Scenario modeling & digital twins

Network decisions are expensive to get wrong, so companies test them on a model first. Scenario modeling asks “what if?” in a structured way:

• What if demand grows 30% in the South?
• What if diesel costs rise 20%?
• What if the Chennai warehouse is out for a month after a flood?
• What if we close two warehouses and open one large hub?

Each scenario is run through the same model, often an optimization that finds the best response, or a simulation that plays out weeks of operations with random demand and delays. Comparing the results shows which designs hold up and which collapse when conditions change. A design that’s slightly more expensive in the base case but much safer under disruption is often the better choice.

A digital twin takes this further: a living model of the real network, connected to current data on stock, orders, capacity and costs. Instead of a one-off study, planners can test decisions as conditions change, such as rerouting around a port closure or seeing the effect of a promotion on warehouse capacity, before committing in the real world.

The value doesn’t come from the software alone. It needs clean data, a clear question, and someone who checks that the model behaves like the real network before trusting its answers.

For a hands-on example, Tributary is a small supply chain digital twin that runs in the browser: change a lever and watch the forecast, network, inventory and a simulated year respond.

A scenario study, step by step

1. Ask a clear question: for example, which of three network designs should we build?
2. Agree the measures: total cost, service, risk exposure, emissions.
3. Validate the baseline: the model must reproduce today’s performance.
4. Design scenarios: the base case plus plausible changes in the key drivers (demand by region, costs, disruptions), alone and in combination.
5. Run every design in every scenario, letting flows re-optimize in each, and compare.

Optimize or simulate?

• Optimization finds the best decisions for given inputs, usually averages. It answers “what should the network look like?”
• Simulation plays out operations over time with random demand, delays and failures. It answers “how will this design actually perform, including service levels, queues and stock?”

A common pattern: optimize to shortlist designs, then simulate the shortlist to check service and resilience under variability.

What makes a digital twin

A network model becomes a digital twin when it is connected to live data (stock, orders, capacity, transport status) and refreshed often enough to support current decisions. Maturity usually grows in steps:

1. Descriptive: an up-to-date picture of the network, so everyone sees the same state.
2. Diagnostic: explaining why performance changed.
3. Predictive: forecasting what will happen, such as a stockout next week if nothing changes.
4. Prescriptive: recommending, and testing, the best response.

Many projects stop at the first step and become dashboards; the value comes from the last two.

Keeping models honest

• Validate against history: replay a past period and check the model reproduces what actually happened.
• Check face validity: walk experienced planners through the model’s behaviour; they will spot impossible flows quickly.
• Match fidelity to the question: a strategic network question needs weekly flows by region, not every pallet movement. Too much detail slows the model and hides errors.
• Keep data fresh: a twin running on last quarter’s freight rates and lead times gives last quarter’s answers.

Decisions under deep uncertainty

When probabilities can’t be estimated with any confidence, as with geopolitical shocks, pandemics or new regulations, expected-cost calculations become guesswork. Approaches that help:

• Robust choices that perform acceptably across all scenarios, as with minimax regret.
• Stochastic programming, which commits to first-stage decisions (sites, capacity) while letting second-stage decisions (flows, inventory) adapt to each scenario.
• Adaptive plans: decide now only what must be decided now, and set signposts that trigger the next steps, such as “if southern demand grows above X, open the third site”.

Using it in the business

Scenario models earn their keep in recurring decisions: S&OP what-ifs, capacity investment cases, disruption playbooks (“if this port closes, here is the reroute and its cost”) and negotiations (“what is a second supplier worth to us?”). The best ones are owned by the planning team and used every month, not commissioned once and left on a shelf.