A retail chain in Tangerang had run a loyalty program for over two years, collecting phone numbers, purchase history, and visit frequency across its branches. Nobody had looked at the data beyond the monthly points-redeemed report. This customer loyalty data case study is about what happened when we finally did, and it's a useful reminder that most SMEs don't have a data problem, they have a looking problem.
The chain sold home goods across six outlets, with a loyalty app that customers used to earn points on purchases. The data existed. The infrastructure existed. What didn't exist was anyone asking a simple question: who are our members, actually, and what do they each need from us right now?
Three segments and one WhatsApp campaign strategy later, repeat purchases lifted measurably within a single quarter, without touching AI, without new software, and without a marketing agency.
The Data Was Already There
When we pulled the loyalty database, it held roughly 40,000 registered members accumulated over two years. The chain had never segmented this list beyond a single undifferentiated broadcast they sent maybe four times a year, "Diskon 20% weekend ini," sent identically to every member regardless of whether they'd shopped last week or eighteen months ago.
This is the pattern I see constantly. A pharmacy chain we worked with had the same issue with inventory across branches, data existed in each branch's system but nobody had connected it into a single usable view. Here it was the same story, applied to customers instead of stock.
The first step wasn't building anything. It was exporting the existing loyalty data into a spreadsheet and looking at three fields that were already being captured: last purchase date, purchase frequency over the past year, and average basket size.
Three Segments, Not Fifty
The temptation with customer data is to over-segment, building twelve tiers with elaborate scoring. We deliberately kept it to three, because three segments a small marketing team can actually act on beat twelve segments that overwhelm them into inaction.
Lapsed customers (no purchase in 90+ days): roughly 14,000 members. These people had shopped before and stopped, which is a fundamentally different problem from customers who never engaged. The message here needed to answer "why should I come back," not "here's a generic discount."
Frequent shoppers (purchased at least monthly): about 6,000 members. These were the chain's best customers, and they were getting the exact same generic broadcast as everyone else. The opportunity here was recognition and small perks that cost little but reinforced loyalty, not blanket discounts they didn't need to be convinced by.
High-basket customers (top 20% by average transaction value, regardless of frequency): roughly 8,000 members, with meaningful overlap with the frequent group but also including occasional big-spend shoppers. These customers responded to different messaging entirely, curated recommendations and early access rather than percentage-off coupons.
The remaining members fell into a general "active but average" bucket that continued receiving the standard broadcast, since building a bespoke message for every customer isn't worth the effort until the top segments prove out.
The Campaign: WhatsApp, Not a New Platform
The chain already used WhatsApp Business for order confirmations. We didn't introduce a new channel, we just used the existing one with different messages per segment, sent through the same broadcast tool the team already knew.
- Lapsed segment got a direct, honest message acknowledging the gap ("Sudah lama tidak belanja, ini yang baru buat Anda") paired with a modest, time-limited incentive to re-engage, not a huge discount that would erode margin, just enough friction removed to bring them back once.
- Frequent segment got early notice of new stock arrivals before the general public, a "thank you for being a regular" framing rather than a discount push.
- High-basket segment got curated bundle suggestions based on their past purchase categories, closer to a personal recommendation than a mass promotion.
None of this required AI. It required someone spending an afternoon segmenting a spreadsheet and writing three different WhatsApp templates instead of one. This is worth stating plainly because plenty of vendors would have pitched a machine learning recommendation engine for a problem that a well-thought-out spreadsheet solved first.
Where AI Actually Entered, Later
AI became relevant only in a second phase, after the manual segmentation proved the concept worked. Once the chain saw the lapsed segment responding, they wanted to test finer-grained targeting, predicting which lapsed customers were likely to respond to a re-engagement offer versus which were genuinely gone. That's a reasonable use for a lightweight predictive model, but it was step two, built on a foundation that manual segmentation had already validated with real revenue data.
This ordering matters more than the tools involved. Businesses that reach for AI or automation before doing the basic segmentation work usually end up automating a strategy nobody has tested. Here, the manual approach de-risked the investment before any technical build happened.
The Results in One Quarter
Measured against the prior quarter's baseline for the same customer segments:
| Segment | Change observed |
|---|---|
| Lapsed customers | Meaningful share returned with at least one purchase |
| Frequent shoppers | Higher engagement with early-access messages than generic broadcasts |
| High-basket customers | Increased response to curated recommendations versus blanket discounts |
The specific percentages varied by branch and category, but the directional result was consistent across all six outlets: segmented messaging outperformed the single undifferentiated broadcast the chain had relied on for two years.
The Practical Takeaway
This customer loyalty data case study isn't really about loyalty programs, it's about the gap between collecting data and using it. Most SMEs already have enough data to run three good customer segments, they just haven't looked. Before investing in AI-driven personalization or a new CRM platform, export what you already have, split it into two or three groups based on fields you're already capturing, and send different, honest messages to each. If that works, and it usually does, you'll know exactly where a smarter, more automated system earns its cost later. If you're weighing whether your current setup can even support this kind of segmentation, it's worth checking whether your business has outgrown spreadsheets as the next step. For businesses that want a partner to run this kind of audit properly, that's the kind of engagement worth exploring through ervandra.com/partner.