Data & drivers
Finding leading indicators in business data
A leading indicator moves before the number you care about. Search queries before orders, bookings before departures, parts receipts before output. The lag is the product; correlation on the same day is just a mirror.
Updated Aug 11, 2026·8 min read
Lead vs coincidental
A leading indicator moves before the number you care about. Search queries before orders. Bookings before departures. Parts receipts before output. Same-day correlation is a mirror. The lag is the product.
A two-day lead
How you actually find them
Lag the candidate (1, 2, 7 days — whatever the decision clock suggests). Put it in a model with the target’s own lags. Ask whether error drops on a walk-forward backtest. A heatmap of contemporaneous correlation will happily rank two series that simply share a Friday.
Negative lags exist. A promo can steal next week’s demand. If your discovery tool can only find “up and to the right,” it will miss the hangover, and ops will notice even if the dashboard does not.
Examples that show up in ops
- Web sessions leading e-commerce orders by a day or two.
- Appointments leading clinic load.
- Bill of lading counts leading warehouse receipts.
- Temperature forecasts leading energy load — a covariate that is itself a forecast.
When to trust a lead
When it survives more than one season, when you can explain the delay in one sentence, and when the series will still exist next quarter. Media mix leads rot when everyone bids on the same keyword. Re-score. Retire without sentiment.
FAQ
- What is a leading indicator?
- A series that reliably moves earlier than your target. If ad clicks rise on Monday and orders rise on Wednesday, clicks may lead orders by two days — worth testing, not assuming.
- How do you test a leading indicator?
- Lag it (1, 2, 7 days…), then ask whether it reduces error on a backtest compared with the target alone. If it only fits the training window, it is a coincidence with a fancy name.
- Can a lag flip sign?
- Yes. A promo can lift demand this week and steal next week's. Driver discovery should be allowed to find negative lags, not only 'up and to the right.'
Keep going
Guide
Forecasting with covariates
A covariate is an extra series the model is allowed to see — a promo flag, a temperature, a price. Used well, it explains swings the target's past cannot. Used as a dump of every column you have, it just overfits last quarter.
Guide
Feature engineering for time series
Most of the work is turning a clock and a few raw series into lags, rolling stats, and calendar flags the model can chew. Do that on a schedule, with no future leaking into the past, or the backtest is theater.
Guide
Why forecasts go stale in production
A forecast starts dying the moment the world moves and the model does not. New promotions, a competitor, a warehouse fire, a quiet change in how you record demand — live systems need a refresh cadence, drift checks, and a way to re-score the champion.
Use case
Shipment volume forecasting
Forecast inbound and outbound volume by lane, facility, customer, and mode before capacity is committed.
Use case
Digital traffic forecasting
Forecast traffic, transactions, and concurrency before campaigns, launches, and seasonal peaks.