Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Glossary term

Baseline

An estimate of what a metric would have done without a given change, campaign or promotion, against which the effect is measured.
AnalyticsconceptUpdated

Definition

A baseline is the expected value of a metric in the absence of a specific change. It is a counterfactual: nobody can observe the version of the period in which the change did not happen, so the baseline estimates it. The effect of the change is then read as actual minus baseline.

The word is also used more loosely for a starting record. The tutorial on taking over a marketing programme asks a new owner to write a baseline of current performance before changing anything, and the guide to measuring AI's impact on team output takes one before AI tools are introduced. Both serve the same purpose: without a fixed record of how things were, no later improvement can be shown.

In practice: history, trend, season and control groups

Baselines come from two places.

The metric's own history. The same period last year, adjusted for the trend between the years; or a clean run of weeks before the change, which reflects the current trend but misses seasonal movement. The guide to measuring a sale or promotion sets these side by side, with unpromoted products and unpromoted regions as further options, and recommends building more than one when there is no randomised control. The guide to SEO forecasting uses the same idea in the other direction: the baseline is the projection of existing organic traffic, with trend and season, before the gains from planned work are added.

A comparison group. Regions, products, customers or periods that did not receive the change give a baseline that moves with the same weather, news and season. The guide to measuring offline media reads branded search and direct visits against a baseline and prefers regions where the media did not run, because a comparison group carries the season with it.

Why the baseline decides the measured effect

Since the effect is a subtraction, every error in the baseline passes straight into the result. A few patterns cause most of the damage:

  • The wrong comparison period. Comparing a sale week with the week before credits the sale with seasonal movement and with everything else that changed.
  • Contaminated periods. A baseline that includes a stock-out, an outage or the days after a sale was announced, when customers wait, starts too low and inflates the effect.
  • Choosing after seeing the result. Trying several baselines and reporting the one that flatters the work produces a number, not a measurement.

The remedy is procedural: write the baseline method down before the result is read, record unusual events as they happen, and report the effect as a range when several reasonable baselines disagree.

Illustrative example — not a real result

A bike repair chain checks its radio campaign against previous springs

Wexmoor Bicycle Repairs ran a four-week radio campaign in April, and bookings across its five workshops rose from an average of 410 a week in March to 610 a week during the campaign. The owner was ready to credit the campaign with 200 extra bookings a week. The marketing manager built a baseline from the previous two springs: in both years, April bookings had risen to about 1.4 times the March level with no radio at all. That put the expected April figure at about 575 a week, leaving roughly 35 a week, not 200, as the possible effect of the campaign. The team decided to test the next burst in two workshops only.

Incrementality is the effect measured against a baseline: the outcomes that would not have happened without the change. A Holdout Group is the strongest baseline, a randomly withheld group that experiences everything except the change. Seasonality is the most common reason a naive baseline goes wrong, because many businesses rise and fall at the same times each year. Demand Pull-Forward lowers the periods after a promotion, so a baseline must look beyond the promotion window to catch purchases that were only moved earlier.

Common confusions

  • The baseline is last week. Last week is one possible baseline, and often a poor one, because it ignores season and other changes.
  • A baseline is a target. A target is what the team wants; a baseline is what would have happened anyway.
  • One baseline is the truth. Without a control group there are several reasonable baselines; their spread is part of the result.
  • The baseline can be set afterwards. Chosen after the result, it tends to match what people hoped to see.
  • A rise above baseline is all incremental. Purchases borrowed from later weeks or other products still need subtracting.

More terms to look up

Where this term is used

Related