Knowing what might happen next is useful. But knowing the best thing to do about it is powerful. This is the jump from predictive to prescriptive analytics. Instead of just forecasting sales or demand, prescriptive analytics tells you the optimal course of action to achieve a specific goal, like maximising profit or minimising waste.

At its heart, this is a game of optimisation. You have an objective you want to achieve, like producing the most widgets possible. This is called the objective function. You also have limitations, like a fixed number of work hours and a limited supply of raw materials. These are your constraints. Industrial optimisation uses mathematical models to find the best possible outcome within those real-world limits.

When the relationships between your variables are straightforward, Linear Programming (LP) is the tool of choice. An LP model works when your objective function and all your constraints can be expressed as linear equations. For example, if each widget adds £5 to your profit, and each hour of labour costs £20, those are linear relationships.

Finding the best solution in this 'feasible region' of possible answers isn't done by guesswork. The Simplex algorithm is a procedure that efficiently finds the optimal point. It works by moving from one corner of the feasible region to an adjacent one, always in the direction that improves the objective function. It continues this until it can't find a better corner, which means it has found the best possible solution.

But what happens when the world isn't so linear? Sometimes, solutions must be whole numbers. You can't schedule 2.5 delivery trucks or manufacture half a jet engine. This is where Integer Programming (IP) comes in. It adds the constraint that some or all variables must be integers, making the problem significantly harder to solve because the neat geometry of the feasible region is lost.

Other times, the relationships themselves are curved. Consider optimising the shape of a vehicle to reduce aerodynamic drag or calculating the financial risk of a portfolio where diversification benefits are not constant. These problems require Nonlinear Programming (NLP), which deals with objective functions or constraints that are not linear. NLP models can be incredibly powerful but are often computationally expensive, demanding a trade-off between model accuracy and the time it takes to find a solution.

Having these powerful mathematical tools is one thing; applying them effectively is another. The DMAIC cycle is a five-step method used in business to improve processes, and it provides a perfect framework for applying industrial optimisation.

• Define: Clearly state the problem. Are we trying to minimise fuel costs, reduce delivery times, or maximise production output?
• Measure: Collect data on the current process. How much fuel is being used now? What are the current delivery times?
• Analyse: Build a mathematical model (LP, IP, or NLP) of the system. This step is where you identify your objective function and constraints based on the data.
• Improve: Use a solver to find the optimal solution for your model. Implement the changes suggested by the solution, such as new delivery routes or a different production schedule.
• Control: Monitor the new process to ensure the improvements are sustained. Set up dashboards and alerts to track performance and make sure it doesn't drift away from the optimal state.

Finally, the real world is full of uncertainty. A supplier might be late, a machine could break, or a sudden storm could close a shipping lane. Stochastic optimisation is an advanced field that incorporates this randomness directly into the model. Instead of finding the single best plan for a perfect world, it aims to find a robust solution that performs well across a range of possible future scenarios, giving businesses the flexibility to handle unexpected disruptions.