Forecasting and optimisation
Integrated forecasting and optimisation
Statistical forecasting and constraint-based optimisation are built into the Orthantic planning workflow. They run on the same data as the plan, and the AI agent can explain their results in natural language.
Forecasting inside the planning process
When forecasts are produced in a separate tool, each cycle requires exporting data, running the forecast and importing the results back into the plan.
In Orthantic, planners run, review and adjust the forecast within the planning model. Results are written directly to the plan, with no transfer between systems.
Statistical forecasting
Select the hierarchy level, horizon and frequency. Orthantic evaluates several statistical methods against your history and selects the best fit for each series.
Screens from Orthantic's demo demand planning model. Select one to see it larger.
Forecast at the level that fits your data
Data is often held at a detailed level, such as SKU by store, where individual series are too sparse or volatile to forecast reliably. Orthantic lets planners choose the level at which to forecast, independently of the level at which data is stored.
- Data at SKU × store
- Forecast at category × customer
- Automatic disaggregation
- Forecast at SKU × store
-
Forecast at a higher levelAggregate
With data held at SKU by store, forecast at category by customer, or any other combination of hierarchy levels, where demand patterns are more stable.
-
Group like productsCombine
Combine similar products, such as variants, pack sizes or low-volume lines, and forecast them as a single group.
-
Automatic disaggregationSpread
Forecasts made at a higher level or for a group are automatically disaggregated back to the detailed level, ready for planning.
-
Forecast at detailDetail
Where volumes support it, forecast directly at the level the data is held, such as SKU by store.
Statistical baseline, consensus plan and actual
Thousands of units. Actual stops after July; the interval band opens from there.
Illustrative figures. The prediction interval widens with the forecast horizon.
Baseline and planner overrides, kept separate
Planners adjust the statistical baseline for promotions, customer changes and market events. Orthantic stores these overrides in a separate layer, so every adjustment remains visible and reviewable.
Because the baseline and overrides are held separately, the accuracy of each can be measured over time.
Linear optimisation
Find the best allocation of any limited resource within the planning model: production and supply, a marketing budget across TV, radio, print and digital, capital across projects, or people across shifts. The constraints are read directly from the plan.
Screens from a demo marketing mix in Orthantic. Select one to see it larger.
Demand, supply, capacity, budgets, costs, expected returns, service levels and business rules come straight from the planning model, so the optimisation always reflects the current plan.
Update a figure such as factory capacity and run the optimisation again, with no export, no separate tool and no downtime.
Combine objectives such as cost, margin, return on investment and service level, weighted by their importance, with continuous, whole-number and yes/no decisions.
The recommended allocation is written back into the model as a scenario to review, adjust and compare before it is adopted.
Balancing competing costs
Setup cost falls and holding cost rises as batch size grows. Total cost is lowest where they balance.
Illustrative figures. The optimiser finds the point of lowest total cost, here a batch size of 4,000 units, while respecting every constraint in the model.
Standalone tools compared with integrated forecasting
| Standalone approach | Orthantic |
|---|---|
| Scripts maintained by specialists | A planning step any authorised planner can run |
| A single method applied across the catalogue | Best-fit method selected per series, evaluated on your history |
| Results imported as static values | Results written back to the model and updated with the plan |
| Overrides merged into the baseline | Overrides held in a separate layer, with reasons recorded |
| Forecast accuracy difficult to measure | Accuracy of baseline and overrides measured separately |
| Forecast changes explained by the specialists who ran them | Planners ask the AI agent why a forecast changed, and it answers with the drivers |
| Runs started in a separate tool | Runs started from the dashboard, or offered by the AI agent for the planner to run |
| Few scenarios, limited by the effort to build each one | Multiple scenarios compared at full detail |
Evaluate it against your own history
Get in touch to arrange a demonstration of forecasting and optimisation, and to discuss your own planning questions.