Skip to content
Shyenova

Supply Chain & Analytics

Digital Twin

Most things sold as a digital twin are a dashboard with a 3D render. The test is whether it can reproduce a quarter you have already lived through.

There is a genuine and valuable idea underneath the term, and it is almost entirely obscured by how the term is sold. A digital twin is a model of your operation, fed by your data, that you can run a decision through before you commit to it. Change the replenishment policy in the model, run a year of demand against it, and see what it does to service and working capital, without changing anything in the real operation and without waiting two quarters to find out.

What gets sold instead is usually a visualisation. A warehouse rendered in three dimensions with live inventory positions is a monitoring tool, and monitoring tools are useful, but they answer no question you could not already answer. The distinguishing property of a twin is that it is predictive: it tells you what would happen under a policy you have not adopted. If it cannot do that, it is a dashboard with better graphics.

How the work runs

Four phases, and what comes out of each.

Durations are indicative and assume reasonable data availability. Where data is poor we say so in the first fortnight rather than modelling on sand.
01

1–2 weeks

Establish what the model has to answer

A twin built to represent everything represents nothing usefully and costs a great deal. We start from the specific decisions it must inform (a policy change, a capacity commitment, a network reconfiguration) because those determine the scope, the resolution and, critically, what can safely be left out. Most of the value of this phase is in agreeing the exclusions.

Outputs

  • Decision list the twin must answer, with the value at stake in each
  • Scope and resolution set per decision, with exclusions stated
  • Honest assessment of which questions need simulation and which do not
02

4–6 weeks

Build the model against real variability

Discrete-event simulation for flow and queueing, Monte Carlo for propagating uncertainty, agent-based only where genuinely warranted. The distributions come from your observed data rather than from assumed forms, supplier lead times in particular are usually far more skewed than the contract implies, and a model that uses the contracted figure will be confidently wrong.

Outputs

  • Simulation model of the relevant flow, capacity and policy logic
  • Distributions fitted to observed data, not to assumed forms
  • Lead-time and demand variability characterised per node
03

2–3 weeks

Validate it against a period you have lived through

This is the step that separates a twin from an expensive animation, and it is the one most often skipped. Given last year's actual inputs, the model must approximately reproduce last year's actual outcomes, service, inventory, cost. Where it does not, something is missing, and finding out now is far cheaper than finding out after somebody has acted on a recommendation.

Outputs

  • Backtest reproducing historical periods within an agreed tolerance
  • Discrepancies traced to missing structure and incorporated
  • Documented statement of where the model is weak and should not be trusted
04

3–4 weeks

Run the decisions, then hand it over

The model earns its cost by answering the questions from phase one, and then by continuing to answer new ones after we leave. We run the initial scenarios with your team, establish the refresh cadence that keeps it representative as the operation changes, and hand it over as something they operate rather than commission.

Outputs

  • Scenario results for the decisions the twin was built to answer
  • Refresh cadence and revalidation protocol
  • Handover with documentation and operator training

What good looks like

The ranges we design toward.

These are the improvement ranges these levers are known to produce where the underlying practice has not been examined recently. They are targets to test against your baseline, not promises.

±5%

Validation tolerance

How closely a twin should reproduce a historical period before anyone acts on its output.

60–80%

Live pilots avoided

Share of proposed policy changes that can be settled in the model rather than trialled in the operation.

Weeks

Time to an answer

Against the two quarters a live pilot takes, during which everything else has also changed.

Deliverables

What you are left holding.

All of it handed over, documented, and with a session for whoever inherits it.
  1. 01Validated simulation model of the operation, in open tooling and handed over
  2. 02Backtest evidence that it reproduces historical periods within tolerance
  3. 03Explicit statement of where the model is weak and should not be relied on
  4. 04Scenario results for the decisions it was commissioned to answer
  5. 05Revalidation protocol so it stays representative rather than quietly aging

What an engagement looks like

A representative shape.

Illustrative. This is the shape of work in supply chain & analytics, not an account of a specific client. We do not publish named results, and we would be wary of a firm that publishes them without the client in the room.

12 weeks

  1. 01Situation

    A consumer goods manufacturer was opening a third distribution centre on the strength of a regional sales forecast, with freight cost per unit rising and no cost-to-serve view below channel level.

  2. 02What we did

    Reconstructed eighteen months of shipment data into a true origin–destination model, built cost-to-serve to customer level, and tested the proposed footprint against demand growth, rate shock and node-loss scenarios.

  3. 03Where it landed

    The third site turned out to be robust only under the most optimistic growth case. Redrawing catchments and re-tendering two lanes captured most of the projected saving without the capital, and the site decision was deferred rather than cancelled.

Questions

The ones we get asked.

How is this different from the digital twin our software vendor sells?

Ask it one question: can it tell you what would happen under a policy you have not adopted? Most platform twins mirror current state beautifully and cannot answer that, because they are visualisation layers over live data rather than predictive models. That is a genuinely useful monitoring tool and it is not a twin in the sense that justifies the investment.

Do we need one?

Often not, and we would rather say so early. If the question is answerable in closed form (a safety stock level, a reorder point, a facility location) use the formula or the optimisation, because both are faster, cheaper and easier to defend. Simulation earns its cost where variability, queueing and interaction between decisions make the analytical answer wrong, which is a narrower set of problems than the term suggests.

How much data does it need?

Enough to characterise variability rather than just averages, which is more than most planning work requires. Twelve to eighteen months of transactional history, actual observed lead times rather than contracted ones, and capacity and downtime records where the constraint is physical. If the observed lead-time distribution is not available, that gap alone usually determines whether the model can be trusted.

Will it stay accurate?

Not on its own. Networks change, suppliers change, product mix changes, and a twin that is not revalidated becomes a confident description of an operation you no longer run, which is more dangerous than having no model at all. We build in the revalidation protocol and the refresh cadence, and we would rather you retire it than run it stale.

Start a conversation

Talk to us about digital twin.

Most useful first conversations are thirty minutes and involve no proposal. Describe the problem you are looking at and we will tell you honestly whether it is one we can help with.

Or write directly to hello@shyenova.com