Customer Win-Back Campaign Case Study in Pakistan

The Client

A Lahore-based consumer-electronics retailer with three physical outlets across the city and an online store running since 2019. The catalogue spans mobile phones, laptops, home appliances, and accessories, with an average order value around PKR 21,000 online — a phone upgrade or a washing-machine purchase sits at PKR 60,000 to PKR 180,000, while accessory orders run PKR 2,500 to PKR 6,000. Online revenue averaged PKR 24M a month across roughly 1,150 orders, supplemented by a marketplace storefront that generated sales but no owned customer relationships.

That last point was the strategic itch. Marketplace orders handed the customer relationship to the platform, while the owned store quietly accumulated a database of 38,400 past purchasers — and did almost nothing with it. Acquisition through Meta ads had worked for years, but the retailer’s blended new-customer acquisition cost had risen from roughly PKR 3,600 to PKR 4,800 over eighteen months, and management could see the ceiling approaching.

Cart-recovery flows had already been rebuilt and were converting, which made dedicated customer win-back campaigns the obvious next retention layer — beyond the cart flow and one newsletter, everything after the first purchase was silence. The brief to WeProms Digital was blunt: the database is the cheapest inventory we own, figure out what it is worth, and get the high-value sleepers back.

The Problem

The diagnostic in week one painted a consistent picture of an asset decaying through neglect:

• 14,200 lapsed purchasers. Customers with no order in 90+ days, including 1,900 high-value buyers (two or more past orders, or lifetime spend above PKR 45,000) who represented an estimated PKR 240M in historical revenue.
• One-size-fits-all newsletter. A single monthly send to the entire list — active buyers, dead addresses, and everything between — averaging an 11% open rate and a 0.8% click rate. Unsubscribes from lapsed recipients were the fastest-growing segment of list churn.
• Decaying deliverability. Spam complaints sat at 0.41%, four times the 0.1% threshold at which mailbox providers start diverting mail to spam. Seed tests placed the domain’s inbox rate at 89%. No SPF, DKIM, or DMARC records existed on the sending subdomain.
• No suppression logic. 4,100 addresses with zero opens or clicks in over twelve months were mailed every month, feeding the complaint and bounce signals that pushed delivery down further.
• Email revenue share in decline. Email drove 13% of online revenue, down from 19% two years earlier, and almost all of it from the newsletter and the cart flow — nothing targeted the post-purchase lifecycle.

The core failure was definitional as much as operational: the retailer had never defined what “lapsed” means for its own category. Electronics is not fashion — a phone buyer on an 18-to-24-month upgrade cycle and an accessory buyer on a 6-week repurchase rhythm were being treated as identically “lapsed” after 90 days. Until lapse tiers reflected the store’s actual repurchase gaps, no win-back offer could land with relevance.

Phase 1 — RFM Diagnosis and Deliverability Cleanup (Weeks 1-2)

The first fortnight produced no sends at all. It produced data hygiene and definitions.

Repurchase-gap analysis. We pulled three years of order history and computed the median repurchase gap for repeat buyers by category: 41 days for accessory-led buyers, 96 days for small appliances, and 19 to 24 months for phone and laptop buyers. These gaps set the lapse tiers — a 30-day-silent accessory buyer is late, while a phone buyer at 30 days is perfectly normal. The base was then scored with RFM-driven customer segmentation — recency, frequency, monetary — which surfaced the 1,900 high-value lapsed cohort and 12,300 standard-tier lapsed customers.

The RFM cut also exposed a sharper problem hiding inside the averages: of the 38,400 contacts, only 6,900 had ever bought twice. The store was not leaking loyal customers so much as never creating them — single-order buyers were 82% of the base, and the lapsed pool was mostly people the store had met exactly once. That reframed the objective: win-back was not about recovering a lost tribe of loyalists, it was about giving one-time buyers their second reason to shop, which is the cheapest loyalty programme any retailer will ever run.

Domain authentication. SPF, DKIM, and a quarantine-mode DMARC record were published on the sending subdomain. Bounce processing was corrected so hard bounces suppressed automatically instead of retrying indefinitely.

List hygiene. The 4,100 addresses with no engagement in twelve months were suppressed into a sunset segment, and 630 role-based addresses (info@, admin@) were removed from promotional targeting entirely. A sunset policy was documented: any contact with no engagement for nine months drops to a monthly digest list and exits flow eligibility.

List-health measure	Before	After cleanup
Spam complaint rate	0.41%	0.07% by week 6
Seed-test inbox placement	89%	97% by week 6
Hard bounce rate	2.3%	0.4%
Dead addresses mailed monthly	4,100	0 (sunset segment)
Authentication	None	SPF + DKIM + DMARC (quarantine)

The sequence mattered. Sending win-back offers into a decayed domain would have poured the programme’s best creative into spam folders. Cleanup first meant the first flow send landed in inboxes at a 97% placement rate.

Phase 2 — Tiered Win-Back Flow Build (Weeks 3-5)

With a clean list and honest tiers, we built the flow architecture in Klaviyo — three lapse-tier flows, each with a separate high-value branch keyed off the RFM bands.

Tier 1 (30–60 days lapsed): the service check-in. No discount. A three-send sequence that reads as service, not selling — an order check-in (“How is the device performing?”), a category-relevant recommendation block (a phone buyer gets screen protection and audio accessories; an appliance buyer gets care tips), and a soft new-arrivals note. For big-ticket buyers, this tier deliberately waits out their long replacement cycle rather than burning an offer.

Tier 2 (60–90 days lapsed): the relevance bridge. Content-led sends built around the customer’s actual category history — buying guides, trade-in programme explainers, and review highlights from customers with the same purchase profile. The angle is re-establishing that the store remembers what this person owns.

Tier 3 (90+ days lapsed): the incentive ladder. Three sends spaced 14 days apart with an escalating offer — 5% capped at PKR 2,000, then 10% capped at PKR 3,000, then a final 15% capped at PKR 4,000. Caps protect margin on big-ticket carts, where an uncapped 10% on a PKR 150,000 order would give away more margin than the order is worth.

The high-value branch. The 1,900 high-value lapsed customers got a different spine: early access to new arrivals 48 hours before the newsletter, and a trade-in upgrade credit — a PKR 3,000 to PKR 8,000 assessed credit against their old device toward an upgrade. Trade-in framing converts the customer’s dormant device into a reason to return that a discount cannot replicate, and it matches how electronics buyers in Pakistan actually upgrade: old handset traded, balance paid across instalments or COD.

Every send carried a preference-centre link letting recipients choose product categories and cadence — 9% of recipients used it in the first month, and preference-setters unsubscribed at roughly a quarter the rate of the default segment. Suppression rules ran in both directions: anyone who purchased while inside a flow exited it immediately, and anyone who clicked without buying entered a 14-day cooldown before the next send, so the incentive ladder never stacked on top of a browser still deciding. Flow gaps were initially set at 7 days; Phase 3 testing pushed them to 14 and 21 days after unsubscribe data showed fatigue.

One structural gap surfaced during the build: roughly 30% of the database had last purchased at a physical outlet rather than online, and their email addresses — collected at the point of sale — had never been merged with order history. Deduplicating POS and online records against phone numbers added 2,300 identifiable customers to the correct tiers, including 210 high-value lapsed buyers the RFM model had missed entirely. For a retailer with outlets and a web store running side by side, a win-back programme is only as complete as the merge between those two worlds.

Phase 3 — Offer, Copy, and Cadence Optimization (Weeks 4-8)

The build was followed by six weeks of structured testing on the variables that actually move reactivation economics.

Bilingual subject lines. Roman Urdu subjects were tested against English across both value bands. On the standard tier, Roman Urdu lifted opens by 9 percentage points — a subject line like “Aap ka pasandeeda brand wapas laa rahe hain” outperformed its English equivalent in every round. On the high-value tier, English with the store name led narrowly, and the flows split accordingly.

Offer economics on the high-value tier. Three offers were tested head-to-head: a 10% capped discount, free next-day delivery, and a complimentary two-year extended warranty on the returned order. The warranty won decisively — it redeemed at roughly twice the rate of the discount code while costing about a third as much to fulfil on the median order. The psychology is specific to the category: a lapsed buyer hesitating on a PKR 90,000 appliance purchase is worried about breakdown risk, not price, and the warranty speaks directly to that worry. The discount framed the return as a transaction; the warranty framed it as a relationship.

Cadence and timing. Sends shifted to Thursday and Friday evenings, ahead of the weekend browsing window, and the flow gaps extended from 7 to 14 days after the first rounds showed unsubscribe rates of 1.1% on the tighter cadence. Timing the tier-two and tier-three sends to land just before the store’s August sale event produced the two highest-revenue send days of the programme.

Offer discipline. Any customer who reactivated entered a 60-day offer blackout so the ladder could not train buyers to lapse deliberately — a win-back programme that teaches discount-waiting behaviour has failed, whatever its dashboard says.

Measurement discipline. A 10% holdout of each tier was suppressed from the flows entirely to keep the headline honest. Held-out customers returned organically at 3.4%, which puts net incremental reactivation near 12.6 points against the 16% gross figure. Both numbers went into the final report; the difference between them is the cost of pretending a win-back programme causes every purchase it coincides with, and knowing that difference is what separates a dashboard metric from a decision-grade one.

Phase 4 — Measure, Sunset, and Compound (Weeks 8-12)

The final phase institutionalised the programme rather than leaving it as a campaign.

A cohort dashboard tracked reactivations by tier and value band, with revenue attributed on a 30-day post-send window and offer costs netted out to a true cost per reactivated order. The numbers that mattered to the finance function: 304 of the 1,900 high-value lapsed customers (16%) placed at least one order in the 90-day window, and 24% of those reactivated customers purchased again inside the same window — the early signal that reactivation was restarting relationships rather than renting one discounted order.

The programme was moved to always-on status with a quarterly review cycle: lapse-tier thresholds recalculated against fresh repurchase-gap data, the sunset policy executed each quarter, and warranty-led offers rotated as new product categories launched. Reporting settled into a weekly one-pager — flow revenue, reactivations by tier, deliverability health, cost per reactivated order — plus a monthly review where the incentive ladder was re-priced against current category margins. When the store ran a site-wide anniversary sale in month three, the win-back tiers were temporarily suppressed from discount sends so returning customers could not be double-discounted into a loss; the flows resumed afterwards with margin intact.

Win-back flows now sit between cart recovery and the loyalty programme as the middle layer of the store’s retention stack — the layer that decides whether a one-time buyer becomes a database asset or a churn statistic.

Final Results

Measured over the 90-day engagement window:

Metric	Before	Day 90
High-value lapsed customers reactivated	0 of 1,900	304 (16%)
Standard-tier reactivated customers	—	713 (5.8% of 12,300)
Win-back attributed revenue	PKR 0	PKR 9.3M in 90 days
Email share of monthly revenue	13%	21% (day-90 run rate)
Win-back flow open rate (blended)	11% (newsletter)	33%
Spam complaint rate	0.41%	0.07%
Cost per reactivated order	PKR 4,800 (new-customer CAC)	PKR 1,200
Reactivated customers buying again in-window	—	24% of the 304

The revenue composition is worth reading closely: the high-value cohort of 1,900 customers produced PKR 6.5M at a PKR 21,500 average order value, while the 713 standard-tier reactivations produced PKR 2.8M on accessory-led average orders near PKR 3,900. Sixteen percent of a 1,900-customer cohort out-earned a 5.8% response from a tier more than six times larger — which is the entire argument for value-band separation in win-back design.

What Made This Work

1. Lapse tiers came from the store’s own data. Accessory buyers and appliance buyers got different flows at different times because their repurchase gaps differ by months. A generic 90-day lapse rule would have spammed phone buyers mid-cycle and reached accessory buyers far too late.
2. Deliverability ran before the first send. Authentication and suppression repaired inbox placement in advance, so the programme’s best creative reached inboxes instead of spam folders. A win-back flow sent into a 0.41% complaint rate is a choreography rehearsal in an empty theatre.
3. The winning offer was a warranty, not a discount. It redeemed at twice the rate of an equivalent discount at a third of the cost, because it answered the actual hesitation of a lapsed electronics buyer — breakdown risk — instead of renting back their order with a price cut.
4. Roman Urdu subject lines matched how customers actually read. The 9-point open-rate lift on the standard tier was the single cheapest performance gain of the entire engagement — it cost nothing but the discipline to test it.
5. Offer blackouts protected the programme from itself. The 60-day blackout and incentive caps kept the ladder from training deliberate lapsing, which is the failure mode that quietly kills most win-back programmes within two quarters.

What Teams Can Apply

1. Define “lapsed” from your own repurchase gaps, not a rulebook. Pull your repeat buyers’ median days-between-orders by category and set tiers from that distribution. For electronics and appliance retailers, expect tiers to separate by product replacement cycle more than by customer type.
2. Clean the list before you mail it. Authenticate the sending domain, suppress twelve-month-non-engagers, and get complaints under 0.1% before the first flow fires. Every send into a dirty list makes the next send harder to deliver.
3. Test non-cash offers before defaulting to discounts. Extended warranties, trade-in credit, and delivery upgrades frequently beat percentage discounts on cost per reactivated order in categories where the buyer’s hesitation is risk rather than price.
4. Run bilingual subject-line tests from week one. Roman Urdu versus English costs nothing to test, and in Pakistani inboxes the lift is routinely the largest single gain available in lifecycle email.
5. Judge the programme on cost per reactivated order against your CAC. Reactivating a known buyer at PKR 1,200 versus acquiring a stranger at PKR 4,800 is the economic case for retention — track it openly and the win-back budget defends itself.