ECommerce Case Study2
consistently appearing early in the customer journey for purchases that subsequently completed through channels that claimed last-click credit. $112K was reallocated away from TikTok and Pinterest in the following budget cycle, toward Google Shopping and Klaviyo email flows. Blended ROAS, measured on the deduplicated model, moved from 3.1x to 4.8x over the subsequent 12 weeks — not because the team spent more, but because they stopped over-investing in channels that had been borrowing credit from others.
The Challenge
Every Friday, the operations manager ran the weekly inventory review. She would log into Amazon Seller Central and export the FBA inventory report. She would log into the W almart Seller Center and export the fulfilment inventory file. She would log into the 3PL's web portal and export the on-hand stock report. Then she would open all three files in Excel and spend four hours attempting to reconcile them into a single view of inventory position. The reconciliation was never clean. The same product would appear under a different SKU code in Amazon versus the 3PL system. Amazon's available inventory figure excluded units in a reserved state that the 3PL counted as available. W almart's on-hand figure included units in transit that Amazon had not yet received. By the time the three systems were reconciled into something usable, the numbers were already a day old, the operations manager had lost most of her Friday afternoon, and the team was starting Monday with a picture of inventory that would shift materially over the weekend's order volume. The business had been stocked out on its three top-performing home organisation SKUs for an average of 11 days per month over the prior year. Each stockout triggered a loss of Buy Box position on Amazon that took between 4 and 9 days to recover after restock — because Amazon's algorithm deprioritised listings that had recently been out of stock. The financial damage was compounding: stockouts caused ranking drops, ranking drops suppressed organic visibility, and suppressed visibility meant the next restock period started with lower baseline sales velocity than before. Where inventory management was breaking down: • Three inventory systems — Amazon FBA, W almart Fulfilment, 3PL — with different SKU codes, different stock count definitions, and no real-time connection to each other • W eekly reconciliation consuming 4+ hours of the operations manager's time every Friday, producing data that was already partially stale by Monday • Stockout detection entirely reactive — the team discovered a stockout when the Amazon listing suppressed or a customer left a negative review, not 10 to 14 days before it happened • No demand-adjusted reorder point model — reorder triggers set as static minimum stock levels established 18 months ago, never updated to reflect seasonal velocity patterns or marketplace ranking dynamics
Excess inventory in the slow-moving pet accessories and personal care categories accumulating
untracked — $620K in stock that had not turned in 90+ days but had never been systematically
identified or flagged
No unified view of fulfilment performance across the three channels — SLA compliance, return
rates, and damage-in-transit rates tracked separately in each system with no comparison possible
The DataVine Solution
The Solution
- The foundation of every decision we made was a single product master. Before any pipeline or dashboard, we needed one agreed identifier per SKU that mapped every variation across the three platforms. That mapping work — correlating Amazon ASINs and FNSKUs to W almart Item IDs to 3PL internal part numbers — was the most painstaking part of the engagement and the most important. Building the unified inventory data foundation • Built real-time API integrations with Amazon Seller Partner API (pulling FBA inventory, stranded inventory, reserved units, inbound shipment status, and sales velocity data), W almart Seller API (pulling fulfilment inventory and order data), and the 3PL's integration layer (pulling on-hand, committed, and in-transit stock) into a centralised Snowflake warehouse on a 4-hour refresh cycle • Built a SKU master harmonisation model mapping every Amazon ASIN/FNSKU combination to the corresponding W almart Item ID and 3PL part number — producing a single product identifier that connected all three systems, with automated daily reconciliation checks to flag any SKU appearing in one system but absent from another • Built dbt transformation models standardising stock count definitions across all three platforms — one agreed methodology for "available to sell," one definition for "in-transit," explicit handling rules for Amazon reserved and stranded units, and documented logic so any analyst could trace any inventory figure back to its source • Integrated historical Amazon sales velocity data — pulling 52 weeks of weekly unit sales by ASIN, enabling the demand-signal model to distinguish seasonal patterns from genuine demand trends Building the inventory intelligence layer • Built a Power BI dashboard with four views: a cross-platform inventory health summary (stock levels, days-of-supply, and reorder status for every active SKU across all three channels), a stockout risk tracker (any SKU projected to reach zero before the next confirmed inbound delivery), an excess inventory report (all SKUs with 90+ days of stock on hand relative to current sales velocity), and a fulfilment performance comparison across Amazon, W almart, and 3PL • Built a demand-adjusted reorder point model incorporating trailing 13-week sales velocity, seasonality indices derived from prior-year data, Amazon ranking sensitivity (SKUs with a history of ranking drop after stockout given an extended safety stock buffer), and supplier lead times pulled from the procurement system — replacing the static minimum stock thresholds with a dynamic model that updated weekly • Built a stockout risk alert layer — any SKU projected to exhaust available-to-sell inventory within the next 21 days, given current sales velocity and confirmed inbound shipments, triggers an alert to the operations manager and the relevant buyer with the projected stockout date, the recommended reorder quantity, and the preferred supplier from the procurement system • Built an excess inventory identification model flagging all SKUs where weeks-of-supply exceeded 16 weeks at current velocity — surfacing each line item with the estimated carrying cost per week, the recommended clearance action (price reduction, Amazon Lightning Deal, W almart Rollback, or 3PL transfer), and the projected carrying cost avoided if actioned within 14 days
Operational Impact
12+ hours of weekly manual platform reconciliation eliminated — replaced by an automated
Monday morning briefing built from a single, methodology-consistent attribution model
$112K in annual ad spend reallocated from over-attributed channels to demonstrably
high-performing ones, based on deduplicated Shapley attribution
Blended ROAS improved from 3.1x to 4.8x over 12 weeks post-reallocation — without increasing
total ad spend
Four ad platforms unified under one attribution framework — zero conflicting conversion counts in
any weekly review
Influencer and affiliate contribution to first-purchase visible for the first time — enabling
data-driven partnership investment decisions
First-ever cohort analysis by acquisition channel — showing which sources were delivering repeat
buyers versus one-time purchasers
WHAT COMES NEXT
DataVines is now building a creative performance prediction model for the brand — using historical creative attributes (format, colour palette, copy length, offer type, product category) and channel-specific engagement signals to generate a predicted attributed-ROAS forecast for new creative concepts before they go live. The goal is to shift creative briefing from intuition-first to a data-informed process where the team enters production knowing which creative directions have historically outperformed in each channel. CASE STUDY 02 · E-COMMERCE · MARKETPLACE RETAIL · INVENTORY INTELLIGENCE & FULFILMENT ANALYTICS They Were Stocking Out on Their Best SKUs and Sitting on $620K of Dead Inventory — at the Same Time. A high-volume Amazon and W almart marketplace seller across five product categories had inventory data locked inside three separate systems that had never been connected. DataVines built the cross-platform inventory intelligence layer that finally made stockout risk and excess inventory visible in the same place, in real time.
The first Monday after go-live, the operations manager opened the dashboard at 8 AM instead of spending her Friday afternoon in Excel. Every SKU across all three platforms, every stock level, every days-of-supply figure — live, current as of 4 AM that morning, and built by no one. The stockout risk tracker immediately surfaced five SKUs that were projected to reach zero inventory within the next 14 days — none of which had triggered any alert in the prior week's manual review because the review had been based on aggregate category-level stock rather than SKU-level velocity analysis. Emergency reorders were placed the same morning. Four of the five SKUs maintained continuous Buy Box eligibility through what would have been their stockout window. The excess inventory report identified $620K of stock across 47 SKUs in the pet accessories and personal care categories that had not turned in more than 90 days. The procurement team had continued ordering these SKUs on their standard cycle because no one had flagged the velocity deterioration. Over the following eight weeks, 31 of the 47 SKUs were cleared through Amazon Lightning Deals, W almart Rollbacks, and targeted price adjustments. W arehouse carrying costs for those SKUs were eliminated.
16+ hours of weekly manual inventory reconciliation eliminated — replaced by a 4-hour-refresh
automated cross-platform inventory dashboard
$620K in excess and slow-moving inventory identified and actioned — reducing carrying cost and
releasing working capital
41% reduction in stockout incidents across all three channels in the first 60 days post go-live
Stockout detection moved from reactive (post-suppression) to proactive (21-day forward-looking
risk model)
Static reorder points replaced by a demand-adjusted dynamic model updated weekly —
incorporating sales velocity, seasonality, supplier lead times, and ranking sensitivity
First unified view of fulfilment performance across Amazon FBA, W almart Fulfilment, and 3PL —
enabling channel-level SLA and return rate comparison for the first time
WHAT COMES NEXT
DataVines is now building a marketplace pricing intelligence layer for the seller — pulling competitor pricing data across all active ASINs and W almart items on a daily basis, comparing against the seller's current prices, and generating repricing recommendations that balance Buy Box competitiveness against margin floor constraints. The goal is to move from manual weekly pricing reviews to a continuously calibrated system that responds to competitive movements within hours, not after the next Friday review cycle. CASE STUDY 03 · E-COMMERCE · HEALTH & WELLNESS · CUSTOMER RETENTION & LIFETIME VALUE ANALYTICS They Were Acquiring Customers at $58 Each. Most of Them Only Bought Once.
More Success Stories
Explore how we helped other brands in the E-Commerce sector.
ECommerce Case Study1
DATA-VINES Client Success Series E-COMMERCE & RETAIL Case Studies Four stores. Four data crises. Four transformations. How DataVines helped e-commerce brands, DTC retailers, and marketplace sellers replace gut instinct and disconnected spreadsheets with real-time intelligence — and star...
ECommerce Case Study3
Python Snowflake dbt Power BI Amazon SP-API · Walmart Seller API 3PL Integration Layer · Procurement API WHAT CHANGED The first Monday after go-live, the operations manager opened the dashboard at 8 AM instead of spending her Friday afternoon in Excel. Every SKU across all thr...