is this just a gym problem?
No. In public reviews of 45 Orlando hotels, front desk and check-in was the dashboard's number-two priority, and its effect on a one- or two-star review was about two points and not statistically significant. Which complaint leads changes from one business to the next, and the hotel, gym and cafe results sit side by side in which complaints cause bad reviews.
Why do dashboards rank by count?
Because a count is the only ranking you can produce without an outcome and without a model. Tag the comments, tally the tags, sort. Any tool that tags text can do that on the first day, so the tally becomes the priority list by default, without anyone deciding it should be.
The NBA's published story describes where most companies start. Before the league moved its fan survey analysis to us, high-level metrics like NPS and OSAT lacked contextual drivers, and analysis skewed toward negative sentiment. A score says fans are less happy and a tally says what they talked about; neither says which part of the experience moved the score.
Why the two orders split, from the things every customer touches to the minor issue that rides along in an angry review, is worked through in correlation vs causation in customer experience data. For a CEO the consequence matters more than the mechanism: the list that is cheapest to produce is the one most likely to send capital to the wrong place.
Ranked by effect on what?
A ranking by effect needs an outcome, and the outcome you pick decides the list. This is where most attempts to fix the dashboard stop too early.
In the same gym reviews, the front desk is the largest cause of an angry review, and billing is third. Ask which members say they are leaving and billing comes first: half of the reviews that raise a billing or cancellation problem say the member will cancel, not come back, or warn others, against 13% for the front desk and about 1.4% across all reviews.
Same reviews, second outcome
Ranked by talk of leaving, billing comes first, at 50%
Share of members raising each issue who say they will cancel, not return, or warn others. Fleet average about 1.4%. These are rates of what reviews say, not causal estimates.
Those leaving rates describe what reviews say. Whether billing actually causes cancellations is a question for the cancellation records, and only the chain holds those. That is the useful edge of public data: it tells you which second list to build from your own. Why gym members say they will cancel reads the billing reviews themselves.
What would a model of why need?
A dashboard holds one layer of the picture: what customers said, counted. To rank causes, you need what they said joined to what they did next, to what the business changed and when, and to what was happening around them: the weather, the calendar, a competitor opening down the street. We call that a customer world model. The gym analysis above uses the first layer plus controls. A company running it on its own data can go further, because it holds the cancellations, the billing disputes and the staffing rosters that public reviews only hint at.
The gym analysis has a limit worth stating. It is observational review text, and its ranking comes from causes that survived refutation tests rather than from a randomized experiment. That is still a far better guide to spending than a count, and it is checkable: fix the front desk at a few clubs, leave others as they are, and the next quarter of reviews will tell you whether the effect was real.
For the statistics behind this, key driver analysis from text runs the same six gym drivers through five methods and shows how far their rankings disagree.
What should you ask for before funding the next fix?
One thing: the list ranked by effect, with a range on every number, on the outcome you actually care about. If the top item by effect is not the top item by count, you have found where the next budget should go. If the ranges overlap, you have found out that "first" is not settled yet, which is worth knowing before you commit a budget to it.
Then ask what else changed in the period. A price change, a new policy, a cold month or a competitor opening can move the same numbers, and a ranking that ignores them will credit or blame the wrong thing. Testing a business decision before you make it shows that step on a worked example.
Where does Dimension Labs come in?
Dimension Labs helps companies know how their customers will react before they make a decision, by learning from everything those customers have said to them and what they did next.
The platform reads every ticket, chat, call, review and survey response, enriches each one with Dimensions built for your business, and joins them to your orders, renewals and locations. On the Prove plan and up it estimates each driver's effect with a confidence interval, which is the ranking this essay asks for. The agent then builds the report or deck from it, so the ranked-by-effect list reaches the people who fund the fixes in the same form every month.
FAQ
What is the difference between a dashboard and a model of why?
A dashboard reports what happened, such as which complaints came up most. A model of why estimates what each factor did to an outcome, holding the other factors constant, and gives a range for each estimate.
Why is the most common complaint often not the most important one?
Because frequency mostly reflects how many customers touch something, and how much it hurts them is a separate measurement. In 9,186 gym reviews, broken equipment was the most-mentioned failure and had the smallest effect on a bad review of the six tested.
What data do I need to rank complaints by effect?
Customer comments with an outcome attached to each one, such as a star rating, a CSAT score or a renewal, plus the location or account and the date. More outcomes and more context give a better ranking.
Upload one decision's worth of customer data, a quarter of reviews or tickets with ratings and locations, and see which complaint actually moves your rating: start with one file. Plans start at $299 a month with a 14-day free trial; effects with confidence intervals are on Prove and up.