What does a supplier scorecard actually tell you that an ERP report does not?

A standard ERP report shows totals - how many orders, how many rejections. A scorecard shows patterns - which supplier causes rejections in specific months, whether a supplier's delivery time is getting worse over time, and which vendor is reliable for one material category but not another. In ForgeFlow, two suppliers account for all 4 quality rejections out of 27 inspections, and both failures cluster in the same month. A flat rejection report makes this invisible; a scorecard makes it impossible to miss.

How much do poor supplier choices actually cost?

The direct cost of a rejected batch is the material cost plus the rework or return logistics. The indirect cost is the production delay - a rejected incoming batch can stop a work order that is waiting for that material, which delays customer delivery. In manufacturing, one bad supplier month can cascade into 2-3 weeks of late orders. The scorecard quantifies this relationship between supplier performance and production outcomes, making the cost visible before it shows up as a customer complaint.

When should we stop using a supplier vs. give them a chance to improve?

The scorecard gives you an evidence-based answer. A D-grade supplier with declining performance across all three dimensions (quality, delivery, lead time) over 6 months is a consolidation candidate. A supplier who scores D on quality but A on delivery and lead time likely has a specific material or process issue worth investigating before replacing them. The data tells you which problem you have, so the conversation with the supplier is about facts, not impressions.

Can we build this without a dedicated data team?

Yes, if you can export purchase orders, delivery records, and quality inspection data from your ERP. Most manufacturing ERPs - SAP, Oracle, ERPNext, Odoo - produce these as standard exports. A developer familiar with Python can build the initial scorecard in a week from those exports. The ongoing maintenance is a monthly data refresh and a review of the output by a procurement manager - no specialist data skills required.

Will this change how we manage supplier relationships?

It changes the basis of the conversation. Instead of a procurement manager saying "we've had some issues with your deliveries lately," the scorecard makes it "you are scoring 0% on-time delivery over the last 6 months across 5 receipts, and your average lead time has increased from 12 days to 19 days." Suppliers respond differently to data than to impressions. For your best suppliers, the scorecard also gives you evidence to prioritise them for preferred supplier agreements.

Manufacturing Series Built on ForgeFlow

AI Procurement Assistant

Score every supplier on lead time, quality, and on-time delivery from 6 months of ERP data

The Problem

Your ERP already knows which suppliers are failing

Every GRN and quality rejection is already in your ERP - but scattered across three modules. No one sees it in one place until a production hold happens and the machine is idle. A supplier scorecard reads those signals before the next PO goes out.

4/27

GRNs rejected

All 4 failures concentrated in just 2 vendors

Paschim on-time rate

Every batch delivered late across 5 receipts

3.6x

lead time spread

Amrit 5.8d vs Vishwakarma 21d

Ranks 8 suppliers by composite score - quality, on-time delivery, lead time
Surfaces monthly rejection patterns: Sudarshan fails Feb + Apr, Paschim fails Mar + May
Squared quality penalty means a 40% rejection rate scores 36, not 60 - properly penalising quality risk
Generates reorder recommendations: A/B vendors auto-approved, D-grade blocked

The Data

4 ForgeFlow datasets, no setup required

Purchase Receipts27 records

GRNs with actual receipt date and PO link - basis for lead time and on-time calculation.

Quality Inspections27 records

1:1 per GRN. 4 rejected batches - Sudarshan in Feb + Apr, Paschim in Mar + May.

Purchase Orders32 records

Scheduled delivery dates for on-time rate calculation across 8 suppliers.

Suppliers8 records

Steel, stainless, tooling, and consumables. 5 active in the scoring period.

The Approach

3 dimensions, 1 score

Quality Score

(1 - rejection_rate)² × 10040%

Squared penalty: a 40% rejection rate scores 36, not 60 - reflects true production stop risk
Calculated from quality inspection records linked to each GRN
Purohit: 100 (0 rejections). Sudarshan: 36 (Feb + Apr hardness failures). Paschim: 36 (Mar + May dimensional failures)

On-Time Delivery Score

(on_time_count / total_receipts) × 10035%

Compares actual GRN date to scheduled delivery date on the purchase order
Each late delivery delays a work order start and idles downstream machines
Amrit + Purohit: 100%. Sudarshan: 60%. Paschim: 0% - every batch late

Lead Time Score

max(0, 1 - avg_lead_days / 30) × 10025%

Average days from PO date to GRN date, normalised against a 30-day cap
Shorter lead times reduce the safety stock buffer needed at reorder
Amrit: 81pts (5.8d avg). Purohit: 74pts (7.8d). Vishwakarma: 30pts (21d - acceptable for tooling)

Results

Full supplier ranking

Two A-grade vendors, two risk vendors, one sole-source tooling supplier. Click any row to expand the analyst note.

Supplier ScorecardForgeFlow - Jan to Jun 2026

Sort by

Score = quality^2 x 40% + on-time x 35% + lead time efficiency x 25%. Quality uses a squared penalty so a 40% rejection rate drops the quality score from 60 to 36, not 60. Click a row to see the analyst note. Data from 27 GRNs and 27 quality inspections across 6 months.

Paschim Steel: remove from approved vendor list

Score 28 (D). 0% on-time, 40% rejection rate, dimensional failures in Mar + May. Route all SS304 and AL6061 POs to an alternative until two clean quarters confirmed.

Sudarshan: conditional - require heat treatment certs for Feb + Apr orders

Score 51 (C). Hardness failures follow a Feb/Apr cadence - identifiable batch cycle issue, not random. Do not expand share until corrective action is verified.

Purohit + Amrit: consolidate volume here

Both score A-grade - 100% on-time, zero rejections. Preferred destination for any volume moved away from Paschim or Sudarshan.

Core Concepts

Key ideas in this playbook

Composite Scoring

Reduces vendor performance to a single number using weighted averaging
Weights reflect business impact: quality (40%) > delivery (35%) > lead time (25%)
Sudarshan example: (36 × 0.40) + (60 × 0.35) + (63 × 0.25) = 14.4 + 21 + 15.75 = 51
Grade thresholds: A ≥ 80, B ≥ 60, C ≥ 40, D < 40

Squared Quality Penalty

Linear quality score (1 - rejection_rate) gives 60 for a 40% rejection vendor
Squared formula (1 - rejection_rate)² × 100 gives 36 - correctly severe
A rejected batch stops production and costs 5-10x the material value in rework
The penalty must be non-linear to reflect real production impact

Monthly Rejection Pattern Detection

A flat rejection rate hides whether failures are random or systematic
pandas groupby on supplier + month surfaces the timing of failures
Sudarshan: Feb and Apr only - matches a batch sourcing cycle, not random variation
Systematic failures have a correctable cause; random failures indicate a broader quality problem

Reorder Recommendation Logic

Trigger: current stock < safety stock + (avg_lead_days × daily_usage)
A/B grade vendor: auto-create draft PO for buyer approval
C grade: create PO with quality flag for buyer to review before confirming
D grade: raise a resourcing task - find an alternative before placing any PO

Architecture

Taking it to production

Weekly scoring job

Pull latest GRNs and quality inspection records from ERP
Recompute composite score and A/B/C/D grade per supplier
Flag vendors with score drop > 10 points week-on-week
Identify items where primary supplier is C or D grade

Reorder alert logic

A or B grade: auto-create draft PO for buyer approval
C grade: create PO with quality flag for buyer review
D grade: raise resourcing task - no auto-PO
Lead time from scorecard sets the reorder point horizon

Run the notebook

The Jupyter notebook builds the full supplier scorecard from scratch using ForgeFlow's 6 months of procurement and quality data. Lead time analysis, rejection pattern detection, composite scoring, grade assignment, and reorder recommendations. No API key required.

Open in Colab