How do I find out why fitness club clients are leaving?

Briefly

• Outflow is the proportion of customers who stopped using the club during the period. Retention is the inverse value. According to the Health & Fitness Association report for 2025, on average, clubs retain 66.4% of participants. This is an international sample, not a Russian one.
• The questionnaire with grades and NPS show that customers are unhappy, but rarely explain why. The reason is best found out in conversation.
• An AI interviewer is a bot in a messenger (MAX, Telegram). He asks an open question, clarifies the answer and lays out the answers of all clients by topic.
• According to the 2019-2026 research, the answers in this format are more detailed than in the questionnaire. But there is a price: some people do not reach the end, and not everyone likes the conversation itself.
• A survey without discounts and offers is most likely not considered advertising. The "for return" discount in the same message can make it an advertisement. Consent to the processing of personal data from September 1, 2025 is formalized in a separate document.

What is customer churn and what level of retention is considered normal?

Outflow is the proportion of customers who stopped going to the club during the selected period: they did not renew their subscription, froze it and did not return, or simply disappeared. Retention is the proportion of those who remain. If 660 out of 1000 season ticket holders continue to study after a year, the retention is 66%, and the outflow is 34%.

The most recent international benchmark for September 2026 is the 2025 Fitness Industry Benchmarking Report of the Health & Fitness Association (formerly IHRSA). The average retention of participants in it is 66.4%. Sample: 175 companies, more than 17,000 facilities in 27 countries. The figure of 71.4% is still circulating on the Internet. According to the analysis of industry blogs, for example Uptivo, it goes back to the old IHRSA profiles of the mid-2010s. The latest HFA data is below.

We did not find an independent open study of the outflow in Russian clubs. There is an assessment of the Getloyalty service, and then only in the retelling. According to the T-Bank blog (18 largest networks, 646 clubs, more than 100,000 reviews), on average only 56% of customers renew their subscription within 90 days after the end. The full report is closed with a form. The selections differ in different versions of the report. Therefore, take this figure as a guideline, not as a norm. For thermal spa complexes, we did not find open data on outflow at all.

Why the outflow is especially important in 2026: according to Kommersant (January 2026), the profitability of the fitness industry in 2025 was about 7%, and up to 5% is forecast for 2026. With such a margin, every departed customer who needs to be replaced with a new one is noticeably hitting profits. The best reference point for you is your own figure: calculate the retention on your database for the past year and compare with it, and not with other people's averages.

Why don't questionnaires and NPS answer the question "why do clients leave"?

The questionnaire with ratings measures the level of discontent, but does not name its cause. The person puts "3 out of 5" and closes the form. You will know that something is wrong, but you will not know what it is.

What is NPS

NPS (Net Promoter Score, Consumer Loyalty Index) is a metric based on one question: "How likely is it that you will recommend us to a friend or colleague?" on a scale from 0 to 10. The method was proposed by Fred Reicheld from Bain & The company in the article "The One Number You Need to Grow" in the Harvard Business Review in December 2003. How to count is described on the Bain website:

• 9-10 points — promoters, that is, supporters;
• 7-8 — neutral;
• 0-6 — critics;
• NPS = percentage of promoters minus the percentage of critics.

If 50 out of 100 respondents put 9-10, and 20 from 0 to 6, the NPS is 30. The names Net Promoter® and NPS® are registered trademarks of Bain & Company, NICE Systems and Fred Reicheld.

What NPS does not show

NPS is convenient to monitor the dynamics: it has become better or worse compared to the last quarter. But it has limitations that are important for working with outflow:

• There is no reason. The number 30 is the same in the club where the shower is broken, and in the club where the beloved coach left.
• The departed are silent. The survey is usually received by existing customers. Those who have already decided not to renew, most often do not respond.
• Short answers. If there is a "Why?" field after the assessment, it is often left blank or one word is written: "expensive", "inconvenient".
• There is no single standard for neighboring metrics. CSAT (satisfaction on a scale of 1-5 or 1-7) is considered by each company in its own way, so it is difficult to compare yourself with others on it.

At the same time, the price of one unsolved problem is high. In the Zendesk CX Trends 2026 report, 85% of customer service managers said that one unresolved problem is enough to lose a customer. To solve a problem, you need to know it by name. This is where talking helps.

What is an AI interviewer in messenger and how does it work?

An AI interviewer is a messenger bot that conducts a short conversation with a client instead of a questionnaire. He asks an open question, clarifies a vague answer and retains the meaning of what was said. The language model (a neural network that understands and writes text) lays out the answers of all clients by topic.

It works in three steps:

1. The question. The bot writes to the client in MAX or Telegram, introduces itself and asks if it is possible to ask a few questions. If the person agrees, the bot asks an open question, for example: "What influenced the decision not to renew the subscription?"
2. Clarification. If the answer is short ("inconvenient"), the bot asks one or two clarifying questions: what exactly is inconvenient, time, road, record for training? So "inconvenient" turns into "evening Pilates groups are always full, you can't sign up."
3. Grouping. The model reads all conversations for the period and puts labels to each: "schedule", "price", "cleanliness of locker rooms", "coach", "relocation". The manager sees not 200 disparate texts, but a list of reasons with the number of mentions and examples of quotations.

Such a bot is connected to the club's account system. Then he knows who to write to and when, and the answer gets into the client's card. How we collect such scenarios is described on the AI Agents page.

How does an AI interview differ from a questionnaire and an NPS

Criteria	Questionnaire with grades	NPS	AI-interview in messenger
What shows	Estimates for the specified items	Willingness to recommend, one number	Reasons in the client's own words
Answers the "why"	Only on items that you have come up with in advance	Almost none	Yes, including reasons you didn't know about
Clarifies the response	I don't	I don't	Yes, 1-2 questions
Customer time (our benchmark)	1-5 minutes	Less than a minute	2-5 minutes
Processing	Automatic, table	Automatic, one number	Grouping by model plus random human verification
What is better for	Regular monitoring of the service	Loyalty dynamics	Analysis of the reasons for leaving and points of discontent

These tools do not replace each other. NPS shows where loyalty has sunk, the interview explains why.

What does research say about AI interviews?

The 2019-2026 studies agree on one thing: in a conversation with a bot, people answer in more detail than in a regular questionnaire. But the format has a price, and you should know about it in advance.

Advantages

• Xiao and co-authors (2019). In the "Tell Me About Yourself" field experiment, about 600 participants were divided in half: some filled out a regular questionnaire in Qualtrics, others communicated with a chatbot. After analyzing more than 5,200 free responses, it turned out that the chatbot gave significantly greater engagement. The answers in it were more informative, more specific and clearer.
• Barari and co-authors (2025). In an AI-Assisted Conversational Interviewing web experiment with 1,800 participants, chatbots based on language models asked clarifying questions. Open answers have become more detailed and informative.
• Wuttke et al. (2026). In the work of "AI Conversational Interviewing" 571 people were interviewed in voice mode, in chat and in the "to choose" format. The conversations revealed motives and reasoning that the standard questionnaire usually does not record. Those who passed the AI interview rated it at the level of a standard questionnaire or higher.

Price

• Respondent's experience. Barari and co-authors got more detailed answers at the cost of a slight deterioration in the participants' impressions. A conversation requires more effort from a person, so it should be made short.
• Not everyone reaches the end. For Wuttke and co-authors, the proportion of completed interviews depended on the format. None of the studies we found guarantees an increase in the number of answers.
• The tendency to agree. When the bot from Barari and co-authors marked up the answers himself right during the conversation, there were a little more false positives. The authors associate them with the tendency of respondents to agree with the wording (the effect of consent).
• The refinements are poorer than a human's. In the pilot of the Lang and Eskenazi AI phone survey (2025), the quality of responses to structured questions approached surveys with a live interviewer. But the clarifying questions of AI were poorer than those of humans.

Conclusion for the club: an AI interview collects good reasons from dozens and hundreds of clients, which a live manager will not have time to do physically. But the conversation should be short, and the conclusions of the model should be selectively checked with your eyes.

How to arrange a survey of departed clients with an AI interviewer?

The process consists of five decisions: who and when to ask, about what, through what channel, with what consent and where to put the answers.

Who and when to ask

The best time to ask a question is immediately after an event that speaks about the risk of leaving. After three months, the person no longer remembers the details and does not want to answer. The club's accounting system sees such events by itself, so the bot can be launched automatically. The dates in the table are our practical recommendations, not research data.

Trigger event	When to write (our guide)	What we want to know
The client refused to renew the subscription or the period expired without renewal	Within 1-3 days	The main reason for leaving and what could change it
N consecutive absences (for example, 3 weeks without visits at the usual rate of 2 times a week)	Immediately after the rule is triggered	What prevents walking: time, health, service, motivation
The first month after the purchase has passed	On the 25th-35th day	Did expectations coincide with reality, which was unclear or inconvenient
Subscription freeze	On the day of freezing	Is it a temporary pause or is the person on the way to leaving

Select the pass threshold based on your database: see how many weeks without visits customers usually do not return. We recommend sending one person no more than one interview per quarter, otherwise the survey will become annoying.

What questions to ask: 5 templates

We advise you to limit yourself to 3-5 open questions: the rest will be done by the bot's clarifications, and a long conversation tires a person.

1. "What influenced your decision not to renew your subscription the most?"
2. "Was there a moment when you thought that the club was no longer suitable for you? What happened then?"
3. "What worked well at the club and what will you miss?"
4. "What had to be different for you to stay?"
5. "Can I give your answer to the club manager if I need to clarify the details?"

The rules for clarifications should be spelled out in the instructions for the model: no more than two clarifying questions on the topic, no dispute with the client, no ratings ("sorry you think so"), no promises on behalf of the club. If a client writes about an injury, conflict, or money to be refunded, the bot thanks and transmits the conversation to a live employee.

Example of a dialog script

This is an illustrative example, not a recording of a real conversation. Club, names and details are conditional.

Bot: Hello, Anna. This is the bot of the club "Example". Your subscription ended on September 12th, and we want to understand what can be done better. Can you answer 3 questions? It will take a couple of minutes.

Client: Come on.

Bot: What influenced the decision not to renew the subscription the most?

Client: It became uncomfortable.

Bot: I understand. Please specify what exactly has become inconvenient: working hours, travel, registration for classes or something else?

Client: Recording. There are no places for evening yoga for three days already, and in the morning I can't.

Bot: Thank you, it's important. How often has this happened in the last month?

Client: Almost every week. As a result, I just stopped walking.

Bot: What did you like about the club and what will you miss?

Client: The yoga coach is great and the pool.

Bot: Thank you, Anna. I will give your answers to the club manager.

In the client's card, after such a conversation, the labels "group registration: no seats", "evening time", "coach: positive", "pool: positive" and a short squeeze appear. Please note: there is no discount or offer to return in the dialog. Why this is important, we will explain below.

Which messenger should I use to send a survey in 2026

As of September 2026, the main channel is MAX, the additional channel is Telegram. According to Mediascope, retold by Habr and 3DNews, in June 2026, MAX's monthly audience was 86.26 million people, Telegram — 75.69 million, WhatsApp — 63.51 million.

• MAX from September 1, 2025 is preinstalled on smartphones according to 156-FZ dated 06/24/2025. But since the summer of 2026, it is not available in the App Store and Google Play, and it is more difficult for new users to install it.
• Telegram has been officially restricted by Roskomnadzor since February 10, 2026, and the messenger is unstable. A complete lockdown was not announced. Keep it as a second channel for those clients for whom it works.
• WhatsApp has not been working in Russia without a VPN since February 11, 2026. It's not worth building a survey on it.

It is better to keep the interview logic separate from the channel so that the same scenario works in both MAX and Telegram. This is how our chatbots work.

How not to turn a survey into an advertisement

A survey without discounts and offers is most likely not considered advertising, but the law does not give an unambiguous answer. We did not find a direct explanation from the FAS about customer surveys. In the letter of the FAS dated 04/09/2025 No. AK/32908/25 about what is considered advertising, surveys are not mentioned separately.

The reference point is given by general norms. The law "On Advertising" (38-FZ, Article 2, Part 2, paragraph 3) does not apply, for example, to reference, informational and analytical materials whose main purpose is not to promote the product. If the message only collects the client's opinion, there is no purpose of promotion in it, in our opinion.

If you add "come back with a 20% discount" to the survey, the message has a promotion goal, and it may be recognized as advertising. Article 18 of the 38-FZ allows advertising on telecommunication networks only with the prior consent of the subscriber, and the distributor must prove the consent. For SMS mailings, Article 44.1 of the law "On Communications" applies separately: the subscriber's prior consent is required, and from 2025 the subscriber can refuse to send through the operator.

Rule of thumb: interview — separately, offer to return — separately. The offer is sent by the manager or another script, and only to those who have agreed to the advertisement. How to qualify a specific message and mailing scheme, check with a lawyer.

What kind of consent is needed for personal data

The client's responses linked to his name and phone number are personal data, and consent is required for their processing. According to Article 9 152-FZ, consent must be specific, informed and unambiguous. From September 1, 2025, it is issued separately from other documents, for example, not a clause in the subscription agreement. Violations in the registration of consent, according to the Guarantor, threaten the organization with a fine of 300 to 700 thousand rubles under Article 13.11 of the Administrative Code.

Remember that these are different consents:

• consent to the processing of personal data — according to 152-FZ, a separate document;
• consent to receive advertising — according to Article 18 of the law "On Advertising";
• consent to SMS-mailing — according to Article 44.1 of the law "On Communications".

In the consent to data processing, it is necessary to explicitly state the purpose of "surveys on the quality of services" and that the answers are analyzed by an automated system. Check the consent text and storage scheme with a lawyer.

Where to store answers

The answers should be stored in the client's card in the club's account system, and not in a separate table from the marketer. Then the administrator, who will see the client at the reception in six months, will know that he left because of an appointment for evening yoga.

Some of the tools for this are already there. For example, in the integration of "1C:Fitness Club" with MAX, customer messages from the messenger are sent directly to the program, and you can write to the client from his card. Mass and trigger mailings for September 2026 are announced there as "soon". At 1C:Fitness Club" has a public API documentation for working with clients and subscriptions. Recording of reason labels and conversation squeezes into the client card can be implemented via the API or configuration refinement. What exactly is available in your version, we check on the audit. If you have a2club or another system, you also need to check the capabilities separately. We make such bundles as part of 1C integration.

What to save:

• date and trigger of the interview;
• cause labels (2-5 pieces from the general reference);
• squeeze in 1-3 sentences;
• full text of the conversation with limited access and clear shelf life;
• check if the client has agreed to re-contact.

How to turn customer responses into actions?

Answers are useful only if they are followed by a solution and repeated measurement. By themselves, the labels in the cards do not change anything. Here is the duty cycle:

1. Collect a summary once a month. Take the top 5 reasons for leaving with the number of mentions and 2-3 quotes for each.
2. Check the markup selectively. We recommend that the manager reads 10-20 conversations in their entirety and checks them against the model labels. If the model confuses the topics, correct the label directory or the instructions.
3. Choose one or two reasons. Take the ones that you can fix: the schedule of groups, cleanliness, the work of administrators. Moving a client to another area cannot be fixed.
4. Assign an action and a deadline. For example, to open a second evening yoga group from October.
5. Measure again. After 1-2 months, see if there are fewer mentions of this reason and if the retention in the affected segment has changed. Compare with your own figure before the changes.
6. Close the loop with the client. If the client has allowed repeated contact, the manager can inform him that the problem has been fixed. If there is an offer to return in such a message, this is already an advertisement, and a separate consent is required for it.

Launch checklist

• The current retention is calculated according to its base.
• 2-3 triggers and frequency are selected (we recommend no more than once a quarter per person).
• 3-5 open questions and clarification rules have been written.
• There are cases when a bot transmits a conversation to a person.
• A separate consent to the processing of personal data has been issued.
• The interview does not contain discounts and offers.
• The bot works in MAX, Telegram — as an additional channel.
• Tags and squeezes are written to the client's card.
• A person responsible for the monthly summary and actions has been appointed.

For clubs and networks where records are already kept in 1C or an industry system, we collect such a cycle in its entirety. Read more about this on the page for fitness clubs.

Sources

1. Health & Fitness Association. 2025 Fitness Industry Benchmarking Report — 30.09.2025
2. Uptivo. Gym member retention statistics — 2026
3. T-Business secrets. The economy of impressions in the fitness industry (Getloyalty data) — 07/16/2026
4. Kommersant. About the profitability of the fitness industry — 01/13/2026
5. Reichheld F. The One Number You Need to Grow. Harvard Business Review — 12.2003
6. Bain & Company. Measuring Your Net Promoter Score — current page
7. Zendesk CX Trends 2026, PR Newswire — 18.11.2025
8. Xiao Z. et al. Tell Me About Yourself: Using an AI-Powered Chatbot to Conduct Conversational Surveys with Open-ended Questions. arXiv:1905.10700 — 05.2019
9. Barari S. et al. AI-Assisted Conversational Interviewing: Effects on Data Quality and Respondent Experience. arXiv:2504.13908 — 09.04.2025
10. Wuttke A. et al. AI Conversational Interviewing: Scaling Up Semi-Structured and In-depth Interviews. arXiv:2606.20064 — 18.06.2026
11. Lang M. M., Eskenazi S. Telephone Surveys Meet Conversational AI: Evaluating a LLM-Based Telephone Survey System at Scale. arXiv:2502.20140 — 27.02.2025
12. Habr. Mediascope: in June 2026, Max for the first time took first place among messengers in Russia in terms of monthly audience coverage — 07/22/2026
13. 3DNews. Max surpassed Telegram in monthly audience, but its growth almost stopped — 07/21/2026
14. ConsultantPlus. Review of Federal Law No. 156-FZ dated 06/24/2025 — 06/24/2025
15. iXBT. MAX messenger disappeared from Google Play — 07/16/2026
16. Fontanka. Roskomnadzor named the reason for the Telegram restrictions — 02/10/2026
17. CNews. The authorities finally nailed WhatsApp in Russia — 02/11/2026
18. Federal Law of 13.03.2006 No. 38-FZ "On Advertising", Article 2 — ed. from 08/04/2026
19. Federal Law No. 38-FZ "On Advertising", Article 18 — current edition
20. Letter of the FAS of Russia dated 04/09/2025 no. AK/32908/25 — 09.04.2025
21. Federal Law No. 126-FZ "On Communications", Article 44.1 — ed. 2025
22. Federal Law No. 152-FZ "On Personal Data", Article 9 — ed. 2025
23. The guarantor. Fines under Article 13.11 of the Administrative Code for violations in the processing of personal data — 2025
24. 1C:Fitness club. Integration with the MAX messenger — September 2026
25. 1C:Fitness club. API Knowledge Base and Documentation — September 2026