Data Department builds optimisation and simulation models for operational decisions with many constraints: fleet sizing, route assignment, production scheduling, stock allocation and programme planning. We use operations research methods, mainly MIP (mixed-integer programming) and simulation, often combined with machine learning forecasts. For a parcel delivery company, the analysis showed that the same service level could be met with about 30% fewer vehicles.
What problems does optimisation solve?
Planners already solve these problems every day, usually with experience, spreadsheets and a lot of phone calls. The decisions we work on:
- How many vehicles to run, and which vehicle serves which customers in which order.
- Which order goes on which production line, in what sequence, to cut changeovers and late deliveries.
- How to allocate limited stock across warehouses, stores or customers.
- How to plan shifts so demand is covered without excess overtime.
- Whether a new depot, line or warehouse is worth the investment.
- How likely a large programme is to hit its milestones, and where it will slip.
Optimisation and simulation answer different questions. Optimisation finds the best plan under your constraints. Simulation tests how a plan behaves when reality varies: demand spikes, machines stop, people are absent, tasks run late. Many projects need both.
Which method fits which problem?
| Problem | Method | What you get |
|---|---|---|
| Fleet sizing and routes | Customer clustering plus MIP, or vehicle routing heuristics for large daily volumes | Fleet size, route assignment, stop sequence |
| Production scheduling | MIP or constraint programming | Sequence per line, expected lateness and changeovers |
| Stock allocation | MIP using demand forecasts as input | Allocation per location and item |
| Capacity and layout decisions | Discrete-event simulation | Queues, utilisation and throughput under each scenario |
| Programme or project delivery | Monte Carlo simulation of resources, schedule and dependencies | Probability of hitting each milestone |
What a project looks like
Discovery (2 weeks). We sit with the people who plan today and write down the objective and every constraint: delivery windows, vehicle capacities, driver hours, machine rules, priorities. We separate hard constraints (never break) from soft ones (break at a cost). Many projects fail here because a key rule lives only in a planner's head. You get a written problem definition, a data check and a go/no-go decision.
Pilot (6–10 weeks). We build the model and run it on real historical weeks, then compare its plans with the plans you actually executed. The business metric is agreed on day one: cost per delivery, fleet size, on-time rate or changeover hours. Planners review the plans and tell us what is unrealistic, and we correct the model.
Production and handover. The model runs as a planning service on your cloud or on-premises servers, connected to ERP (enterprise resource planning), transport or MES (manufacturing execution system) data. Planners can override any decision and re-run the plan. We train your team to adjust constraints without us.
What data does an optimisation project need?
- Orders or demand, with locations, quantities and time windows.
- Resources and capacities: vehicles, lines, shifts, warehouses.
- Times and costs: travel times, processing times, changeover times, cost per hour or kilometre.
- Business rules and priorities, written down during discovery.
Results from real projects
For a parcel delivery company, we combined custom clustering with mixed-integer programming to size the fleet and assign routes under delivery windows, priorities and vehicle capacity. The analysis showed that the same service level could be met with about 30% fewer vehicles. Read the fleet and route optimisation case study.
For a global networking company, working through a US partner, we built a simulation of a large network migration programme covering resources, schedule and technical dependencies. It estimated the probability of hitting each milestone and gave early warning, so the plan and client commitments were reset in time. Read the programme delivery simulation case study.
Optimisation is often only as good as its inputs. When demand is uncertain, we pair it with demand forecasting.
When is optimisation NOT the right tool?
- Nobody can state the constraints. If the rules change from planner to planner, agree them first.
- The problem is small. An experienced planner may already be close to optimal on twenty deliveries a day.
- Time and cost data are poor. A model built on guessed travel or processing times produces confident but wrong plans.
- The real issue is organisational. If departments optimise against each other, a model will not settle that.
Frequently asked questions
Which solvers do you use?
Open-source solvers such as OR-Tools or HiGHS, or commercial solvers such as Gurobi, depending on problem size and your licensing preferences.
Can planners override the model?
Yes, always. The model proposes, the planner decides, and overrides are logged so we can learn which rules the model is missing.
How long until we see results?
Discovery takes 2 weeks and the pilot 6–10 weeks. You see the model's plans on your own historical data, compared with what you actually did.
What drives the cost of an optimisation project?
Problem size, the number and complexity of constraints, how fast plans must be produced, data quality, and the depth of integration with your planning and ERP systems.