Time-series structureEvaluationProduction trust

AI Forecasting for Real Business Decisions

Forecasting only matters when a decision changes.

Nine live models to play with — from pulling demand apart to the production trust gate.

Hosted by Tim Löhr · Data Engineer at Meta

01
Estimatereality before reality is known
02
Decideowner, policy, horizon, cost of error
03
Shipbaseline, acceptance, monitoring, feedback

Time-series structure

Pull demand apart into trend, season, and noise.

Every demand line is really three things added together — separate them before you model.

Objective

Read any series as trend + seasonality + noise, and know which parts a model can actually predict.

Why

You model the trend and season; the noise you can only buffer for — chasing it is overfitting.

Connects

The foundation for every slide after — next we add signal, choose a level, then measure what's left.

Exogenous regressors

The future you already know.

The biggest accuracy win is often telling the model what you already know is coming.

Objective

Add known-future events — promo, price, launch, holiday — as inputs and watch error fall.

Why

A driver only helps if it's known at forecast time — otherwise it's leakage, or noise you can't use, like weather.

Connects

Decomposition read the past; drivers inject the known future onto the same series.

Where you forecast is a choice

Forecast where it's reliable, act where it hurts.

The same demand exists at many levels — SKU, family, region, total.

Objective

See why higher levels forecast more accurately — and why you must still act at the SKU.

Why

Bottom-up keeps detail but is noisy; top-down is stable but its fixed share is wrong lower down.

Connects

Drivers worked on one series; here we choose which level of the tree to trust.

Industry workflow

Daily, weekly, or monthly changes the data, the model, and the action.

Daily, weekly or monthly isn't given — it's a choice.

Objective

See how the grain changes the data shape, the model that fits, and the action it drives.

Why

Coarser is smoother and lower % error, but you lose the within-period timing you act on.

Connects

Hierarchy aggregated across products; cadence aggregates across time.

Evaluation

Test on the future you are pretending you can't see.

A forecast's error only counts if it's measured on data the model never saw.

Objective

Hold out unseen time and see how one lucky split can flatter a weak model.

Why

A single split can lie; rolling over many windows gives an honest, stable read.

Connects

This is how you'd fairly test every choice from the earlier slides.

Evaluation

One error, three verdicts.

One set of errors can produce several different verdicts.

Objective

See how MAE, RMSE and MAPE disagree, and pick the one that matches your cost of error.

Why

RMSE punishes big misses; MAPE explodes on small demand — there's no universal best.

Connects

Backtest is when you measure; metrics is what you measure.

Uncertainty and buffers

Use the range to choose the business action.

A forecast is a range, and the range is a business input — not a hedge.

Objective

Pick a coverage, reveal reality, and size the buffer to the more expensive error.

Why

Wider bands are safer but cost inventory; the build point moves toward the costlier miss.

Connects

This is where a forecast finally becomes a decision — the payoff of everything before.

Explainability

Whichever model you pick is also a choice about why.

Every model is also a choice about whether you can say why.

Objective

Compare a readable glass-box against an opaque black-box on the same series.

Why

The black box edges the error, but you can't audit, override or defend its number.

Connects

Metrics scored accuracy; this scores trust — which sometimes matters more.

Model choice

No model wins everywhere.

No single method wins everywhere — the champion changes with the data.

Objective

Pick two models and a dataset, then switch the data and watch the winner flip.

Why

Don't choose a model in the abstract; choose on your decision and your data.

Connects

Closes the loop — the honest workflow of backtest + metrics beats betting on one model.