Coloplast Volume Manufacturing · Cartago, Costa Rica

Cartago Supply Chain: From Spreadsheet to Decision System

A site proposal to close a planning gap the global ERP will not close — and to package the answer so the other four sites can use it.

Michael BernaldHead of Supply Chain
Volume ManufacturingLa Lima Free Zone, Cartago
August 2026Proposal for Global Operations
The weekly call→ the gap the ERP will not close→ what is already built→ what it saves→ a 90-day ask
01 The operating reality

The call we make every week

Material planning at this site is not a reporting problem. It is a prediction problem with a twelve-week penalty for being wrong.

~800
SKUs reviewed each cycle
Every one is a planner's judgement call
~12 wk
Replenishment lead time
Most suppliers in Europe
5
Coloplast sites we can pull from
Hungary ×2 · Portugal · China · Costa Rica
Detect a shortage outside the lead-time window and it costs nothing. Detect it inside the window and only three levers are left — all three expensive.
Lever 1
Air freight
Premium rate, on every affected line.
Lever 2
Pull from another plant
Covers us here, opens the same gap there.
Lever 3
Take the line impact
The outcome nobody wants to report.
Site material planning, Cartago
02 How we do it today

JPD920 — the shortage list

This is not disorder. It is a well-designed system that has reached its ceiling.

Inputs
  • Item master · work orders · BOM pick
  • Open purchase orders — already from the data lake
  • Requirements & forecast · material prices
  • Material expiry dates
  • Intercompany — still manual out of E1
Process
  • One line per item
  • Inventory + in transit + open orders
  • − weekly consumption
  • Rolled forward ten weeks
  • Alternates on a separate tab — by hand, by each planner
Output
  • Weekly balance · covers column
  • Urgencia N+1 / N+2
  • Balance Negativo 10 Semanas
  • Summary Unicos vs Alter
  • KPI by Planner · HeatMap
The discipline is real. Ten-week forward visibility, alternates mapped, per-planner performance already measured. Very few sites can describe their shortage process this precisely. That is the foundation everything below is built on.
JPD920 shortage_list_CostaRica.xlsx · site material planning
03 Where it breaks

Five named failure modes — not five complaints

1

Volume

~800 SKUs reviewed by hand, every cycle. Attention is the scarce resource, and it is spread evenly across items that do not need it.

2

Person dependence

Each planner runs the analysis their own way. The method lives in people, not in the process — so it cannot be reviewed, improved, or handed over.

3

Fragility

One mistyped cell propagates through the formulas, and nothing flags it.

4

Silent omission Costliest of the five

A material that was never added to the sheet is invisible. There is no shortage alert for a row that does not exist — and it is the one failure mode that can be eliminated outright.

5

Alternates live on a separate tab

The substitution that would solve a shortage has to be found by hand, at the moment of most time pressure.

Site material planning, Cartago
04 The structural gap

The gap the ERP will not close: material expiry

E1's MRP does not consider material that is expiring. It will not tell me the material is going out of date and that I will not have it. Material planning, Cartago — August 2026
What the MRP countsas coverage
Consumed before expiry
Expires on the shelf
What the line canactually consume
Consumed before expiry
the shortage nobody is warned about
Stock that expires before it is consumed is not inventory. It is scrap that is still counted as coverage — and today the expiry date is pulled separately and patched in by hand.
A global ERP roadmap will not close a gap that is specific to medical devices — which is precisely why the site should close it.
Oracle JD Edwards EnterpriseOne (E1) · site material planning
05 What is already in place

Cartago is not asking for a platform

The expensive part of this problem has already been paid for. This proposal capitalises on infrastructure that exists today.

Working today
  • Databricks data lake in production — open purchase orders are sourced from it, not rekeyed
  • Power BI and Cognos deployed and in daily use
  • Veeva for change control
  • Access is provisioned locally in minutes, by the team itself
  • E1 stays the system of record. Nothing here replaces it or writes to it
Honestly, not yet solved
  • Intercompany is still pulled manually out of E1
  • An ageing OLAP cube sits alongside the lake and cannot be queried flexibly
  • Missing tables in the lake are queued behind global IT
  • Two disconnected sources feed one weekly decision
  • No single place where coverage, expiry and alternates are seen together
Budget consequence: the data platform spend is already committed and running. What follows is an application on top of it, not a second platform.
Site IT footprint, Cartago
06 The actual constraint

What is missing is not technology

People here already build their own tooling around the spreadsheet. Individual capability is not the gap.

Time
The department has no analyst function, and analysis is not what it exists to do. The work happens in the margins of the planning job.
Consequence: the analysis competes with the weekly cycle, and loses.
Ownership
Personal workflows, built by whoever needed them. Nothing is anybody's job to maintain, so nothing outlives the person.
Consequence: the method leaves when the person leaves.
Continuity
Additional data-lake tables sit in the global IT queue. Progress depends on a queue the site does not control.
Consequence: the site's pace is set somewhere else.
The unit of work. Turning personal workflows — one planner's method, on one machine, understood by one person — into a standard the whole team runs. That is not something the department can do in the margins of the planning job.
The constraint is small, nameable and fixable. That is the whole reason the ask at the end of this deck is narrow.
Supply chain organisation, Cartago
07 Proof, not description The turn

This already exists: coverage in one screen

Everything up to here is what we do by hand. This is the same weekly decision, computed — built against the real column structure of JPD920.

Shortage coverage view
Working prototype · all data shown is synthetic
  • Of ~800 materials, the ones that demand action this week
    Ranked, not listed. Attention goes where the reaction window is closing.
  • Coverage week by week
    The projected balance for each material, ten weeks forward, against the real lead time.
  • Shorts the MRP does not see
    Because the material expires before it can be consumed — counted as stock today.
  • Premium-freight exposure
    What each projected shortage will cost if it is caught inside the window.
Prototype built on the JPD920 column contract · synthetic data
08 The approach

Levels of autonomy — a ladder, not a leap

Each rung has its own validation gate. No rung removes the qualified person from the acceptance path.

L0
Manual spreadsheet — where we are today
Today
L1
Automated ingestion, one source of truth — no rekeying, no broken lookups, no silent omissions
Pod 1
L2
Exception detection and a ranked shortage list — of 800 items, the 30 that need attention this week
Pod 1
L3
Scenario simulation and recommended action — alternate material, inter-plant pull, or expedite
Later
L4
Supervised draft purchase order in E1 — prepared by the system, approved by the planner. One touch, never zero
Later
Funding request covers L1 and L2 only. L3 and L4 are shown so the direction is legible, not so they are approved today.
Proposed capability ladder
09 Being precise about AI

Where the intelligence actually sits

Roughly 80% of the value here is ordinary software engineering. Saying so is what makes the remaining 20% credible.

Sources Ingestion · plain software Coverage engine · plain software Where AI earns its place — the 20% Output Oracle E1 Databricks lake Expiry dates Intercompany manual today Ingestion layer one column contract no rekeying every material, always Coverage engine weekly balance, ten weeks out covers, adjusted for expiry alternates in the same model 01 Omission & anomaly detection the row that is not there 02 Scenario simulation alternate · inter-plant · expedite, costed 03 Root-cause narration why this item is going short Planner decision queue ranked by reaction window with the reason attached with the options attached A person decides Ordinary software engineering — ingestion, storage, interface, access control. ~80% of the build. Machine judgement, only where a formula cannot do the job. ~20%.
Precedent for the shape, not the number. Sanofi's supply-chain tool forecasts shortages, narrates a root cause, proposes an action — and a human acts.
Sanofi, publicly reported · pattern cited as evidence the shape works, not as a result forecast for this site
10 The deliverable, concretely

What the planner sees on Monday

Not a dashboard. A prioritised decision queue — one operational view of the site's material position, answering six questions in order.

The questionWhat the view answers
Where do I start? Of ~800 materials, the ranked handful that will go short inside the reaction window.
Why this one? Week-by-week projected balance, and the specific driver — demand, a late order, or expiry.
How long do I have? Coverage in weeks, measured against the real lead time for that supplier.
What are my options? Mapped alternate, inter-plant availability, or expedite — surfaced on the same row.
What did I miss? Materials absent from the list that should be on it.
What is it costing us? Premium-freight exposure attached to each projected shortage.
The covers column and the HeatMap tab we maintain today are the seed of this view. It is the same idea, computed rather than typed.
Proposed Pod 1 output
11 Boundaries

What this site owns — and what it does not

This proposal stays inside site authority. It consumes from the rest and delivers into it at defined edges.

Inside this proposal — site authority

  • Site master planning
  • Production and raw material planning
  • NPI
  • Non-BOM purchasing
  • Import / export
  • Warehouses
  • Logistics

Outside it — named owners, delivered to at defined edges

  • Global S&OP and supplier planning — Denmark
  • Direct spend negotiation — global category team
  • ERP and data platform — global IT
  • R&D and PLM — Humlebæk
  • MES programme — New Digital Wave
  • Validation — Site Quality + Global QA
This proposal does not ask to govern anything in the right-hand column.
Scope definition
12 Why this travels

Pods: solve it once in Cartago, ship it to four sites

15–20% → 20–25%
Cartago share of global volume by 2030
70–75% → 60–65%
Hungary share over the same period
The method has to survive a site that is growing — and be portable to the sites handing the volume over.
01
Column-level input contract
Not a live connection
02
One job only
Three jobs get negotiated, not transferred
03
Its own assurance file
The receiving site inherits the argument
04
No site hard-coding
Calendars, units, plant codes are configuration
05
Runs on what every site already has
No new infrastructure to buy
The expiry gap exists wherever E1 runs. Proposed sequence: Cartago → Portugal → Hungary ×2 → China. Portugal first — it is ramping now and has no legacy spreadsheet to displace, so adoption cost is lowest.
“From this department we export technology, simply because we solved it here first.”
Volume allocation: Capital Markets Day 2025 · Precedent: J&J DePuy Synthes Suzhou scaled standardised digital solutions across J&J sites
13 The business case

The money: air freight is the number

This pays for itself in avoided premium freight long before it pays for itself in planner hours.

12-week lead time→ detect inside the window→ air, inter-plant pull, or line impact⇒ every week of earlier detection converts air to sea
Annual premium-freight saving = (expedites/year attributable to late detection)  site data needed × (air cost − sea cost per shipment)  site data needed Inventory reduction = safety stock recalculated on measured lead times − safety stock on material-master lead times
Lever 1 · fastest
Air converted to sea
Lever 2 · largest
Safety stock on measured lead times
Lever 3 · weakest
Planner hours released
Both numbers are deliberately blank. They exist in our own expedite history and nowhere else. A benchmark borrowed from another company makes this slide easy to dismiss; our own figures make it hard to argue with.
GOP7: “gross margin accretive… capex and inventory reduction” · Group inventories DKK 3,942m, working capital 25% of sales vs Impact4 target ~24%
14 Quality posture

Light assurance by design

The system is advisory by construction. Three properties decide the whole validation argument:

Does not set process parameters Does not determine product acceptability Is not the record
FDA Computer Software Assurance — final, Feb 2026
Names AI/ML tools inside the risk-based framework. The guidance's own worked example is an ERP: the same system is not high process risk when a qualified person reviews the output, and high risk when nobody does.
ISO 13485 §4.1.6
Validation effort must be proportionate to risk. That is the text of the standard, not an interpretation of it — and it is the clause this posture is built on.
Intended-use statement, day one→ Risk-based assurance plan→ Site Quality reviews as it is built→ Assurance file travels with the pod
Stated precisely: this is light assurance, not full CSV. It is not a claim that validation does not apply.
FDA CSA final guidance (Feb 2026) · ISO 13485:2016 §4.1.6
15 The road after this

After Pod 1

There is a roadmap, and it converges with the corporate programmes as a consumer, not as a competitor.

Pod 2 · candidate
Warehouse coverage and replenishment
11,000 pallet positions, ~5,000 of them external. The same coverage logic applied to a different object.
Needs: warehouse movement history in the lake.
Pod 3 · candidate
Non-BOM spend analytics
Squarely inside site authority. No overlap with the global category team, and no negotiation mandate implied.
Needs: nothing outside the site.
Pod 4 · candidate
Logistics
The least instrumented area we run today, and the one where a first measurement is worth the most.
Needs: carrier and transit data we do not capture yet.
Long-term destination: a single material view across the five sites, co-owned with Global Operations — built by consuming the global platform, never by duplicating it.
Each pod is a separate decision. None of them is being asked for today.
Indicative sequence — not part of the current request
16 The request

What I am asking for

A commitment that is narrow, measurable and reversible.

Scope
Pod 1 — material coverage and shortage detection, including expiry
L1 and L2 on the ladder. Nothing beyond it.
Terms
90 days, with defined exit criteria
An explicit go / no-go before any Pod 2. Owned by this organisation.
Two things needed from corporate
1 · A budget line under GOP7
2 · A named global IT contact for the missing data-lake tables
How we will know it worked
Expedites avoided · share of shortages detected at ≥12 weeks · zero silent omissions · planner hours released (last, and the weakest of the four)
No public case exists for supply-chain AI in a Costa Rican medical device plant. That is the opportunity, not the risk — whoever does it first sets the standard the other sites adopt.
Michael Bernald · Head of Supply Chain · Coloplast Volume Manufacturing Costa Rica
o = index