Forecast models

Logistified offers a handful of forecast models. Most operators don’t pick one — AutoAuto model selectionLogistified runs every forecast model against your sales history, scores each one on how well it would have predicted the recent past, and picks the winner per variant. Re-checks daily. Your manual overrides stick. Read more → picks the best model per variant automatically. This page lists the models, gives starter advice for choosing one, and explains what changes when you switch.

On this page

• The models at a glance
• How to choose — start simple, then turn on Auto
• No per-variant model picker
• Tuning is automatic
• Cold-start behavior
• What changes when you switch
• See also

The models at a glance

Model	When it’s a good fit	Plan
AutoAuto model selectionLogistified runs every forecast model against your sales history, scores each one on how well it would have predicted the recent past, and picks the winner per variant. Re-checks daily. Your manual overrides stick. Read more →	You don’t know — let Logistified pick the best model per variant. Default.	Essential
AverageAverage modelThe simplest forecast model. Adds up all your sales in the chosen history window and divides by the number of days. Good for variants whose demand is flat and predictable. Beginner-friendly because the math has no hidden parameters. Read more →	Flat, predictable demand with no trend.	Essential
Moving averageMoving average modelA forecast that uses only the most recent N days of sales instead of the whole history. Smooths out random ups and downs while still responding to slow drifts. Pick this when demand changes gradually but you don't see a clear up- or down-trend. Read more →	Smooth slow drifts; reduces random noise.	Essential
Weighted recentWeighted recent modelA forecast that weighs the last few days of sales more than older days. Useful when something recent changed — a new collection page, a paid ad, weather — and your near-term demand differs from your historical average. Read more →	Recent days dominate — recent change matters more.	Essential
Exponential smoothingExponential smoothing modelA forecast that gives every past day a weight, but the weights shrink the further back you go. Recent days matter most; ancient days barely register. Good when you want recency without throwing data away. Read more →	Weighted recent with a smoother decay curve.	Essential
Holt (Double exponential smoothing)Holt modelHolt's method — level plus trend, with separate smoothing for each. Better than double exponential smoothing when level and trend change at different speeds. Enhanced-plan feature. Read more →	Smoothing plus a trend component.	Enhanced
Linear regressionLinear regression modelFits a straight line through your sales history and extends it forward. Use it when demand is clearly trending up or down over time. Sensitive to outliers — a single big sale day can tilt the line noticeably. Read more →	Clear up- or down-trend.	Essential
SeasonalSeasonal modelA forecast that captures repeating patterns — Christmas, summer, end-of-month. Logistified picks up the cycle automatically from your history. Enhanced-plan feature. Read more →	Repeating short-cycle patterns — weekly or monthly.	Enhanced
MixedMixed modelA blend of multiple forecast models per variant. Operationally heavy — usually picked by hand for unusual demand shapes. Rarely auto-selected. Read more →	Blend of multiple models.	Essential

How to choose — start simple, then turn on Auto

Don’t start with the most complicated model. Start with Average (or Moving average). Look at the suggestions for your top variants. Once they roughly match your gut on a chunk of the catalog, switch to Auto and watch how the demand forecast changes — Auto will usually do better than your manual pick, but you’ll trust it because you’ve already validated the simple case.

The same advice for the history period: start with a fixed 90-day manual window. See History period.

No per-variant model picker

You won’t find a model dropdown on the variant detail page — that UI doesn’t exist. Auto already adapts the model per variant automatically based on each variant’s history shape.

If you want different model behavior for a subset of variants, create a separate view scoped to those variants and set the model at the view level. Views are how model + history-period choices are scoped in Logistified.

Tuning is automatic

Smoothing factors, window lengths, seasonality cycle lengths — all tuned internally per variant. There are no operator-facing knobs. The variant detail page shows the ranked models with their internally-chosen parameters as a read-only diagnostic, not an editor.

Cold-start behavior

Variants with very little history fall back to AverageAverage modelThe simplest forecast model. Adds up all your sales in the chosen history window and divides by the number of days. Good for variants whose demand is flat and predictable. Beginner-friendly because the math has no hidden parameters. Read more → until enough data accumulates. Specifically:

• Empty history (no sales at all): the system uses Persistence as a “tomorrow looks like today” sentinel until the first sale lands.
• Less than ~7 days of history: Average.
• Less than ~30 days: Moving average or Average, depending on signal.

Auto won’t pick a more sophisticated model until enough data is available — by design. Overriding at the view level on a thin-history variant tends to produce noisy suggestions.

What changes when you switch

Switching the model at the view level re-runs the day-by-day inventory simulation for the variants in that view. After the simulation:

• The reorder point.
• Days until reorder.
• The reorder date.
• The reorder quantity.
• Days of stock.

…all update. The recompute is fast — usually within a minute of saving the view.

Tip

Beginners often pick the most complicated model first. Resist. Start with Average and a fixed 90-day history; once the suggestions match your intuition, switch to Auto. Auto will usually do better — but you’ll trust it because you’ve already validated the simple case.

See also

• Automatic model optimization — what Auto does and when it runs.
• History period — the second big lever in forecast settings.
• Reorder suggestions — how the model’s output flows into the suggestion.
• Views & filters — how model + history-period choices are scoped.

Last updated: 19 May 2026