Success Story | AI Demand Planning
How Made By Mary reached zero stockouts in 6 weeks
12-month SKU forecasts aware of promotions and retail timing.
Multi-level BOM demand resolved to components with weeks-of-supply logic.
Grouped draft purchase orders written straight back into Fulfil in one click.
From kickoff to a live planning system
Unplanned stockouts after go-live
Faster to create and approve purchase orders
Better category forecast accuracy (weighted MAPE)
Demand Planning at Scale
Made By Mary is a direct-to-consumer jewelry brand. It pairs made-to-order personalization with ready-to-ship collections, while seasonal gifting, influencers and promotions drive sharp demand spikes.
Scale raised the bar on planning. Thousands of SKUs and overlapping raw materials pushed beyond what spreadsheets could coordinate, and sales history sat apart from promotion effects, so plans leaned on judgment more than data.
Rudder Analytics built AI demand planning on the brand’s Fulfil ERP data. The system produces retail-aligned 12-month SKU forecasts, consolidates multi-level BOM demand, and writes grouped draft purchase orders back into Fulfil.
Strain Across the Planning Chain
Manual spreadsheets could not keep pace with a fast-growing product mix. A planning transition added urgency to building something durable and repeatable.
Fragmented, Off-Cycle Forecasts
Sales history and promotion effects sat in separate spreadsheets, creating timing errors and scattered SKU-level forecasts.
Cash in the Wrong Stock
Buying inputs without consolidated demand tied up cash and reduced capacity to buy newer, faster-moving SKUs.
Double-Counted Components
One component fed several finished SKUs. Buyers placed duplicate orders and reconciled by hand across BOM levels.
Manual PO Creation
Creating and grouping purchase orders was slow and manual, with limited traceability once orders were placed.
Buys and Budget Apart
Purchase quantities and budget allocations were set separately, so buys did not always reflect cash priorities.
A Planning Handover
A planning transition left the process without a dedicated owner, adding urgency to a durable, documented system.
AI Planning on Top of Fulfil
Fulfil stayed the single source of truth, and the planning layer sat above it without a middleware rewrite. Forecasts, constraints and inbound stock feed one buy engine that drafts purchase orders straight into the ERP.
ERP & Sales Data
Fulfil BigQuery schema and transactional exports, refreshed on schedule with ad-hoc pulls on campaign days.
Predictive Engine
12-month SKU forecasts with promotion flags and 4-4-5 retail-week markers, scored on holdout validation.
Auto PO Into Fulfil
BOM-resolved demand, WOS logic and an LLM agent produce grouped draft POs approved in one click.
One Planning Dataset
Fulfil’s BigQuery schema and exports feed a central dataset with scheduled ETL, campaign refreshes and logged validation.
Retail-Timed Forecasts
Monthly 12-month SKU forecasts use 36–48 months of history, promotion flags, 4-4-5 markers and per-SKU confidence scores.
Plan-to-PO Quantities
Category and SKU demand come from recent contribution weights, with saleable and purchasable demand plus scoped overrides.
One Line Per Component
Finished-good demand rolls into multi-level input needs and one consolidated buy quantity per component.
Protected Working Capital
Quantity on hand, open POs, expected inventory and target Weeks-of-Supply determine what is needed without overbuying.
One-Click, Auditable POs
An AI agent blends forecasts, safety stock, lead times, MOQs and inbound POs into vendor-aware recommendations logged in Fulfil.
Impact in the First Quarter
Across the pilot and first three months live, early warnings and automated recommendations produced zero unplanned stockouts. Working capital came back, and the team scaled without adding planning headcount.
The live system delivered zero unplanned stockouts, improved category forecast accuracy by approximately 25%, and made PO creation and approval approximately 80% faster.
| Metric | Before | After | Impact |
|---|---|---|---|
| Deployment time | Spreadsheet planning | 6 weeks to live | New capability |
| Unplanned stockouts | Recurring | 0 | Eliminated |
| Category forecast accuracy | Spreadsheet baseline | ~25% better | +25% |
| Average on-hand inventory | Baseline | ~20% lower | −20% |
| PO create & approve time | Manual, per-line | ~80% faster | −80% |
| Planning & reconciliation | Baseline | ~60% lower | −60% |
Made By Mary now plans on its own ERP data, promotion-aware and fully auditable. Buys track budget and cash, procurement cycles run shorter, and in-stock performance holds as the product mix keeps growing.
Engagement summary · Rudder Analytics
Data and AI for Brands Outgrowing Their Stack
Rudder Analytics builds that platform—from the first pipeline to reporting a board can act on.
