How a Mid-Sized E-Commerce Brand Reduced Checkout Abandonment by 42% Using an AI-Powered Cart Recovery System

## Overview Wallet and Worn is a Pune-based direct-to-consumer (DTC) lifestyle brand that sells curated wallets, leather accessories, and everyday carry goods through its Shopify storefront and a rapidly growing Instagram Shop presence. Founded in 2019 by two university friends with a shared love of minimalist design and responsible manufacturing, the company had grown from a small Kickstarter-funded launch to a team of 38 employees and annual revenue of approximately 17 crore by early 2025. Despite strong brand affinity, a loyal Instagram following of 182,000, and a Net Promoter Score (NPS) routinely in the 72–78 range, Wallet and Worn was struggling with one of the most stubborn metrics in DTC e-commerce: cart abandonment. Industry benchmarks place average checkout abandonment at roughly 70%; Wallet and Worn's rate sat at 71.3% — just above the benchmark but translating, at their traffic volumes, to an estimated 2.4 crores in recoverable revenue per year. The founders had tried the usual remedies — exit-intent popups with a flat 10% discount, abandoned-cart email reminders at a fixed 1-hour and 24-hour cadence, and even a basic retargeting campaign on Meta. None of them moved the needle in any meaningful or lasting way. By late 2024, the team had begun to suspect that the problem was not the offer, but the approach itself: blanket discounting was eroding margin, one-size-fits-all timing was ignoring behavioural signals, and treating every shopper identically was leaving money — and goodwill — on the table. That suspicion set the stage for a collaboration with Webskyne's growth engineering team to reimagine cart recovery from the ground up. --- ## The Challenge Cart abandonment is one of the most complex problems in e-commerce because it is not a single problem — it is the aggregate symptom of dozens of micro-failures in the buyer journey. The Webskyne team began the engagement with a two-week diagnostic sprint to decompose Wallet and Worn's abandonment funnel and understand exactly where, how, and why shoppers were dropping off. The diagnostic revealed four compounding failure points. ### 1. The Just Browsing False Positive The existing abandoned-cart script triggered on every user who added an item to cart and then navigated away — regardless of whether they had even entered an email address. This flooded the recovery pipeline with low-intent signals and wasted messaging budget on people who had never indicated a serious purchase intent. ### 2. Rigid Timing Cadences The 1-hour / 24-hour email cadence had been configured as a static rule with no personalisation. Research shows that the optimal recovery messaging window varies dramatically by product category, cart value, time of day, and individual shopper behaviour. Sending a reminder two hours before bedtime to a user who typically shops at 2 PM is not just sub-optimal — it is actively counterproductive and contributes to unsubscribes. ### 3. Channel Mismatch Email was the sole recovery channel. Wallet and Worn's audience skews young (18–34) and is highly active on WhatsApp and Instagram DMs. For these shoppers, an email landing in a crowded Gmail promotions tab is easily missed — whereas a WhatsApp or SMS message arrives with a 98% open rate within minutes. ### 4. Discount Dependency The flat 10% code applied universally had trained a segment of shoppers to game the system: add to cart, wait for the recovery email, use the discount, abandon the next cart, repeat. This was hollowing margin and creating a discount-dependent brand perception that clashed with Wallet and Worn's premium positioning. --- ## Goals The engagement charter, co-signed by Wallet and Worn's co-founders and Webskyne's lead architect, established four primary goals. | Goal | Baseline | Target | Measurement Window | |---|---|---|---| | Reduce cart abandonment rate | 71.3% | <= 60% | 90 days post-launch | | Increase checkout completion rate | 28.7% | >= 40% | 90 days post-launch | | Maintain or improve margin per recovered order | 8,420 | >= 9,200 | 90 days post-launch | | Reduce recovery message unsubscribes | 1.8% | <= 0.6% | 60 days post-launch | A secondary, strategic goal was not explicitly revenue-based: to make every recovery interaction feel genuinely helpful rather than pushy, reinforcing the brand relationship instead of undermining it. --- ## Approach The Webskyne team adopted a four-phase methodology: Diagnose -> Design -> Build -> Optimise. Critically, they avoided a rip-and-replace approach — Wallet and Worn's Shopify storefront, Klaviyo account, and data infrastructure were left largely intact. The work centred on building a new intelligence layer that sat above the existing tools rather than replacing them. ### Phase 1: Signal Amplification (Weeks 1-2) The most important insight from the diagnostic sprint was that the event data Wallet and Worn was already collecting contained far richer behavioural signals than were being used. The team built a Behavioural Signal Index (BSI) — a composite score assembled from 12 distinct behavioural inputs collected during a single session: 1. **Cart value in rupees** — Higher-value carts receive earlier, more personalised attention. 2. **Number of unique items in cart** — Multi-item carts suggest genuine intent to shop. 3. **Interaction with checkout form fields** — Form field focus without submission signals active hesitation, not passive browsing. 4. **Product page dwell time** — Shoppers who spend more than 90 seconds on a product page score higher on purchase intent. 5. **Scroll depth on the product page** — Deep scrollers are reading descriptions, comparing dimensions, weighing the decision. 6. **Returning visitor flag** — Returning visitors who add to cart have a 3x higher recovery probability. 7. **Referral source channel** — Organic search traffic has higher intent than paid social traffic for this brand. 8. **Device type** — Mobile shoppers on 4G connections drop off more often at the payment step due to payment flow usability issues. 9. **Time of day and day of week** — Checkout completion rates vary significantly by temporal context. 10. **Previous purchase history** — Existing customers who abandon a new cart have the highest recovery probability of any segment. 11. **Product category velocity** — Items that have sold out or are in low stock create urgency without needing a discount. 12. **Geo-location delivery estimate** — Showing accurate delivery timelines in the recovery message dramatically improves conversion for location-sensitive shoppers. Each signal was assigned a weight determined through a cohort analysis of 4,200 historic recovery journeys. The BSI was then normalised to a 0–100 scale, with scores above 70 classified as High Intent, 40–69 as Medium Intent, and below 40 as Low Intent. This became the primary segmentation axis for all downstream messaging. ### Phase 2: Dynamic Timing Engine (Weeks 3-4) The next failure point to address was the rigid timing cadence. The Webskyne team built a Dynamic Send Optimiser (DSO) — a reinforcement learning model that recommends the optimal send time for each individual user on each individual channel based on their response history. The DSO was trained on six months of send-log and open-rate data across Wallet and Worn's Klaviyo and SMS providers. It learns that User A opens recovery emails reliably at 8:15 PM IST on weekday evenings, while User B consistently engages with SMS messages within 10 minutes of checkout abandonment if sent before 12 PM IST. The optimiser adjusts continuously — if a user's response pattern shifts (say, they start travelling for work and opening emails at 2 AM), the model adapts within a 72-hour observation window. The DSO also gates channel selection: if a user has never opened a recovery email but has clicked through a Meta retargeting ad, the engine routes the next recovery attempt through Instagram DM or WhatsApp instead of email — dramatically improving open and CTR rates. ### Phase 3: Content Studio and Personalisation Layer (Weeks 5-7) With signal quality and timing handled, the team turned to message content. The key insight here was that a generic "Did you forget something?" email performs worse than no email at all for high-intent shoppers — it feels patronising and suggests the brand is not paying attention. Conversely, a generic message can work well for low-intent shoppers who may genuinely need a nudge. The team built a six-variant content studio, with message tone, imagery, urgency framing, and offer type varying dynamically by BSI segment: - **High Intent (BSI >= 70):** No offer. The message focuses on product refinement — restocking alerts, delivery estimates, and social proof (customer reviews or "28 people also looked at this"). The framing is helpful and factual. - **Medium Intent (BSI 40–69):** A soft, value-add framing. Free shipping threshold reminder ("You are 349 away from free shipping"), or a product comparison table. No discount code to avoid training discount dependence. - **Low Intent (BSI < 40):** A gentle curiosity nudge featuring the most-viewed product from the abandoned cart, with low-pressure social proof ("This wallet was the 3rd most-loved product this month"). Each variant is further personalised with the shopper's first name, their city (for delivery estimate accuracy), and a product recommendation based on browsing patterns from the same session. ### Phase 4: Preview, Instrumentation, and Rollout (Weeks 8-10) The final phase involved building a comprehensive instrumentation layer — detailed event logging, A/B test scaffolding, and a real-time dashboard for Wallet and Worn's growth manager to monitor segment-level performance. Before the public rollout, the team ran a two-week shadow mode where the new engine ran alongside the old system, generating logs but not sending messages. This produced 18,000 simulated recovery events and allowed the team to validate that the BSI correctly categorised 89% of historic recovered orders with zero manual review, that the DSO's recommended send times outperformed the legacy fixed cadence by 34% in open rate in the shadow run, and that no user received more than three recovery messages in a 72-hour window, significantly reducing unsubscribe risk. Only after these validation criteria were met did the team activate the new system in production. --- ## Implementation ### Technical Architecture The solution was implemented as a serverless event pipeline hosted on AWS. The pipeline ingests Shopify webhook events (cart.create, checkout.started, checkout.completed) via AWS API Gateway, which validates, deduplicates, and enriches each event payload before passing it to a Signal Processor Lambda. This Lambda computes the Behavioural Signal Index for each abandoned cart and persists the result along with cart data to DynamoDB. A DynamoDB Stream triggers a Recovery Orchestrator Lambda that selects the optimal channel (Email via Klaviyo, SMS via Twilio, WhatsApp Business, or Instagram DM via Meta Graph API), composes the personalised message payload, and dispatches it. AWS infrastructure costs came to approximately 4,200 per month at 125,000 monthly active users — negligible relative to the recovered revenue. Development and build took ten calendar weeks with a team of one backend engineer, one ML engineer, and one growth strategist from Webskyne. ### Channel Configuration Recovery messages are dispatched across four channels, weighted by BSI segment and user channel preference. Email via Klaviyo covers all segments as the primary channel with a custom HTML template per segment variant. SMS via Twilio is reserved for Medium and High Intent users, triggered only through the DSO. WhatsApp Business is limited to High Intent users exclusively, providing a two-way conversation surface for resolving payment friction in real time. Instagram DM via Meta Graph API is used for Medium Intent users who arrived through Meta-referred traffic. --- ## Results The new cart recovery system was rolled out to 100% of Wallet and Worn's traffic on a Monday in early February 2025. The first 30 days produced results that exceeded all 90-day targets in the charter within four weeks. ### Primary Metrics (90-Day Performance) The 42% relative reduction in cart abandonment — from 71.3% to 41.4% — was the headline figure. But the margin improvement was perhaps the more consequential business outcome. By moving away from blanket 10% discount codes and replacing them with personalised urgency, social proof, and shipping incentives, the average order value of recovered orders rose from 8,420 to 9,680 — a 15% uplift. Combined, this meant the quarterly recovered revenue of 1.08 crore was not just higher in volume — it was structurally healthier. ### Secondary and Observability Metrics Beyond the charter metrics, the instrumentation layer captured several signal-quality indicators that confirmed the BSI model was working as designed. BSI Segment Accuracy stood at 89% — 89% of recovered orders landed in the BSI segment predicted at the moment of abandonment. The DSO Open Rate Lift delivered a 34% improvement in email open rates over the legacy cadence in the first six weeks. Segment-specific unsubscribe rates showed High Intent users at 0.12%, compared to 0.61% for Medium Intent and 0.89% for Low Intent. Channel Diversification produced 23% of all recovery conversions from non-email channels in the first quarter — a meaningful shift from a pre-launch baseline of under 4%. Customer NPS increased from a pre-launch mean of 75 to 82, with the most common verbatim positive comment being "They really seem to know what I want." ### Revenue Impact (Annualised) Month 1 produced 28.3 lakhs recovered — 4.2 times the baseline projection. Month 2 produced 35.7 lakhs recovered — 4.8 times baseline. Month 3 produced 44.1 lakhs recovered — 5.1 times baseline. Projecting conservatively at the 90-day run-rate, the annualised incremental revenue impact is estimated at 3.6 to 4.3 crores over the prior year's trajectory. --- ## Key Learnings The engagement produced eight lessons. ### 1. Signal quality matters more than message volume Sending more recovery emails does not solve abandonment — it trains users to ignore them. The BSI framework showed that sending one perfectly-timed, relevant message to a high-intent user outperformed three generic messages in driving conversion. Quantity, as it turned out, was not a strategy. ### 2. The no offer offer works It is deeply counterintuitive, but the highest-converting recovery segment — users with BSI >= 70 — performed best when shown no discount whatsoever. Personalised urgency, accurate delivery estimates, and genuine social proof converted at 18.4%, compared to 12.7% for the same segment with a flat 10% discount attached. Offering a discount to someone who already intends to buy not only leaves margin on the table — it can feel jarring and reduce trust. ### 3. Timing is not absolute — it is personal The DSO demonstrated that the best time to send a recovery message varies by as much as 14 hours between individual users. Acknowledging this and building a system that adapts to individual rhythms was responsible for approximately 40% of the open-rate improvement observed in the first six weeks. ### 4. WhatsApp does not replace email — it complements it Some teams viewed channel diversification as an either/or decision. Wallet and Worn's results suggest the better framing is guest list logic: WhatsApp for High Intent users who need to resolve a specific friction point, email and SMS for the broader audience. The WhatsApp channel contributed 14.2% of recovery conversions in its first full quarter despite being active for only 22 days in the initial rollout window. ### 5. Instrument everything, before you need the data The shadow mode ran for two weeks before production activation, during which no real messages were sent. This produced a baseline of 18,000 simulated events that allowed the team to validate the BSI model's accuracy and identify timing edge cases before any real users were affected. Teams that skip pre-production shadow runs are, in effect, debugging in production. ### 6. Recovery is not just about the message Perhaps the most impactful technical intervention in the entire project was not messaging-related at all: a subtle UX fix to the guest checkout flow that reduced frictions for new customers who chose not to create an account during checkout. This single change — a compact address autocomplete widget — reduced guest checkout abandonment by an estimated 6.2 percentage points on its own, before any recovery messaging was sent. Fix the funnel before you optimise the follow-up. ### 7. Margin protection is a feature, not a constraint The charter's decision to set a margin requirement for recovered orders — rather than simply maximising revenue — was quietly one of the best decisions in the project. By removing blanket discounts and instead optimising for relevant messaging that earns the conversion, the team protected margin while growing revenue. In e-commerce as in most businesses, the most durable growth is profitable growth. ### 8. Build the loop, not just the campaign The most lasting outcome of this project was not the headline numbers — it was the feedback loop built into the DSO and BSI systems. Every recovered order, every unopened message, and every unsubscribe feeds into the model. The system improves itself over time, requiring no manual campaign tuning or seasonal A/B test churn. Weeks after launch, the team convened a 30-minute monthly review instead of a weekly sprint meeting, because the model had stabilised beyond the level that manual intervention could meaningfully improve in the short term. --- ## Looking Ahead The Wallet and Worn team has already begun planning Phase 2: expansion of the BSI model into post-purchase cross-sell and re-engagement journeys, extension of the WhatsApp recovery agent into an AI-powered product recommendation assistant, and integration of the DSO with the in-store loyalty programme for omnichannel recovery journeys. For founders and growth leads facing a similar basket-abandonment challenge, the most important starting point is not a new tool or a discount strategy — it is a rigorous diagnosis of why people are leaving. The answers are usually already in your data. And once those answers are in hand, the right approach is rarely to shout louder. It is to listen more carefully — and speak at the right moment, in the right tone, with the right message. --- This case study was produced by the Webskyne editorial team in collaboration with Wallet and Worn's co-founders. All revenue figures are based on internal performance data shared with editorial approval. The AWS architecture, BSI scoring model, and DSO framework described above are part of Webskyne's proprietary Growth Engineering methodology.