Skip to main content

Production Planning Assistant

Find the 80% on-time miss rate hiding in your ERP work order history and fix it with a lead time calibration model

The Problem

80% of work orders miss their target date

Production planners at FlowForge enter planned end dates manually - usually 1 to 7 days. Every item in the catalogue actually takes 18 days to produce. Customer delivery promises are broken before the work order is saved.

80%
work orders late
40 of 50 WOs missed plan across 5 months
14.2d
average overrun
Systematic - not random - across 7 item families
17x
worst planning gap
FG-YK planned 1 day, takes 18 days actual
  • All 9 item families complete in ~18 days actual - regardless of complexity
  • Planning estimates range 1-34 days - set by gut feel, not data
  • The Heavy Forging Press (36% of all job cards) sets the throughput floor
  • Calibrating WO templates to 20 days eliminates the 80% miss rate entirely

The Data

4 ForgeFlow datasets, no setup required

Work Orders50 records

Planned and actual start/end dates across 9 item families - the core dataset for lead time analysis.

Job Cards250 records

5 operations per WO linked to workstations - used to identify the Heavy Forging Press bottleneck.

Stock Entries190 records

Raw material consumed and finished goods produced per WO - material consumption patterns.

Sales Orders115 records

Customer delivery commitments - used to flag orders at risk from unrealistic lead times.

The Approach

3 steps from ERP dates to calibrated plan

1Measure actual vs planned lead time
  • Compute actual_lead_days per work order using actual_end_date - planned_start_date
  • Group by item_family to get average actual lead time per product group
  • Compare to planned_lead_days - the gap is the systematic planning error
  • ForgeFlow: planned range 1-34d, actual range 17-19d for all 9 families
2Find the capacity constraint
  • Count job cards per workstation to identify the bottleneck
  • Heavy Forging Press handles Die Forging + Flash Trimming per WO = 90 jobs total
  • 36% of all 250 job card slots sit on a single machine
  • The press queue explains why all items share the same ~18-day actual cycle
3Build the calibration table
  • recommended_lead_days = ceil(avg_actual + 2-day buffer) per item family
  • Output: one row per item code with a data-derived planned lead time
  • Load into ERP as WO template default - replaces manual date entry
  • ForgeFlow result: 20-day standard fixes 7 under-planned families, recovers 16-day buffer from FG-KP

Results

The planning gap by item family

7 of 9 families under-planned, 1 over-buffered, 1 correct by coincidence. Click any row to see the analyst note.

Sort by
Item
Planned
Actual
Gap
On-time

From 50 closed work orders, Jan-Jun 2026. All items average 18 days actual. Click any row for analyst note. Planned days reflect current WO template defaults.

FG-YK + FG-CS: 17-16 day gaps - worst in catalogue

Both planned with 1-2 days. Press queue affects them identically to complex machined parts. Fixed by 20-day standard.

FG-KP: 16 days of wasted buffer

Planned at 34 days, completes in 18. Always on time but wastes planning capacity. Reducing to 20 days recovers 14 days without affecting delivery.

FG-CR: accidentally correct - 20 days planned, 18 days actual

On time by coincidence, not calibration. The model makes this the standard for all families.

Core Concepts

Key ideas in this playbook

Planned vs Actual Lead Time
  • Planned lead time = planned_end_date - planned_start_date (what the planner entered)
  • Actual lead time = actual_end_date - planned_start_date (what production took)
  • Overrun = actual - planned. Positive means late, negative means early
  • Computing per item family reveals systematic errors vs random variation
The 18-Day Production Floor
  • 5 sequential operations share fixed equipment - induction, press (x2), inspection, machining
  • Heavy Forging Press is used twice per WO: Die Forging then Flash Trimming
  • All items wait in the same press queue regardless of BOM complexity
  • Simple forgings and complex machined parts both take ~18 days because the press sets the pace
Lead Time Calibration
  • Formula: recommended_days = ceil(avg_actual + safety_buffer) per item family
  • Safety buffer absorbs minor variation without over-padding (2 days used here)
  • Output is a lookup table: item_code -> planned_lead_days
  • Loaded into ERP WO templates to auto-fill planned dates instead of manual entry
Sales Order Delivery Risk Flag
  • Flag: if (delivery_date - order_date) < calibrated_lead_time then at_risk = True
  • Runs before a WO is created - catches impossible commitments at order entry
  • No ML needed - pure date arithmetic against the calibration table
  • Allows sales team to renegotiate or expedite before the shop floor is impacted

Architecture

Taking it to production

Weekly calibration job

Cron script queries last 90 days of closed WOs, recomputes avg_actual_lead_days per item family, updates the calibration table. Adapts automatically to seasonal throughput shifts.

WO creation validation hook

On WO save, checks planned_lead_days against calibration table. Raises a warning if the planned duration is >20% below the historical standard - planner can override but the error is visible before it hits the shop floor.

Daily sales order risk report

Joins open SOs to calibration table, flags any where delivery_date - order_date < calibrated_lead_time. Sent each morning to sales and planning so commitments can be renegotiated before escalation.

FAQ

Common questions

Why are work orders late even when the team is working hard?

Late delivery is usually a planning problem, not a production problem. If the planned completion date was set to 3 days and production genuinely takes 18 days, the order is late before anyone picks up a tool. The team cannot work faster than physics allows. In ForgeFlow, 80% of work orders miss their planned date because estimates were set using gut feel rather than historical production data. The fix is in the planning system, not on the shop floor.

How can lead time estimates be this wrong in a business that has been running for years?

Planning templates are usually set once - often by whoever configured the ERP - and rarely updated. As the business changes (new equipment, different supplier lead times, higher volumes), the templates stay the same. Nobody flags the error because "always late" becomes the normal state. The ERP data contains the evidence: every closed work order records the planned date and the actual completion date. Nobody has looked at the gap across 50 orders until this analysis.

What does an 80% late rate actually cost in customer terms?

Every late delivery is a customer conversation that should not have to happen - an expedite request, a delivery apology, or a missed contract SLA. In make-to-order manufacturing, late deliveries create downstream scheduling problems for the customer's own production. Over time, a pattern of late delivery erodes trust and gives customers a reason to dual-source or switch supplier entirely. The production planning model does not just fix an internal metric - it fixes the promise made to the customer at order entry.

Will this require changing the ERP or buying new software?

No new software required. The calibration model reads from your existing ERP export (closed work orders with planned and actual dates) and outputs a table: item code, recommended lead time in days. That table loads back into your ERP as work order template defaults - a configuration change, not a development project. The weekly recalibration job is a scheduled script that runs against the same export. Most businesses can implement this within 2-4 weeks.

How do we catch a bad delivery commitment before it reaches the shop floor?

When a sales order is entered, compare the customer's requested delivery date against the calibrated lead time for that product. If the customer expects delivery in 10 days and production takes 20 days, the commitment is impossible before a work order is even created. Flagging this at order entry - rather than discovering it two weeks later - gives the sales team time to renegotiate the date or expedite materials before the customer is already waiting.

Want this running on your business?

We implement these systems on your ERP, ecommerce platform, or existing data. Most projects go from data to production in 4-8 weeks.

Talk to us

Manufacturing Series

AI Procurement Assistant
Production Planning AssistantCurrent
Supplier Intelligence Dashboard
Production Workflow Agent
Playbook details
SeriesManufacturing #2
CategoryProduction / Operations
DifficultyIntermediate
Read time25 min
PlatformForgeFlow (ERPNext v16)
Tech stack
Pythonpandasnumpymatplotlib