One method for every SKU
A spreadsheet applies the same moving average to a steady seller, a seasonal line and a product that sells twice a month. Each needs a different model, and nobody tests which one fits.
DEMAND FORECASTING
A forecast that ends in a purchase order, not a spreadsheet
Miridia Planner is the forecasting module of Miridia, the supply chain automation platform. It picks the best model for each SKU from your own history, shows whether each planning touch helped, and turns the plan into reorder points and draft purchase orders in the same system.
WHAT GOES WRONG
Where a spreadsheet forecast costs you
Stockouts hide demand
When the shelf is empty, the sale never enters the history. The history then understates demand, the next forecast is lower, and the next order is smaller. Gruen and Corsten found that the manufacturer carries an average of 35% of an out-of-stock sale.
Nobody knows if the overrides helped
Sales adds a promotion, the planner rounds up, and finance trims the total. Each change feels right, and none is measured, so the same habits continue whether they improve the forecast or not.
The forecast stops at a file
The forecast lives in one tool and the purchase orders in another. Someone copies the numbers across, and the reorder points in the system stay where they were first set.
8.3% average out-of-stock rate across the FMCG industry
35% of an out-of-stock sale is carried by the manufacturer, on average
From The Cost You Cannot See, with the source of each figure in the report.
THE MODELS
A pool of models, tested on your history
Each SKU gets rolling-origin backtesting against its own history. The champion must beat the naive and the seasonal-naive benchmarks before Planner trusts it.
- Classical
- AutoARIMA, AutoETS, Holt-Winters and Theta, through Nixtla StatsForecast
- Intermittent demand
- Croston, for products that sell in small, irregular quantities
- Machine learning
- XGBoost and LightGBM, with conformal prediction intervals
- Ensemble
- A weighted blend of the best candidates, which often beats any single model
A short or intermittent history goes to a smaller pool of models, so a product with little data is not overfitted.
HOW IT RUNS
From history to purchase order
- 01
History cleansed
Planner reads the sales history from your connected channels, or from a validated CSV upload. It proposes outlier caps, stockout imputation and gap fills, and the planner approves each one. The raw history is always kept.
- 02
Champion selected
Each SKU is backtested against its own history, and the model with the best result becomes its champion. A champion that cannot beat the naive benchmarks is not trusted.
- 03
Plan agreed
Planners add growth assumptions, promotion uplifts and overrides on top of the baseline, which stays intact. A consensus cycle reconciles sales, finance and operations, and publishes a fixed snapshot.
- 04
Orders drafted
The plan sets the lead-time demand, the safety stock, the reorder point and the days of cover. Planner turns its suggestions into draft purchase orders, one for each supplier, rounded to the minimum order quantity.
FROM HISTORY TO PURCHASE ORDER01History cleansed
02Champion selected
03Plan agreed
04Orders drafted
Connected by Miridia
THE PLATFORM
What Miridia does about it
Forecast value added at every layer
Planner measures the accuracy of the naive forecast, the statistical forecast, the planner-adjusted forecast and the consensus. If an adjustment makes the forecast worse, the number shows it.
Demand cleansing that you approve
Detectors propose outlier caps, stockout imputation and gap handling. The planner approves each proposal, and the raw history stays unchanged beside the cleansed one.
Scenarios before you commit
Test a price change or a promotion against the current forecast with an elasticity-based scenario. When sales signs off, promote the scenario into the live plan.
Plans for each location
Forecast at each location, with hierarchical reconciliation. Two-echelon replenishment shows which stock each location needs and when, and it splits the result into transfer and purchase order sheets.
Retailer sell-out as demand
For a retailer that sends sell-out data, Planner forecasts what consumers buy in its stores instead of the purchase orders that it sends, and it derives the orders to expect.
Segments from real revenue
ABC-XYZ segmentation ranks each SKU by revenue and by the variability of its demand, so the planning effort goes to the products where it pays.
HONEST FIT
When Miridia fits, and when something else fits better
Miridia is a good fit when
- You forecast in a spreadsheet, or in a tool that does not reach the purchase order.
- You plan hundreds or thousands of SKUs with different demand patterns.
- A stockout or an overstock has cost you a customer or tied up cash.
- You want to know whether the adjustments of your planners improve the forecast.
Choose something else when
- You plan for thousands of stores and need automated store replenishment at that scale. An enterprise planning suite such as RELEX or Blue Yonder
- You sell a few dozen steady products and reorder from a simple rule. Reorder points in Miridia Core, without Planner
QUESTIONS
Demand Forecasting: common questions
Which forecasting models does Miridia Planner use?
AutoARIMA, AutoETS, Holt-Winters and Theta through Nixtla StatsForecast, Croston for intermittent demand, and the gradient-boosted trees XGBoost and LightGBM with conformal prediction intervals. A weighted ensemble of the best candidates is also tested.
How does Planner choose a model for each SKU?
Planner runs rolling-origin backtests of each candidate against the history of the SKU. The model with the best result becomes the champion, but only if it beats the naive and the seasonal-naive benchmarks.
What is forecast value added?
Forecast value added compares the accuracy of each step of the forecast with the step before it. Planner measures it for the naive, the statistical, the planner-adjusted and the consensus forecast, so you see which touches help and which hurt.
Can I start with a CSV file?
Yes. Planner accepts a validated CSV upload of your sales history. When your channels and systems are connected, it syncs the history live instead.
How does the forecast reach a purchase order?
The plan sets the lead-time demand, the safety stock and the reorder point of each SKU. Planner exports supplier-grouped order sheets, or it turns its suggestions into draft purchase orders, one for each supplier, in one step.
Is Planner part of the Starter plan?
No. Miridia Planner is an advanced module with its own price. The core platform includes reorder points, safety stock and automated purchase orders without Planner.
MAKE. MOVE. MEASURE.
See Planner on your own history
Bring a sales export. We load it into a demo account and show you the champion model and the forecast value added for your own SKUs.