营销与数据工程最常误解的 24 个营销数据术语

Imagine you’re a marketer planning a win-back campaign and you ask your data team for a list of “inactive customers.” You’re expecting people who haven’t bought anything in 90 days. The list comes back based on users who haven’t opened your app in 30 days, because that’s how “inactive” is defined in the data.

Or you ask for “real-time” audience updates to avoid sending “we miss you” to someone who just placed an order. You’re picturing a list that refreshes every 15 minutes. The data team hears “updates within seconds” and comes back with questions about new infrastructure, ongoing costs, and a longer timeline.

These differences affect what gets built, how much it costs, and when marketing can use it. Agreeing early on what a “customer” means, what makes an audience ready, and how fresh a signal needs to be can prevent misunderstandings from becoming campaign rework.

This guide covers 24 terms that marketing and data engineering often interpret differently, organized around five practical questions about campaign execution. Use the paired definitions as conversation starters: surface assumptions, clarify what each side needs, and agree on meanings that fit your business before work begins.

Get all 24 terms in the printable pdf here.

Same word, different meaning printable guide pdf

1. Where does the campaign data come from?

Information in a marketing tool can depend on several systems and processing steps. Understanding that path helps you explain what you need and work out where a change belongs. For example, take a field called “last purchase date.” Before using it to identify lapsed customers, agree on whether it includes canceled orders and which system supplies the date.

Term	What marketing often means	What data engineering often means
Event	Something a customer did that you can target on.	A timestamped record with a schema, written once. The schema can change, and that affects ingestion and everything downstream.
Data source	The tool or application where you see the data.	The system that produced it. Which system, how it arrived, and which version counts.
Field	A box in the segment builder.	A column with a type (what values are allowed), a nullability rule (can it be empty), and an owner.
Raw data	An unformatted export.	The untouched landing layer, preserved so downstream can be rebuilt.
Sync	The data shows up in the tool.	A job with failure modes, retries, and last night's failed records. Scheduled, change capture or event-driven.
Integration	Two tools are connected.	A direction of travel with auth, field mapping, schema handling, rate limits, retries, monitoring and an owner.

Agree on: Where the value comes from, what it includes, and who maintains it.

2. Who does a customer record represent?

Customer identity determines who gets included, excluded, or worse – counted twice. Start by establishing what one record represents and how the team decides that two records belong together.

Marketing’s knowledge of the customer relationship helps engineering make better decisions about identity. In the win-back example, a person and a household are both useful ways to organize customer data. The campaign needs an explicit choice between them. A household view might suit an offer limited to one per home. A person view might suit an individual loyalty benefit.

Term	What marketing often means	What data engineering often means
Customer	A person.	An entity at a grain. Person, account, household, or device, and which identifiers represent it.
Profile	Everything known about someone, in one view.	A join, several tables stitched on a shared key, computed at a moment and already slightly stale.
Deduplication	Removing the doubles.	Identity resolution. A policy with thresholds. Too keen merges two people, too cautious splits one.
Golden record	The truth.	The output of rules a person wrote, which can be changed.

A shared data foundation lets teams maintain clearly named views for those purposes and reuse the appropriate one. Each needs a documented meaning, the identifiers used to connect records, and an owner for the matching rules.

Agree on: Whether the campaign targets a person, account, household, or device, and how the relevant records are matched.

3. What makes an audience ready to activate?

An audience moves through several states: the rules that define it, the records that qualify and the records a destination can use. Each state answers a different question about campaign readiness.

Suppose 50,000 people match a segment, but the email platform only accepts 47,000 of those records. In this example, the difference could reflect missing destination identifiers, rejected records, or exclusions applied during delivery. The team needs to account for that difference before deciding the audience is ready.

Term	What marketing often means	What data engineering often means
Segment	A group of people to message.	Needs to know if you mean the definition or the result. Those cost differently.
Audience	Interchangeable with segment.	Often the exported copy of a segment, the version pushed to the destination, not the logic behind it.
Model	A prediction, like propensity to churn.	The data model. Tables and how they relate. The worst offender on this page.
Computed trait	A useful number on a person, like lifetime value.	Something recalculated on a schedule, at a cost, for every person, forever.
Lookalike	Expanded reach, built from a seed audience.	A downstream black box they cannot inspect, explain or debug. Expect hesitation about depending on it.
Activation	Launching a campaign, or putting an audience to use.	Delivering eligible records to a destination with the right identifiers, permissions, freshness and format.

Keeping the definition, evaluated membership and delivery result identifiable makes this easier. Marketing can check whether the rules express the intended audience. Engineering can check whether the correct records reached the destination. Both can distinguish expected exclusions from a problem that needs fixing.

The same precision helps with predictions. If the campaign needs a propensity score, name the prediction and what it’s intended to estimate. That gives the team something more specific to build and validate than “the model.”

Agree on: The audience selection rules, the prediction or trait required, and how delivery will be validated.

4. When is the data fresh enough?

Freshness depends on the customer decision you’re trying to make. A single campaign can have several timing requirements, even when its data comes from one platform.

Consider a churn score calculated each night and sent to an engagement tool every 15 minutes. More frequent delivery doesn’t make the prediction 15 minutes old. It’s still based on the nightly calculation.

Now add a customer who purchases this morning. The campaign may need to remove that person from a win-back audience before the next send, while the churn score can wait for its next scheduled calculation. Those are separate requirements with different consequences for the customer.

Term	What marketing often means	What data engineering often means
Real time	Soon enough that it does not feel broken.	Sub-second streaming, an always-on architecture, and a cost line that never stops.
Refresh	The numbers change.	A pipeline run with dependencies, a schedule, and someone on call for it.
Latency	The delay before you see something.	Several delays that add up, measured at points you have not specified.
Live	It is working now.	It is in production rather than in a test environment.

To make freshness meaningful, identify the start and end of the delay: from the purchase happening, to its availability in the data platform, from that availability to a membership change, or from the membership change to the destination being able to act on it.

These distinctions help the team invest in speed where it changes the experience. They also give marketing a better way to describe a problem than saying the whole campaign needs to be “real time.”

Agree on: Which customer event starts the clock, where the updated data must be usable, and the acceptable delay.

5. Who is responsible when the data is used?

Getting access to customer data brings questions about its meaning, permitted use and ongoing ownership. Those responsibilities continue when an audience leaves the data platform.

Term	What marketing often means	What data engineering often means
Source of truth	The dashboard you trust.	The designated system, governed dataset or semantic definition. There can legitimately be different ones per subject.
Consent	The customer ticked the box.	A per-purpose permission that must travel downstream and be enforceable at query time.
Copy	Downloading or exporting the data somewhere.	Any additional persisted version. Replication, materialised tables, caches, extracts, destination-side storage.
Governance	The reason things take longer.	The reason the company has not had to disclose a breach.

In practice, governance also covers data quality, ownership and appropriate use. A useful starting point is to identify the designated source for each decision. The system that establishes customer identity may differ from the one that records a channel preference or calculates revenue.

Then follow the data to its destination. An audience that was eligible when exported may need an updated suppression before a later send. A saved copy can outlive the campaign that created it. Teams need to know how changes reach the destination and who owns the work when they do not.

Databricks CustomerLake, the Agentic CDP, brings customer data platform capabilities natively into Databricks and its governance foundation. Working from that shared foundation gives marketing and data teams a place to share trusted customer context.

Agree on: The authoritative source for each decision, the permitted use and the owner of updates at the destination.

Action item: Make one agreement useful beyond one campaign

For teams building on Databricks, the next step is to make these critical agreements reusable. Teams can document business definitions in Unity Catalog, then make them discoverable for analysis in Genie, audience building in CustomerLake, or in ML model development. Each new campaign can build on definitions that marketing and data engineering have already agreed and put into practice.

Choose a term that has caused confusion in a recent campaign. Work through one customer example with your data team, then record the decision where the next person using the data can find it.

For “customer,” that means naming what a record represents and the identifiers behind it. For “last purchase date,” it means documenting which transactions count. For “audience ready,” it means agreeing on what must be checked before launch.

Connect the agreed meaning to the dataset or workflow that implements it, and identify who should be involved when it changes. Documentation and implementation need to evolve together.

This becomes especially useful as marketers ask natural-language tools and AI agents to work with customer data. “Find our most valuable lapsed customers” still leaves wiggle room about value, inactivity, identity, and permission. Without explicit instructions, AI will make its own decisions – but a different decision every time. Giving AI tools explicit business definitions and context will produce better and more consistent results.

A useful test is whether someone joining the next campaign can understand the decision without finding the people who attended the original meeting. If they can, the team has created something it can reuse.

About this series

Helping Marketers Talk to Data Engineering is a six-part series developed in partnership with Matthew Niederberger of MartechTherapy. The 24 definitions in this article are reproduced from his independently produced guide, sponsored by Databricks.

Read Matthew’s original article, Same word, different meaning, for his perspective and the printable reference to bring into your next meeting.

Next in the series: How to write a data request someone can act on.