AUTOMATES 80% OF REPLENISHMENT DECISIONS

AI Replenishment Co-Pilot

Category: Forecasting

Tools: AI · Python

Scope: 847 SKUs · 5 Categories

Impact: 80% Auto-Decision Rate

Period: Jan 2025 – Mar 2026

Key Metrics

SKUs Monitored

847

↑ +12% YoY

AI Auto-Approved

80%

↑ was 0%

Stockouts Prevented

↓ 71% reduction

Avg AI Confidence

87%

↑ +6pp

Orders Generated

156

This month

WC Released

€284K

↑ vs manual

Service Level

98%

↑ from 94.1%

Avg Lead Time

8.4d

↓ 1.8d saved

Context

Problem

Manual replenishment at scale is failing

Planners were manually reviewing 847 SKUs weekly, spending 35 hours per planner on routine order decisions. Stockouts ran at 79 per quarter and excess stock tied up €1.2M in working capital due to inconsistent safety-stock policies.

Solution

AI Co-Pilot with confidence-gated automation

Built a Python-based AI engine combining XGBoost demand forecasts, dynamic ROP calculation, and a confidence scoring model. Orders above 85% confidence auto-approve; below that threshold, planners receive ranked recommendations with AI reasoning to review in minutes.

Findings

80% automation with 98% service level achieved

The system auto-approves 680 of 847 SKU replenishments weekly, cutting planner review time by 62%. Stockouts fell 71%, working capital released €284K, and service level climbed from 94.1% to 98.2% — validating the confidence-gate threshold.

AI Decision Feed

LIVE AI RECOMMENDATIONS — 12 PENDING

Updated: 2026-03-21 08:14

Live Policy Simulator

Adjust parameters to see real-time AI policy recalculation. Formulas: ROP = (AvgDemand × LeadTime) + SafetyStock · EOQ = √(2DS/H) · SS = Z × σ × √LT

Average Daily Demand (units)

65 units/day

Lead Time (days)

8 days

Demand Std Dev (units)

18 units

Target Service Level (%)

97.5%

Unit Holding Cost (€/unit/yr)

€8/unit/yr

Order Setup Cost (€/order)

€85/order

Dataset & Charts

Stockout Risk by Category

AI Confidence Distribution

Days of Stock vs Reorder Point

Decision Status Split

AI vs Manual ROP Comparison

Service Level: Actual vs Target

12-Month Service Level Trajectory

SKU Replenishment Table

SKU ID	SKU Name	Category	Supplier	Curr Stock	ROP	AI Rec Qty	Confidence	Days of Stock	Stockout Risk	AI Status

Company Advice

⚡ For Leaders & Analysts Deploying AI Replenishment

Start with the confidence gate at 85% — do not set it higher than 90% until you have 90+ days of approval history to validate model drift.
Always log override reasons. After 30 overrides, mine the pattern: if 70%+ share a root cause (e.g. "promotional uplift"), feed that signal back into the model as a feature.
AI replenishment works best when demand forecasting MAPE is below 20%. Deploy the Forecast Accuracy Scorecard (Project 18) as a prerequisite diagnostic.
Safety stock set by a static formula will underperform a dynamic ML safety stock model — especially for lumpy or seasonal SKUs. Quantify the gap before go-live.
Planner buy-in is the biggest failure mode. Present the system as a co-pilot, not a replacement. Share the "AI was right" vs "Human was right" scorecard monthly.

Key Takeaway: Automation rate is a vanity metric — service level is what matters. An 80% automation rate achieving 98.2% service level outperforms 95% automation at 96% service level every time. Tune for outcomes, not throughput.