Using AI and Optimisation Tools in Network Design
A city warehouse looks perfect on a map until fuel prices rise, a highway floods, and your fastest-selling SKU starts moving in a different region. Network design is where supply chain decisions stop being neat spreadsheets and become real commitments - rent, trucks, service levels, risk and customer promises all locked into one operating shape.
- Network design decides where facilities sit, what capacity they hold, and which lanes serve which markets.
- AI improves inputs - demand forecasts, risk signals, lead-time estimates, cost patterns and disruption alerts.
- Optimisation chooses the best feasible design - it minimises cost or maximises service while respecting constraints.
- The core model is: objective function + decision variables + constraints + scenarios.
- Do not optimise only for transport cost. Include inventory, facility, handling, service penalty, tax/regulatory and risk costs.
- A strong answer compares baseline vs scenarios, not just “AI will find the best warehouse location.”
- The biggest interview win: say clearly that AI predicts; optimisation decides; managers validate.
Big Picture - AI Predicts, Optimisation Decides
Think of network design as a layered decision pyramid. The bottom layers create reliable inputs; the middle layer converts business rules into mathematical constraints; the top layer is the actual network choice a manager can defend.
Core Explanation - What AI and Optimisation Actually Do
The most practical way to explain this topic is to separate prediction from decision.
AI helps estimate uncertain inputs: future demand by region, expected lead times, probability of disruptions, order clustering, delivery density and exception patterns. Optimisation tools then use those inputs to choose the best network configuration under constraints.
A network design model typically asks:
- Which plants, warehouses, cross-docks or dark stores should we open, close or expand?
- Which customer zones should each facility serve?
- What capacity should be assigned to each node?
- Which transport lanes should be used, and at what frequency?
- How should the design change under demand growth, disruption or cost inflation?
The Four-Part Model Interviewers Expect
When you say “AI and optimisation tools,” make it concrete. A serious answer has four parts.
AI vs Optimisation - Do Not Mix Them Up
This distinction is where many candidates sound vague. AI is not the same as optimisation. They work together, but they answer different questions.
If you want to go deeper into the demand and replenishment side that feeds network models, revise Using AI for Inventory Optimisation and Replenishment.
Metrics That Prove a Network Design Is Better
A network design is not “better” because the software says so. It is better only if it improves measurable business outcomes. There is no universal benchmark across industries, so the right comparison is usually against the company’s current network and promised service level.
A Small Worked Example - Two Warehouse Options
Assume a company is comparing two network options for one region. The numbers below are illustrative, not industry benchmarks.
Decision logic: Option B has higher fixed and inventory cost, but lower last-mile cost and lower service penalty. It saves ₹1.5 lakh per month in this simplified model and improves delivery reliability. A good candidate will then ask: is the demand stable enough to justify the second site, and can the company staff and operate it reliably?
Definitions to Say Clearly
- Network design: Choosing facilities, lanes and capacities to meet demand at the lowest feasible cost and risk.
- Optimisation model: A model that minimises or maximises an objective while obeying constraints.
- Decision variable: A choice the model can change, such as opening a facility or assigning a customer zone.
- Constraint: A rule the solution must obey, such as capacity, budget, SLA or labour availability.
- Scenario: A plausible future demand, cost or disruption case tested before committing to a network.
Case Study - Delhivery and the Network as the Product
Delhivery shows why network design is not just a support function in logistics - the network itself becomes the operating advantage.
Situation: In parcel logistics, demand is fragmented by city, pin code, seller location, delivery promise and shipment type. A weak network creates empty miles, overloaded hubs, missed promises and expensive manual firefighting.
The move: Delhivery’s model depends on designing a national logistics network that combines pickup points, sortation centres, line-haul connections, destination processing and last-mile delivery. Optimisation matters because the company has to decide which parcels should move through which nodes, how much flow each hub can absorb, and how to balance consolidation efficiency against delivery speed.
Why AI matters: AI can improve the inputs - demand forecasting by lane, ETA prediction, exception detection and shipment clustering. But the real decision still needs optimisation: which hub handles which flows, where capacity should be added, and how service levels change if demand shifts.
Outcome and lesson: The strategic lesson is not “more warehouses are always better.” The primary driver is intelligent flow design - deciding where to aggregate and where to stay close to demand. Supporting drivers include automation, route planning, standard operating processes, capacity planning and real-time exception handling. That is the complete answer interviewers like: network advantage comes from the system, not one magic tool.
When to Use Which Optimisation Approach
You do not need to name every algorithm in an interview, but you should know the broad tool families and when they fit.
How AI Changes Using AI and Optimisation Tools in Network Design
AI is changing network design in three concrete ways.
- Better demand sensing: Models can combine order history, seasonality, promotions, weather signals and local events to create sharper regional demand inputs for the optimiser.
- Faster scenario generation: Generative AI can help planners frame disruption scenarios - fuel increase, port delay, labour shortage, demand migration - that are then tested in optimisation or simulation tools.
- Live exception intelligence: ML-based ETA prediction and anomaly detection can flag lanes, hubs or regions where actual performance is drifting from the designed network.
Practical student workflow: Load a company annual report, a short description of its distribution network and your class notes into NotebookLM. Ask it to generate: “What network design trade-offs does this company face, and what data would an optimiser need?” Then use ChatGPT to convert the output into a 5-step interview answer with objective, variables, constraints, scenarios and metrics.
AI can make a network design look scientific even when the input data is biased, outdated or incomplete. Always ask: which assumptions are driving the recommendation?
Interview Relevance
“Suppose an e-commerce company wants to redesign its warehouse network for faster national coverage. How would you use AI and optimisation tools?”
Use this sentence: “I would not let AI directly choose the network; I would use AI to improve forecasts and risk signals, then use optimisation to select the best feasible design.”
Common Mistake
Mistake: Saying “AI will find the best warehouse locations” without defining cost, constraints, service levels or scenarios. It costs candidates because it sounds like tool worship, not managerial thinking. Fix: Always frame the answer as objective function, decision variables, constraints, scenarios and validation.