Past customers not coming back is rarely just a marketing failure. Before adding vouchers or running ads to win them back, a manager should check three things in order: whether the repeat customer number is counted correctly, whether the customer's last purchase had a problem, and whether anyone contacted the customer at the moment they began buying less often.
Is your repeat customer rate counted correctly?
Usually not, because each channel identifies customers with a different key. Stores and Facebook orders through Nhanh.vn have phone numbers. Shopee and TikTok Shop usually mask the buyer's phone number, leaving only the name or the marketplace account email. The same person buying at a store and then on Shopee becomes two customers.
| Channel | Usable identification key | Trap |
|---|---|---|
| Store (POS) | Phone number | Staff enter an internal or shared number |
| Facebook, website through Nhanh.vn | Phone number | Numbers can be partly masked, and masked numbers cannot be merged |
| Shopee | Buyer account name | Duplicate names, cannot be linked to other channels |
| TikTok Shop | Buyer email / account | No phone number |
The result: a repeat rate calculated across all channels is usually lower than reality and hard to compare between channels. The right way is to calculate it separately per channel with one fixed key, not using one key to fill in for another. A customer with no key counts as one customer per order, accepting a low but honest number. Merge across channels only when the customer's accounts are truly linked, for example through web login or a points program.
Did the customer's last purchase have a problem?
Customers often do not come back because something went wrong last time: an order cancelled for being out of stock, a failed delivery, goods returned for the wrong size, or a one-star review nobody answered. These points are scattered in many places: order status on the marketplace, delivery status in the sales software, reviews on the product page.
The management question is: among lost customers, how many had a failure point on their last purchase? If that share is high, money should go into operations and products, not advertising. Also read how to separate returned orders from failed deliveries, because lumping the two together points at the wrong team to fix it.
Which customer group is dropping off, and since when?
RFM answers this: Recency is how long since the last purchase, Frequency is how many purchases, Monetary is how much was spent. Customers who buy a lot and spend a lot but have not returned for a long time are the most worrying group. Losing one person from this group each month costs more than losing ten one-time buyers.
Do not set one shared "long time no purchase" threshold for everyone. Each customer has their own buying rhythm. A customer who usually returns after three weeks but has not bought in six is a signal; for a customer who buys once a quarter, six weeks is nothing unusual. Compare the current gap with that customer's own rhythm.
How can AI and automation help?
AI speeds up two things: reading the RFM picture and drafting contact scripts. For example, a prompt for ChatGPT or Claude, with names and phone numbers removed from the data:
Dữ liệu là bảng phân khúc RFM 12 tháng (số khách, chi tiêu, khoảng cách mua trung bình theo phân khúc).
Phân khúc nào đang thu hẹp nhanh nhất? Khách rời đi chủ yếu từ phân khúc nào?
Đề xuất 3 việc cho từng phân khúc đáng lo, mỗi việc kèm chỉ số đo được sau 30 ngày. Không bịa số.
The rest is operations, which AI cannot do for you: someone has to call, message and record the result. From the Pro plan up, the Chăm sóc khách hàng (Customer care) page of Quantri360 generates tasks from drop-off points: customers who messaged but did not order, abandoned carts on the web, cancelled orders, failed deliveries, returned orders, bad reviews, and loyal customers past their buying rhythm. Each task is assigned to one person with a handling deadline.
How do you measure whether care is working?
Measure by valid orders placed after contact, compared with a control group that was not contacted. Without a control group you cannot tell whether customers came back on their own or because they were called. Quantri360 lets you set a control ratio per brand, holding back a portion of customers from task assignment for comparison.
Metrics to track:
- The repeat customer rate per channel, with one fixed key per channel.
- The number of loyal customers past their buying rhythm with open tasks, and the time to first contact.
- The rate of valid orders after contact, compared with the control group.
- The share of lost customers with a failure point on their last purchase.
See also the overview page, in-depth customer analysis and sending ZNS by order to message customers at the right time after purchase. See the pricing for plans that include the customer care page.
Read more: data-driven marketing, the overview article for this topic.
In short
- Calculate the repeat rate separately per channel with one fixed identification key, without substituting keys.
- Check failure points on lost customers' last purchase before spending more on marketing.
- Use RFM and each customer's own buying rhythm to spot loyal customers who are buying less often.
- Assign contact tasks to specific people, with deadlines and recorded results.
- Measure care by valid orders after contact compared with a control group.