RFM Segmentation Case Study in Pakistan

The Client

A Lahore-based D2C brand selling pet supplies — imported kibble, cat litter, grooming products, tick-and-flea spot-ons, and veterinary consumables — to dog and cat owners across Lahore, Karachi, Islamabad, and Rawalpindi. The store ran on Shopify with Klaviyo collecting emails, fulfilled its own orders from a Lahore warehouse, and did roughly 3,800 orders a month at an average order value of PKR 4,600. Around 70% of orders were cash-on-delivery, which matters more to this story than it first appears.

The category is close to ideal for retention work. A medium-breed dog finishes a 12kg bag of kibble in about two months. A single cat goes through 5kg of litter in three to four weeks. Deworming repeats every quarter, and tick-season spot-ons repeat monthly through the summer. A customer who buys once and disappears is not a small sale lost — it is twelve months of predictable reorders lost, usually to a marketplace seller who shows up first on the reorder search.

The brand’s owner engaged WeProms after a board-like moment with their accountant: revenue was growing, but gross margin had compressed for five straight months while Meta acquisition costs climbed. The diagnosis they arrived with was “we need better ads.” The diagnosis they left with was that the store was buying customers once and then ignoring them.

The Problem

Four retention failures were compounding under the surface:

• One list, one message. Every email the brand sent went to all 41,000 customer records. A customer who had spent PKR 180,000 over two years and a customer who grabbed one discounted bag in 2024 received the same newsletter, at the same cadence, with the same coupon. Champions were being trained to wait for discounts they did not need.
• Discount dependency. 41% of orders carried a discount code. Many of those codes had leaked to deal-hunting Telegram and Facebook groups, so a meaningful slice of “customers” were arbitrageurs who bought only when a code was live. Every blanket discount also cut the margin on full-price buyers who would have converted anyway.
• Replenishment reorders going elsewhere. The store had no mechanism to know when a customer’s kibble was running out. When reorder moment arrived, the customer searched, found a marketplace listing with free delivery, and bought there. The brand had already paid to acquire that customer once.
• Corrupted customer data. Cash-on-delivery buyers overwhelmingly check out as guests, often with typo’d or variant emails. The same human existed three or four times in the database with one order each. Every retention metric computed on that data — repeat rate, average orders per customer, time between purchases — was quietly wrong before any strategy question was asked.

This is a pattern we see constantly in Pakistani ecommerce: acquisition machinery that works harder every quarter, attached to a retention layer that was never built. The engagement brief was to fix the second half.

Phase 1 — Data Audit and RFM Baseline (Weeks 1-2)

Nothing gets scored until the customer table is honest. Phase 1 was deliberately unglamorous.

Identity deduplication. We pulled 24 months of order history and collapsed customer records using matched email, normalized phone number, and delivery address fingerprints. Of 41,000 records, roughly 7,400 were fragments of existing buyers — the COD guest-checkout problem in pure form. The result was 33,600 unique customers, of whom 21% turned out to have two or more orders once their fragments were merged. The store’s true repeat behavior had been invisible, not absent.

Email matching. The deduplicated base was matched back to Klaviyo. Deliverable, consented email coverage rose from 63% of buyers to 78% after we added a post-purchase account prompt on the thank-you page and a phone-number-based recovery path for guest orders.

RFM scoring. Every customer was scored on recency (days since last order), frequency (orders in 12 months), and monetary (trailing 12-month spend), each graded into quintiles. The composite score produced the familiar pyramid: a small top tier carrying a disproportionate share of revenue. The baseline report made the misallocation obvious — the top 9% of customers generated 34% of revenue and received exactly the same treatment as everyone else.

Baseline lock. Before anything shipped, we froze the numbers every later claim would be judged against:

Metric	Baseline (pre-work)
60-day repeat purchase rate	19.4%
60-day customer LTV	PKR 6,480
Email share of online revenue	11%
Orders carrying a discount code	41%
Average orders per customer	1.6
Unique customers (deduplicated)	33,600

Phase 2 — Segment Architecture and Platform Wiring (Weeks 3-5)

With clean scores in place, we built the six operating segments that would carry the whole program — the core of our customer segmentation and personalization service. The test for every segment was operational: if it did not have a distinct treatment, it did not deserve to exist.

Segment	Customers	Revenue share	Core treatment
Champions	9%	34%	Early access, priority stock, zero discount codes
Loyal regulars	17%	27%	Replenishment reminders, bundle and top-up offers
Potential loyalists	23%	18%	Second-order nudge, lifestage education
At-risk	14%	12%	Win-back ladder with timed, tiered incentive
New and unprofiled	21%	8%	Welcome series with pet-profile capture
Dormant	16%	1%	Low-frequency quarterly reactivation

Platform wiring. Segments synced to Klaviyo as dynamic smart lists, refreshed nightly from the scored customer table. Custom properties were added per profile: pet type, lifestage, breed size, last category purchased, and the customer’s own median purchase interval where two or more orders existed. Every property doubled as a personalization token and a targeting filter.

Offer policy. The document that changed the economics of the program was one page long: which segments may receive discounts, of what depth, at what frequency. Champions and loyal regulars receive early access and bundles, never percentage-off codes. At-risk receives a ladder that escalates from content, to a category best-seller, to 10%, then 15%. Dormant gets one deeper offer per quarter and nothing more. Discounting stopped being the default lever and became a targeted one.

Profile capture. New buyers entered a welcome series that asked one question in exchange for value: “What pet is this for — dog or cat, and what age?” A single progressive question in the second welcome email, with the answer feeding lifestage flows. Profile completion across the new-buyer cohort reached 62% by week eight.

Phase 3 — Personalized Campaigns and Flows Live (Weeks 4-8)

Phase 3 turned architecture into revenue. The build leaned on the lifecycle flow patterns we deploy across Pakistani D2C stores, tuned to this catalog’s physics.

Replenishment engine. The flagship flow. For every customer with a calculable interval — their own median where possible, otherwise the category default — a reminder fires seven days before the projected empty date, with a second nudge three days after it. A small-breed dog owner gets a 3kg kibble reminder at day 23; a two-cat household gets litter at day 24; deworming repeats at day 83. The email shows the exact product, the exact size, the customer’s usual quantity, and a reorder button that pre-fills the cart. By week eight this single flow was producing more attributed revenue than the old weekly newsletter had in a month.

Lifestage transitions. Puppies and kittens switch to adult food around twelve months — historically the moment this brand’s customers churned to a marketplace, because nobody owned the transition. A lifestage flow now triggers at month ten for buyers flagged as puppy or kitten: a switch guide, a transition-feeding schedule, and a bundle offer timed to the switch. The equivalent moment exists in the senior lifestage and in seasonal tick-and-flea protection, which in Pakistan runs hard from May through August.

Win-back ladder. At-risk customers — good historical value, silent for 1.5 times their own interval — entered a four-step ladder: a genuinely useful piece of content (feeding guide, seasonal checklist), then a category best-seller email, then 10% off, then 15% off. Of the 2,340 at-risk and dormant customers the ladder mailed, 326 reordered within 60 days — a 14% reactivation on a segment the store had previously reached only with the same blanket newsletter everyone ignored.

Category cross-sell by household profile. The pet-type and lifestage properties unlocked a second revenue layer beyond replenishment. Cat-litter buyers received odor-control and grooming attach offers timed to their litter cycle, not a generic seasonal blast. Dog owners finishing a grooming shampoo showed predictable interest in coat supplements within the following month. Because cross-sell emails referenced the household’s actual pet profile, they read as service (“for a two-cat home”) rather than advertising, and the attach rate on replenishment orders rose from 11% to 17% over the window — margin that arrives without a fresh acquisition cost attached to it.

Dynamic campaign content. The weekly campaign remained, but its blocks became conditional. Dog owners see dog imagery and dog products; cat owners see the cat edit. Champions see new-arrival kibble drops before the site. Lapsed bargain hunters — the dormant segment — see the quarterly deep offer instead of weekly noise, which cut list fatigue: unsubscribe rates fell from 0.84% to 0.51% per campaign even as send relevance rose.

Program element	Week 4 state	Week 8 state
Flows live	2 (welcome, cart)	9 (incl. replenishment, win-back)
Email revenue share	13%	17%
Discount-coded orders	38%	31%
Campaign CTR	1.9%	3.4%
Unsubscribe rate per send	0.84%	0.51%

Phase 4 — Measure Against Holdouts and Compound (Weeks 8-12)

The discipline that makes the final numbers believable: every automated flow went live with a 10% randomly withheld control group. The replenishment flow’s lift is measured against customers whose kibble ran out in silence, not against a hopeful baseline. Every number in the results table below is flow-versus-control or cohort-versus-cohort.

Segment migration tracking. The scored table re-ran nightly, so we could watch customers climb. Over the quarter, 2,100 customers moved up at least one tier — potential loyalists becoming loyal regulars, at-risk customers recovered into the loyal tiers. Migration up is the compounding asset; a one-off revenue spike is not.

Cadence governance. Each segment carries a maximum of two marketing sends per week, replenishment and transactional flows excluded. Fatigue stayed low because frequency followed value, not the calendar.

Cohort LTV dashboard. The owner’s weekly view became one chart: 60-day LTV by acquisition month, split by email-engaged versus non-engaged. By week twelve the engaged curve ran visibly above the non-engaged curve, and the discount-rate line had fallen thirteen points from baseline.

Retiring the leak. With the offer policy in force, the leaked-code problem was attacked directly: codes became single-use and segment-bound rather than universal strings that deal groups could repost, and expiring seasonal codes replaced the permanent 10% welcome voucher. Combined with reserving incentives for at-risk and dormant tiers, this is what let discount share fall thirteen points without a volume dip — the buyers who actually needed a discount to reorder still received one, and the buyers who did not, stopped getting one.

Handoff. The engagement closed with the segment table, offer policy, and flow library documented so the brand’s two-person marketing team could run the program without agency dependency: refresh the RFM scores monthly, review segment migration weekly, and hold the cadence caps. A 90-day review calendar was set for the two moments the program depends on — pre-tick-season inventory planning in April and the Q4 gift-season campaign, where segment-specific offers now replace the old blanket sitewide sale.

Final Results at 90 Days

Metric	Before	After	Change
60-day customer LTV	PKR 6,480	PKR 7,970	+23%
60-day repeat purchase rate	19.4%	24.6%	+5.2 points
Email share of online revenue	11%	19%	+8 points
Discount-coded orders	41%	28%	-13 points
Lapsed customers reactivated	—	326 of 2,340	14% in 60 days
Customers moving up a tier	—	2,100	Over the quarter

These are illustrative outcome ranges built from the patterns WeProms sees across Pakistani D2C retention engagements — not audited third-party figures. They exist so a growth team can sanity-check what a segmentation program should plausibly return in this market, at this order volume, in one quarter.

What Made This Work

1. Identity cleanup preceded scoring. Merging 7,400 fragment records changed every downstream number. RFM scores computed on fragmented data systematically understate frequency — the exact dimension that identifies your best customers — so the unglamorous dedupe pass was the highest-leverage work in the engagement.

2. Every segment had a treatment, not just a definition. Six segments, six playbooks. Analytical segmentations that stop at a slide deck change nothing; the program worked because each segment mapped to a concrete offer policy, cadence, and flow set.

3. Replenishment was timed to the customer, not the calendar. A fixed 30-day reminder is wrong for almost everyone — it arrives either uselessly early or uselessly late. Interval-based timing with a buffer made the reminder feel like service rather than marketing, which is why it earned opens instead of unsubscribes.

4. The offer policy protected margin on purpose. Pulling percentage-off codes out of the champion and loyal tiers cut discount-coded orders by thirteen points without volume loss, because those customers were never price-blocked to begin with. Discount budget was redeployed to the at-risk ladder, where it actually unlocks orders.

5. Holdouts made the claims falsifiable. With 10% controls on every flow, lift was measured, not assumed. When the replenishment flow underperformed for cat-litter buyers in its first version, the control comparison caught it in week five and the copy was rebuilt — before the dashboard had a chance to flatter it.

What Teams Can Apply

For Pakistani D2C and ecommerce teams sitting on unsegmented order history:

1. You already have the data RFM needs. No customer-data platform is required — 24 months of Shopify or WooCommerce orders plus email is enough for a first scoring pass. Start with recency, frequency, and monetary quintiles and resist building more dimensions than you can treat differently.
2. Deduplicate before you score. COD guest checkouts fragment your best repeat buyers into fake one-time customers. Match on email, normalized phone, and address before computing any retention metric; if your repeat rate looks implausibly low, this is usually why.
3. Time consumable reminders to the customer’s own interval. Median days between past purchases, minus a buffer, beats any fixed schedule. Where you lack history, use category physics — bag sizes, household sizes, seasonal cycles — as the default.
4. Write down who is allowed to get a discount. A one-page offer policy is the cheapest margin protection available to a Pakistani D2C brand. Blanket codes leak to deal groups and train full-price buyers to wait; tiered, targeted incentives do neither.
5. Hold out 10% of every flow from day one. It costs almost nothing in deferred revenue and converts every later claim about lift from an opinion into a measurement. If a flow cannot beat its own control group, fix or kill it.

WeProms Digital has run this framework across pet, baby-care, beauty, food, and electronics ecommerce in Pakistan. The segment definitions and flow set change with each catalog; the sequence — clean identity, operational segments, interval-timed personalization, holdout measurement — stays the same.