Fashion E-Commerce — Recovering Inbox Placement After Aggressive List Growth

A French fashion e-commerce brand based in Paris grew its email subscriber base from 180,000 to 1.4 million in seven months through a combination of aggressive lead-capture tactics: full-page exit-intent pop-ups offering a 15% first-order discount, co-registration partnerships with three complementary lifestyle brands, and a high-visibility giveaway campaign offering a €5,000 wardrobe prize. The marketing team celebrated the growth as a strategic win. Six months later, the email program that had been the brand's best-performing revenue channel was functionally broken.

The warning signs had been dismissed for months. Gmail open rates had dropped from 24% to 9% across the expanded list. Klaviyo's deliverability dashboard showed a complaint rate climbing from 0.08% to 0.4% — well above the 0.3% threshold at which Gmail begins to treat a sender as spam by default. The brand's primary sending domain was rated BAD in Google Postmaster Tools. By the time our engagement started, Gmail inbox placement had fallen to 55%, Yahoo to 48%, and the Black Friday campaign — the single most important sending event of the year — was six weeks away.

This engagement started with a hard conversation. The marketing team's assumption was that an infrastructure migration — dedicated IPs, better authentication — would resolve the problem while preserving the full subscriber list. That assumption was wrong. Infrastructure expertise is not a workaround for poor list acquisition. It is an amplifier of good practice; it cannot rescue a domain whose reputation has collapsed because most recipients never genuinely wanted the mail. The recovery path required accepting that 84% of the list was unrecoverable and rebuilding the program from the engaged core.

1. Week 1: Immediate sending halt to toxic segments

   All sends to co-registration acquisition (440,000 addresses) and giveaway entrants (510,000 addresses) were stopped immediately. These two segments combined accounted for 68% of the list and more than 90% of the complaint rate. Continuing to send to them was compounding the reputation damage faster than any recovery could offset. Sends continued only to the 450,000 organically-acquired subscribers — the users who had signed up directly on the website for a discount or newsletter subscription.

2. Week 1–2: New sending domain provisioning

   A new sending domain (mail.brand.fr, distinct from the primary brand.fr) was provisioned on 2 dedicated IPs in Frankfurt, with SPF, DKIM (2048-bit), and DMARC configured from the start. The new domain had zero prior sending history, which is both a liability (no established reputation) and an asset (no damaged reputation to overcome). The strategy treated the new domain as a fresh brand-email identity — a transplant, not a patch.

3. Weeks 2–4: Re-engagement campaign to the engaged core

   A two-email re-engagement sequence ran from the new domain to the 220,000 subscribers who had opened or clicked in the previous 90 days. The first email was a plain-text re-introduction explaining that the brand was rebuilding its email program around engagement; the second offered an explicit opt-in to continue receiving updates. Of 220,000 targeted, 142,000 opened, 61,000 clicked the opt-in button, and 58,400 completed a double opt-in confirmation. This 58,400-subscriber segment became the seed list for the new domain's warmup.

4. Weeks 4–11: Gradual warmup and domain retirement

   Sending from the new domain ramped from 5,000 messages per day in Week 4 to 40,000 per day by Week 9. The old domain was wound down progressively: no acquisition-driven sends from Week 1, no marketing sends from Week 4, no transactional sends from Week 8, full retirement at Week 11. By Week 11, Google Postmaster Tools rated the new domain HIGH reputation.

Technical Assessment: Infrastructure Layers Examined

The pre-migration audit examined the three layers that matter for a reputation-recovery engagement: list quality (the root cause in this case), authentication configuration, and sending infrastructure. List quality was the finding that shaped every other decision.

List Composition Analysis

A 90-day engagement analysis segmented the 1.4M subscriber base by acquisition source and engagement history. The results were stark. Organic sign-ups (450,000 addresses) showed a 22% 90-day open rate — healthy by fashion e-commerce standards. Co-registration leads (440,000) showed a 4% open rate and a 0.8% complaint rate on their most recent campaign. Giveaway entrants (510,000) showed a 2% open rate and a 1.1% complaint rate. The complaint-rate differential was the critical signal: even if the acquisition segments were sent to carefully, they would continue to produce complaint rates incompatible with Gmail's reputation thresholds.

This kind of segmentation is a basic operational practice that the brand's in-house team had not performed before the engagement. The dashboard-level aggregate numbers (0.4% complaint rate overall) had obscured that the problem was concentrated in two specific acquisition sources. Without segment-level visibility, the team had been trying to "fix deliverability" in aggregate — an unsolvable problem framing.

Authentication Baseline

SPF was present but included two third-party services that were no longer in active use — legacy includes that had never been cleaned up. DKIM was configured with a 1024-bit key that predated current ISP recommendations of 2048-bit minimum. DMARC was at p=none, which meant that during the reputation crisis, the brand's domain was exposed to spoofing — an incident the brand had not experienced but was statistically likely within months given the heightened attention on its domain. Authentication was not the cause of the reputation problem, but advancing it was a necessary step before moving to enforcement.

Sending Infrastructure Review

The brand's existing ESP (a major North American shared-IP provider) was not the direct cause of the reputation problem, but the shared-IP model limited the recovery path in two ways. First, the old domain's reputation could not be rehabilitated on shared IPs because the IPs themselves had no stable reputation from the brand's traffic — reputation was a rolling aggregate of all tenants on the pool. Second, when the new domain went live, using the same shared pool meant inheriting whatever reputation damage the pool had accumulated from other tenants during the same period. The migration to dedicated infrastructure on the new domain gave the recovery program a clean slate — not because shared IPs are categorically worse, but because in a reputation-recovery context, the ability to isolate the brand's behaviour from every other sender's is essential.

Infrastructure Rebuild: Configuration Decisions

Three decisions defined the shape of the rebuild:

New domain, not same-domain repair. The marketing team initially resisted the new-domain approach, worried about losing the brand recognition of brand.fr in the sender address. The operational argument was stronger: a domain with a BAD Gmail reputation will take 6–12 months of clean sending to reach HIGH again, if it can be recovered at all. A fresh domain reached HIGH in 11 weeks. The brand identity in the sender-name field (Brand Name <hello@mail.brand.fr>) preserved recognition; the underlying sending domain being mail.brand.fr rather than brand.fr was imperceptible to recipients.

Confirmed opt-in as the floor for the new list. The 220,000 subscribers targeted by the re-engagement campaign produced 58,400 who completed double opt-in. The marketing team wanted to also include "soft-engaged" subscribers (those who had opened but not confirmed) on the new domain. This was declined. The value of the new domain's reputation depended on every subscriber on it being demonstrably interested. Introducing passively-engaged subscribers — even those with some activity history — would reintroduce the ambiguity that caused the original problem. The 58,400 figure was a reduction from 1.4M, but it was a reduction to a list that would not destroy the infrastructure it ran on.

Suppression registry shared across domains. Every unsubscribe, complaint, or hard bounce on either the old or new domain flows to a central suppression registry consulted before any send on either domain. This prevents a scenario where a subscriber who unsubscribed from the old domain is later sent to on the new domain because the new domain's list was compiled independently. The registry is redundant during the transition — by Week 11, the old domain is retired — but it prevents the class of error that would undo the recovery.

Operational Monitoring: What Changed Permanently

The recovery was not just an infrastructure migration; it was a program redesign. Three operational practices were embedded to prevent the pattern from recurring:

Complaint-rate thresholds as acquisition-source gates. Any new acquisition source must produce a 30-day complaint rate below 0.2% or it is retired. Co-registration was replaced with a referral program where existing engaged subscribers invited others; the referred subscribers showed a 0.04% complaint rate — lower than the organic baseline. Giveaways were discontinued as an acquisition channel; where the brand still runs occasional giveaways, the entrants are added to a separate sending domain used only for giveaway follow-ups, with no crossover to the primary marketing stream.

Monthly list-quality audit. The operations team reviews per-acquisition-source engagement metrics monthly. Any source with a declining 90-day open rate, or with a complaint rate trending above 0.15%, is flagged for investigation. This is the monitoring the brand had lacked during the original list-growth period; if it had been in place, the co-registration and giveaway issues would have been visible by the end of Month 2 rather than Month 7.

Engagement-based suppression with seasonal adjustment. Subscribers with no opens in 90 days are removed from the primary marketing stream and offered one re-engagement campaign from a reactivation segment. The 90-day threshold is relaxed during the fashion industry's natural quiet periods (mid-January, mid-summer for this brand's audience). This prevents accidental suppression of otherwise-healthy subscribers during periods when open rates drop industry-wide rather than brand-specifically.

Subscriber count is a vanity metric when engagement is not controlled for. A list of 1.4 million where most recipients never wanted the mail is not an asset — it is a structural risk that degrades the reputation needed to reach the 15% who do want it. Infrastructure expertise cannot fix this; it amplifies whatever list quality it is given. The recovery for a reputation-damaged domain is often not "rehabilitate the domain" but "transplant the program to a new domain and commit to the practices that prevent recurrence".

The acquisition tactics that drove the original list growth — co-registration, giveaways, aggressive pop-ups with vague consent language — are not categorically broken. What failed was using them as cold acquisition sources to people who had no prior relationship with the brand. When re-examined as warm-introduction channels to existing audiences, some of the same tactics became legitimate again. The infrastructure work made the recovery possible; the acquisition-strategy change made it durable.