Skip to content

Multi-objective optimisation

Low cost and fast delivery work against each other, and one “optimal” plan hides who set the balance. We calculate the best options, and you choose.

  • Free 30-minute call
  • No commitment
  • You own the code
Tools
Gurobi
Used in
2 documented use cases
Delivery
Documented source code, yours to keep

In short: Multi-objective optimisation is planning for several goals at once, such as cost in francs and delivery time in hours. It returns a Pareto front: the set of plans where you can only improve one goal by giving up some of another. You then choose the plan that fits.

The problem

Finance wants the cheapest plan. Sales wants next-day delivery everywhere. Your sustainability team wants fewer kilometres on the road. Your analysts build one “optimal” plan, and each side asks whose priorities it follows. Nobody can say. How much each goal counts was set inside the model, and nobody discussed it.

What changes

Today, you get one plan and a debate about whose numbers it favours. With our model, you see the best possible plans for cost in francs and delivery time in hours, and for kilometres if you need them. The people who answer for the decision explore these plans in an interactive chart. They choose a plan they can defend, and never have to accept a number they can’t question.

How it works

Multi-objective optimisation plans for several goals at once. Its result is a Pareto front: the set of plans where you can only improve one goal by giving up some of another. Any plan off that curve is wasted: another plan does better on one goal and no worse on the rest. A single “optimal” plan would hide who chose the weights and why, so we show you the whole curve instead.

We calculate the plans on the Pareto front with three methods: the weighted sum, the epsilon-constraint method and direct Pareto-front generation. A weighted sum folds all goals into one score, with weights you set. The epsilon-constraint method optimises one goal and sets an upper limit on each of the others, such as at most 48 hours of delivery time. Pareto-front generation calculates the plans on the curve directly, one after another.

Why sliding the weights can miss good plans

Most guides show the weighted sum and stop there. With yes-or-no decisions in the model, that’s risky. An answer on Operations Research Stack Exchange shows it with a small example: a weighted sum finds only some of the plans on the Pareto front. The other plans on the curve are never found, whatever weights you set. In optimisation, people say the curve bends inward there. One of these plans can be exactly the one your board would pick.

Setting an upper limit on one goal at a time, the epsilon-constraint way, reaches those plans too. Weights also need a common scale. If cost runs in millions of francs and delays in minutes, cost decides almost everything. Another answer on the same forum suggests a question for the decision maker: which trade would you accept, say ten minutes of delay for a given amount? The weights then follow from the answer. Sometimes one goal simply comes first. Gurobi can then optimise the goals in order. A more important goal may then get worse only by a small amount you set. The honest boundary: if one goal clearly matters far more than the others, a single-objective model is simpler. We’ll tell you when that’s the case.

Trade-offs made visible

Imagine every possible plan as a point on a chart. Further right means cheaper. Further up means faster delivery. Most points are wasted, because for almost every plan there’s another that is both cheaper and faster. The plans that survive that test form the Pareto front. On that curve, you can only improve one goal by giving up another. The Pareto front doesn’t tell you which plan is best. That depends on how much a faster delivery is worth to you in francs. It shows you which choices exist, and you set that weighting yourself.

  • Weighted-sum scalarisation: all goals folded into one score, with weights you set.
  • Epsilon-constraint: optimise one goal, with an upper limit on each of the others.
  • Pareto-front generation: calculate every plan on the Pareto front, one after another.

In the interactive chart we deliver, you set how much weight cost and delivery time get. The highlighted plan then moves along the Pareto front. Your stakeholders use it to see what faster delivery costs and pick a plan they’ll defend.

When it isn’t the answer

The honest boundary: if one goal clearly matters far more than the others, a single-objective model is simpler. We’ll tell you when that’s the case.

Common questions.

Something else? Ask us directly.

A weighted sum adds all goals into one score. Then the goal with the bigger numbers quietly decides. Cost in francs will outweigh delay in minutes. One fix is to translate each goal into money, if you can honestly put a price on it. Another is to ask the decision maker which trade they’d accept, say ten minutes of delay for a given amount. The weights then follow from that answer.

Yes. Gurobi calls this hierarchical, or lexicographic, optimisation. It optimises the top goal first, then the next one without making the first worse. You can allow a small tolerance. Then the top goal may get a little worse if a less important goal gets better in return. That suits rules like “fast delivery first, then cost, then kilometres”.

For planning models with yes-or-no decisions, you rarely need a special package. The epsilon-constraint method is a loop around your normal model: set a limit on one goal, optimise the other, move the limit. Gurobi also has blended and ranked objectives built in. Dedicated packages mostly target black-box problems where no model can be written down.

Sometimes they can, and then we’ll say so. Standard tools are built for an average problem, though. With Multi-objective optimisation, we start from yours: its size, its rules and the exceptions nobody wrote down. That is why the plan fits your operation better.

You don’t need either to start. We build the model from the data you’ve got, messy or not, and cleaning it up is part of the job. You get a documented system your team can maintain, and we train them until they can run it long after we’ve left.

It starts with a free 30-minute call about your operation and the model that would fit it. Then we build a working prototype and run it on your numbers. You’ll see results before you commit to anything bigger.

Tell us what you’re planning.

In a free 30-minute call with a specialist, you find out whether optimisation can help.

  • Free, no commitment
  • You talk to a specialist directly
  • A clear go or no-go