AI agents: Retention and loyalty

What it does: Finds at-risk customers in the export and suggests who to message and why.

What a person decides: The list and the text.

For whom: Manager.

How to measure: Customers who came back.

Example result: “Drifting away: Aizhan and Marat — used to come monthly, now absent for three months; Gulnara — after a complaint about a haircut. Message Aizhan and Marat first; for Gulnara, a call from the manager first; don't message Olga — she asked not to be contacted.”

What it does: Birthday greetings and personal reasons to reach out.

What a person decides: The text and the gift.

For whom: Customer.

How to measure: Visits from occasions.

Example result: “Happy birthday, Gulnara! Thank you for being with us for three years now — it's a real pleasure to see you again and again. Our gift to you: 20% off any service through the end of the month — just mention it to the front desk when you book.”

What it does: Awards points and reminds customers about them.

What a person decides: Program rules.

For whom: Customer.

How to measure: Bonuses redeemed; return visits.

Example result: “Aigerim, you have 4,500 bonus points — you can use them to pay up to 30% of your next visit. They're valid until December 31, so now's a great time to use them. Shall I find a convenient time to book?”

What it does: A short survey; low scores go to the manager the same day.

What a person decides: The response.

For whom: Customer, manager.

How to measure: NPS; response time to low scores.

Example result: “Category: unhappy — to the manager today. Why: Aidana gave 2 of 5 and wrote ‘waited 40 minutes, nobody even apologized’ (survey reply on October 8, 14:20); she's been a customer since 2023. Suggest the manager call her before end of day and ask how to make it right.”

What it does: Sends a reminder before expiry and offers a renewal.

What a person decides: Renewal terms.

For whom: Customer.

How to measure: Renewal rate.

What it does: A week later, checks whether the issue is resolved and invites the customer back.

What a person decides: Compensation.

For whom: Customer.

How to measure: Customers who came back after a complaint.

Example result: “Hi Aidana! A week ago you told us you had to wait a long time for your appointment. We've changed the schedule so there's now buffer time between patients. We'd love to see you again — if you like, I'll find a time when the doctor is running on schedule.”

What it does: Flags the most valuable customers and suggests ways to show appreciation.

What a person decides: What to offer.

For whom: Manager.

How to measure: VIP retention.

Example result: “Action: assign a dedicated manager to the 12 customers who made up 30% of revenue from closed deals over six months: a call from the manager every quarter and priority booking. Start with Bereke LLP and Aigul Saparova — their purchases over 2 months are half the usual. Impact: top customers feel valued and don't quietly leave. Risk: others shouldn't feel second-class — no public ‘statuses.’ Fallback: if purchase frequency doesn't change in a quarter, keep only the calls.”

What it does: Splits the customer base into groups by behavior.

What a person decides: Who gets what message.

For whom: Manager.

How to measure: Revenue by segment.

Example result: “Situation: 545 customers in the base: regulars — 340 (with us over a year, no complaints), new — 120 (under 3 months), drifting away — 85 (flagged ‘at risk’ last month). Key point: a third of those drifting away had complaints. Options: 1) personal messages to those drifting away — slower but more precise; 2) one promotion for everyone — fast, but loses the personal touch. Recommendation: personal messages to those drifting away, starting with those who complained, and a welcome series for new customers. Today: approve the texts for both groups.”

What it does: Warns in advance based on signs of churn.

What a person decides: The action.

For whom: Manager.

How to measure: Churn.

Example result: “Score: 4 of 5 — high churn risk. Why: per front-desk check-ins, Yerlan came 2 times in September versus 6 in August; his membership ended October 1 and wasn't renewed; on September 25 he complained about noise in the gym. What to do: a call from the manager this week — start with the complaint, not the renewal.”

What it does: Prepares topics and events for a customer chat or channel.

What a person decides: Posts.

For whom: Staff member.

How to measure: Community activity.

Example result: “Customer chat plan for October. Week 1: poll ‘Which service should we add?’ — 3 options drawn from common customer questions. Week 2: a specialist answers the question of the week. Week 3: a 20-minute live Q&A. Week 4: poll results and what we've already done. Chat rules: no third-party ads, complaints go to direct messages. To decide: a prize for taking the poll, and the budget.”

What it does: A letter to customers about new prices: from when, how much, and why, plus what stays the same for loyal customers.

What a person decides: Sending and terms for loyal customers.

For whom: Customer.

How to measure: Churn after the price increase.