Learn

The vocabulary behind every Lab — from match rate to response curves — explained the way you'd explain it to a customer, then again the way you'd explain it to their data team.

First-party data

Data a business collects directly from its own customers — purchases, sign-ups, app usage — as opposed to data bought or licensed from someone else.

Identity resolution

The process of recognising that different pieces of data — an email here, a device ID there — belong to the same real person.

Match rate

The percentage of your customer records that a platform successfully recognises and connects to a real user it knows.

Lookalike audience

A group of new prospects who share characteristics with your existing best customers, generated by a similarity model.

Propensity modelling

Scoring individual customers by how likely they are to take an action — usually converting — based on their behaviour.

Brand lift

The measured change in how people perceive a brand — awareness, consideration, intent — as a result of a campaign.

Footfall attribution

Measuring whether advertising actually increased visits to a physical location, like a store.

Incrementality

The share of results that genuinely happened because of advertising, as opposed to results that would have happened anyway.

Attribution

The practice of assigning credit for a conversion to the marketing touchpoints that led up to it.

Marketing mix modelling

A statistical approach that estimates how much each marketing channel — and other factors like price and seasonality — actually contributed to sales.

Response curves

A chart showing how much extra return you get from a channel as you spend more on it.

Diminishing returns

The principle that each additional dollar spent on a channel tends to generate less incremental return than the dollar before it.

ROAS vs. incremental ROAS

ROAS measures revenue against spend using platform-attributed conversions. Incremental ROAS measures it using only the conversions that genuinely happened because of the advertising.

Control groups

A comparison group that doesn't receive the treatment being studied — like advertising exposure — so its behaviour reveals what would have happened without it.

Correlation vs. causation

Two things happening together doesn't mean one caused the other.

Multi-touch attribution

A family of rules for splitting credit for a sale across every digital touchpoint a customer interacted with — not just the last one.

Customer lifetime value

What a customer is worth over the whole time they buy from you, not just on their first order.

Data clean room

A locked room where two companies can compare notes on their data without either one handing over the raw list.

Holdout group

A group of people deliberately kept from seeing a campaign, so you have something to compare the exposed group against.

Adstock

The idea that an ad's effect doesn't vanish the moment it stops running — it fades out gradually.

Saturation

The point where extra spend on a channel stops meaningfully moving the needle because it's already reaching everyone it's going to reach.

Seed audience

The starting group of known good customers you show a model so it can go find more people like them.

Identity graph

A map connecting the different digital identifiers that all belong to the same real person.

Walled garden

A platform — like Meta, Google, or Amazon — that lets you advertise and measure results inside it, but won't let raw user-level data out.

Triangulating measurement methods

MMM, MTA, and incrementality testing often each give a different answer for how well the same channel performed — triangulating means using all three together instead of trusting whichever number you saw first.