Selected work: entrepreneurial project September 2020 to March 2022

DD
Analysis

MRP and DDMRP inventory simulation and optimisation

Supply chains are complex, and controlling inventory without taking on risk has become critical. Yet we keep thinking locally.

DD Analysis made a cross-functional view possible, beyond company silos: simulate the whole network before touching real parameters, then compare inventory policies to find the best trade-offs between availability, stock and risk across upstream, downstream and internal operations.

Sold Optimisations invoiced to paying clients
Built C# simulator, algorithms and web interface
Project closed March 2022

The problem is a network problem

Conceptual diagram, without real data

Supplier

A lead time or a received quantity drifts from the plan.

Plant

Bills of materials and network levels absorb or pass on the gap.

Warehouse

Cutting stock here may simply move the problem elsewhere.

Store

Demand varies, and peaks travel back up the chain.

01
Decision
lab

A decision lab for the supply chain

I built a C# simulator able to reproduce, day after day, how a multi-level supply chain behaves.

Four capabilities of the simulator

01 Represent products, bills of materials and the different levels of the network.
02 Simulate demand, forecasts, consumption peaks and delays.
03 Vary lead times and the quantities actually received.
04 Compare several inventory policies.

How it was delivered

A web interface let users upload their data, run the calculations and retrieve the results.

The simulator was therefore not an analysis script: it shipped as a tool usable within an engagement.

DD Analysis diagram: on the left, the configuration levers (DAF, buffer profile, ADU horizon, peaks, buffer type, variability and lead-time factors); in the centre, the simplified buffer mechanics with ADU, green, yellow and red zones, orders and daily stock; on the right, the tracked objectives (workload, service level and average stock). Inspect
LeversBuffer mechanicsTracked objectives
DDMRP model · levers and objectives

DD Analysis connected the configuration levers to the buffer calculation rules, then to the tracked objectives: service level, average stock and workload.

Original diagram from the project.

02
Decision
evidence

Comparing before changing the real system

On a real network, a parameter is rarely set once. Simulation showed the effect of a setting before applying it.

DD Analysis archive: for two references, the simulated inventory trajectory with the initial ADU 40/20 parameters and with new ADU 30/30 parameters, stockouts circled, and the resulting service and stock in each case Inspect
Simulation · impact of parameter settings

Two seemingly close settings can produce very different inventory trajectories and stockouts. The simulator made those effects comparable before changing the real system.

Project archive, with data and curves kept as they were.

03
A product
on the market

A product that was genuinely sold

For eighteen months I defined the value proposition, built the product and its algorithms, prospected, sold the solution and delivered the engagements with paying clients.

≈ €3,000

Price charged per optimisation

An accessible price, but too low against the work required and the potential value for clients.

≈ 20%

Average inventory-reduction target

An average target, alongside fewer delays. Not a universal guarantee, nor a certified result for every client.

That price made the solution easy to reach. In hindsight I had undervalued the offer: I wanted it accessible to as many as possible, but that very idealistic positioning could not sustain a business.

A useful solution also has to be priced and sold in line with the value it produces.

04
Analysed
case

Results and analysed case

A complete analysis deck exists in the project archive. Its figures are kept here as an analysed case, distinct from the average target of around 20%.

−14%
inventory in the analysed case
+0.3 pt
service level in the analysed case

An analysed case kept with its limitation: its exact context still needs documenting.

These figures are neither an average nor a promise of results.

Three statuses not to be confused

Target
≈ 20% inventory reduction aimed for on average.
Scenario
The figures of one analysed case, in its own context.
05
Planning
families

Grouping products by how they actually behave

In some planning processes, simply reorganising the planning families, without changing anything else in the system, can cut inventory by 5 to 10% at constant service level.

So I extended DD Analysis with a clustering tool that goes beyond ABC/XYZ classification: it analyses product characteristics and consumption history to identify which products should be managed in a similar way.

The real trade-off

Optimising product by product can give the best theoretical result, and create hundreds of rules that nobody can maintain day to day.

The goal is not only the mathematical optimum, but a recommendation the teams can actually apply.

Four terms of the trade-off

01 The inventory gains obtained.
02 The service level preserved.
03 The number of families to manage.
04 How simple it stays to maintain for the operational teams.
Original infographics, 2020-2022
DD Analysis infographic: a baseline scenario with 12 families managed by ABC/XYZ compared with a proposed scenario of 20 families managed by ABC/XYZ, life cycle and lead time. Results: stock €520k against −8% at €480k, service 97.7% against +1 point at 98.7%, sensitivity 124 against −15% at 105. Inspect
Segmentation compared

In this scenario, a new family segmentation cuts inventory by 8%, improves service by one point and reduces the system’s sensitivity by 15%.

An analysed scenario, not a guaranteed average

DD Analysis infographic: for 1.5k products and €410k of stock, the service level obtained by number of families: 9 families 96.7%, 10 families 96.8%, 15 families 97.6%, 20 families 98.2%, 25 families 98.5%. Inspect
Number of families

More families improve performance step by step, but also increase the maintenance load for the teams.

DD Analysis output: distribution of products by cluster across several behavioural dimensions, including lead time, quantity and value of the ABC classes, XYZ class and coefficient of variation. Inspect
Multidimensional clustering

Products are grouped across several behavioural dimensions, beyond their ABC/XYZ class alone.

The clustering took product characteristics and their history into account, then proposed several family counts to compare.

06
From expertise
to offer

From domain expertise to an offer that sold

DD Analysis was neither an academic exercise, nor a mere case study, nor development alone. The project brought six dimensions together.

01 Turning supply chain expertise into a usable product.
02 Modelling complex systems and comparing several scenarios.
03 Writing the algorithms and the application in C#.
04 Translating a mathematical optimum into recommendations teams can maintain.
05 Building an offer, presenting it, selling it and delivering the engagements.
06 Drawing the lessons of a business model that created value but was not profitable enough.