Planning with hard limits

Planning

Planning with hard limits

A model that finds the best plan within the rules you can't break, and shows the people who run it why.

“We plan in spreadsheets and can only compare a few options, so we play safe.”

Diagram. Existing systems: Lab results, Ingredient stock and lots, Recipe specs, Production plan, Ingredient costs. They feed Blending model, refined through Ingredient data, Spec rules, Recommended blend. It serves the Planning team (lot alerts), the Quality team (batches against spec), the Plant managers (line performance).

Illustrative example.

  1. The problem

    “We plan in spreadsheets and can only compare a few options, so we play safe.”

  2. The solution

    1. 01Process mapWe sit with the people who make the plan today.
    2. 02RulesWhat must never happen, such as an off-spec batch, and what you would prefer.
    3. 03ModelA solver finds the plan that meets every rule at the lowest cost.
    4. 04Planner viewPlanners see the plan and the reasons for it. If they change it, they see what the change costs.
  3. Previous work

    Cocoa ingredient waste

    We built a model that picks each day's cocoa bean blend to meet spec with the least waste. It went into full production and saves €3.5M a year per plant.

    Full story

    The client asked for a forecast. The real question was which beans to blend each day, given what had arrived and what the recipes allowed.

    National 4G rollout

    A national mobile operator had to choose which of its cell sites to upgrade for 4G, and in what order, within a fixed budget. The work was won against a top-tier strategy firm and delivered on time.

    Full story

    John designed the solution while at a telecoms analytics company.

  4. First steps

    1. ScopeWe pick one plan that is made by hand today and costs money when it's wrong.
    2. WalkthroughYou show us how it's made now, including rules nobody has written down.
    3. PilotThe model runs beside your planners, recommending only, until you've seen where it does better.
    More detail

    This is optimisation, which is older than AI and often the better tool. It finds the best plan that breaks none of your rules, and shows its working.

    The first model is rarely the last. On one job our first approach worked for two stages of the recipe and failed after that, so we changed solver early.

    A plan is only as good as its data and rules. Some rules are only in people's heads until we write them down together.

  5. Get in touch

    Tell us which plan you make by hand and which rules it must keep. We'll tell you if a model would do better.

    Have a chat