learn
/
methodology
The marketing engineer: the person who builds the system marketing runs on
A marketing engineer builds the system marketing runs on: data, brain, harness and tools. What the role is, salaries up to $400K, skills and how to start.
A marketing engineer builds the system marketing runs on: data, brain, harness and tools. What the role is, salaries up to $400K, skills and how to start.
By Gabriele Franco, co-founder of Cassandra
Get one practical idea every week to grow with the same team
A marketing engineer is the person who builds the system marketing runs on. Instead of pulling reports, writing briefs and moving budgets by hand, they build the machine that does it: connected data, a brain that holds the brand's context and learns, a harness where AI agents work under rules, and tools that act in the ad accounts with approval.
What you'll find in this guide
This guide is based on my talk at GA Summit 2026, "The AI marketing system that learns what works". You can watch it here, or read on.
Companies like Figma, ClickHouse, Legora, Harvey and Profound are hiring for a role most marketers had never heard of a year ago. The posted base salaries go from $127,000 to $400,000 a year. That's roughly double what most marketing roles pay.
So the obvious question: is AI about to take our jobs, or is it creating new ones? Look at those job postings and the answer is already written. The job isn't disappearing. It's moving up one level, from doing the work to building the thing that does the work.
What is a marketing engineer?
A marketing engineer builds and runs the system that makes marketing decisions and carries them out. Their output is not a campaign or a report. Their output is the machine that produces campaigns, reports, tests and budget moves, every day, in parallel.
The easiest way to see the difference is to look at a normal week.
The marketer as executor | The marketing engineer as system builder | |
|---|---|---|
What they do | Pull the report, write the brief, shift the budget, manage campaigns in Google Ads and Meta | Define how the work should happen, once, and let the system repeat it |
Limit | The hours in their day | The quality of the system they built |
Work pattern | One task at a time, every week again | Many tasks in parallel, every day |
Responsible for | Their channels and their tasks | The outcome: growth |
Here's the shortest way I know to say it: the marketer defines how the work should happen. The system repeats it at speed.
The title itself is not brand new. Scott Brinker already listed "Marketing Engineer" as a typical title of the marketing technologist "maker" archetype in 2020. And the term marketing engineering goes back to 1998, when Gary Lilien and Arvind Rangaswamy used it for computer models that support marketing decisions: response models, segmentation, forecasting, marketing mix. What changed in 2026 is that the models can now act, not only advise.
Why are companies paying up to $400,000 for marketing engineers?
Because one person with the right system produces what used to take a team.
When a marketer delegates tasks to AI agents (in Claude Code, for example), output goes up about five times. Nothing was outsourced and nobody was hired. The work simply stopped depending on one person's hours.
But five times is the small number. The big number comes when you stop delegating tasks and start building a system that understands what to do next. More on that below: in our work it moves a team from about 1 test a month to 30.
That changes who is responsible for what. When the system runs the operations, the marketing engineer becomes responsible for the outcome of the whole growth engine. In the past that responsibility was spread across a team of six or seven people. Now it sits with one. That's why the salary is where it is.
Where did the marketing engineer come from?
Every business function went through the same sequence: first it gets data, then it gets an engineer.
2012, data engineering. Paid social took off and companies realized better decisions needed better data. Data engineering split from business intelligence at companies like Facebook and Airbnb.
2023, GTM engineering. With AI, sales operations got its engineer: the go-to-market engineer, who solves sales ops problems with automation and AI. The title is now on the order of a hundred job listings a month.
2026, marketing engineering. The same sequence reaches marketing operations. The postings already exist, often under adjacent names.
Searches tell the same story. On Google Trends (worldwide, November 2023 to July 2026), interest in "marketing engineering" stayed flat for two years and then broke away: the July 2026 peak is about five times the level of a year earlier and ten times the 2024 low.
Marketing is simply the next function to get one.
What does a marketing engineer build? The four layers
A marketing system has four layers. Think of them as a body: data is what it sees, the brain is how it thinks, the harness is how it works, the tools are its hands. Nothing above works without the layer under it.
Layer | Job | What the marketing engineer builds |
|---|---|---|
1. Data | Shows what is happening | One version of reality: ad platforms, revenue, CRM and goals connected and refreshed |
2. Brain | Decides what to do about it | The brand's context plus a learning loop that gets better every week |
3. Harness | Controls how the work runs | The place where agents work, with memory, rules, evaluation and approvals |
4. Tools | Let the system act | Permissions to read results, draft assets, change budgets, segment audiences |
Layer 1: data. Our data is still a mess
Nothing changed here. Ads live in one platform, revenue in another, goals in a spreadsheet nobody updated. The more data you have, the better decisions you can make, and the same is true for an AI agent. The first job of a marketing engineer is to give every agent one version of reality: the same numbers, updated often.
If you're starting from zero, connect the ad accounts and analytics to the AI you already use: Meta Ads, Google Ads and GA4 each take about 10 minutes.
Layer 2: the brain. Your context, and a loop that keeps improving it
In a world where intelligence can be bought per token, why are humans still relevant? Because our brain holds a much wider context than any AI agent: the brand, the competitors, the industry, every decision we made and what happened after. That's what we call judgment.
The brain layer is the attempt to write that context down so a machine can use it. Brand, competitors, industry and past decisions go in. Then a learning loop runs on top: decision, hypothesis, result, learning, and the learning feeds back into the context every day. The more learnings it has, the better its next decision. We wrote a full explainer on what a marketing brain is.
Layer 3: the harness. Where the work actually gets done
If the brain is how the system thinks, the harness is how it works. It's the environment where AI agents live: they use the brain, keep a memory, follow rules, get evaluated over time and ask for your approval before they do certain things. You delegate, the agent learns the skill, it executes. We explain this layer in what is a marketing harness.
Layer 4: the tools. Let the system act
If we stopped at the harness, we'd have a very advanced system that tells us what to do. And then we'd do the dirty work ourselves, as the servant of our own AI. The point is to close the loop: the system changes things in the real world. Tools are what let it do that.
Tool | What it does | Typical permission |
|---|---|---|
Ads | Change a budget | Needs approval |
Analytics | Read campaign results | Read only |
Creative | Draft new assets and briefs | Draft only |
CRM | Segment an audience | Read only |
Prepare a send | Needs approval |
Here's the rule: you choose what the system can touch, and what waits for you. Start everything that spends money on "needs approval". If you want the exact settings in Claude, ChatGPT, Meta and Google Ads, read how to stop an agent from changing your budget.
Why are today's marketing teams so slow?
Not because people are slow. Because every learning passes through six people and five handoffs.
Here's how long one learning loop takes in a fast team today:
An analyst collects the data. Two days later there's a meeting.
The analyst analyses it. Three days to understand it and bring it to the decision maker.
A manager decides. Two days to get the executive team aligned.
A specialist executes the change.
The market responds. Seven days to see results.
Someone analyses the results. Five more days before the CMO learns something.
19 days to learn one thing. Six owners, five handoffs, and context lost at every one. And that's a fast team.
How does a marketing engineer turn 19 days into 7?
By removing the handoffs, not by making people work faster.
Step | Today | With the system |
|---|---|---|
Data | 2 days | 2 minutes |
Analysis | 3 days | 5 minutes |
Decision | 2 days | 20 minutes (you) |
Execution | days | minutes |
Market results | 7 days | 7 days |
Learning | 5 days | 5 minutes |
Total | 19 days, 6 people | about 7 days, 1 person |
Everything before the market goes from days to minutes. The only wait left is the market itself, not your team. And because there are no handoffs, the system can run many loops in parallel instead of one at a time.
Here's what one test looks like end to end:
Observe what changed: 2 minutes.
Hypothesize why it changed: 5 minutes.
Design the experiment: 20 minutes. You approve.
Launch it into the market: 7 minutes.
Measure the lift: 5 days (that's the market, not you).
Validate the learning and decide the next step: 5 minutes. You approve.
That's 39 minutes of your time per test.
Where does the marketing engineer still have to approve?
In two places, and they are not optional.
First, the design. Before anything changes in the real world (a budget, an ad, a campaign) you approve what the system wants to do. You're responsible for the outcome, so you check that every change is the highest quality it can be.
Second, the learning. After the market responds, you check that what the system concluded matches what actually happened. This one matters more than people think: the system decides its next move from its learnings. Approve a wrong learning and every decision after it is built on sand.
A useful way to think about it: the marketing engineer is not the operator of the machine. They're its editor. They don't write every line, but nothing gets published without them.
What changes when it works: 25 times the tests
A human-paced team runs about 1.2 tests a month with 10 people involved. Mapped into a system, the same company can run up to 30 tests a month with one person approving. That's 25 times the tests with a tenth of the team.
Tests are how a marketing team learns what actually drives growth, so this is not a productivity number. It's a learning-speed number. The team that learns 25 times faster doesn't just save time, it makes better decisions every month.
What can a marketing engineer's system actually run?
Most people start with the obvious two: managing ads and designing creatives. But that's not what a marketing team does all day. Then the system takes on analytics, conversion rate optimization, competitive intelligence and reporting. And then it runs the hard ones.
Starts with | Then | Then the hard ones |
|---|---|---|
Managing ads | Analytics | Marketing mix modeling |
Designing creatives | Conversion rate optimization | Incrementality testing |
Competitive intelligence | Conversion lift | |
Forecasting | ||
Budget allocation | ||
Strategic decisions |
Think about how a company buys a new capability today, say marketing mix modeling. Vendor selection, procurement, legal, two months of negotiation, a signature from your boss, another month of onboarding. In a system, a new capability is a new skill: you add it, and it runs on the data and context the system already has. We show one in practice in building your messaging from customer words with Claude Code.
And because every company decides differently, the last piece is the custom workflow: you write down the logic you use to make a specific decision, toward a specific objective, and the system repeats it at scale.
The limit stops being your team. It becomes your imagination.
Marketing engineer vs marketing ops, GTM engineer and growth engineer
Role | Focus | Main question |
|---|---|---|
Marketing operations | Keeps the stack running: tools, integrations, lead routing, campaign setup | Is the system working? |
GTM engineer | Builds automation for the sales and RevOps stack: enrichment, outbound, CRM | Does pipeline get built without manual work? |
Growth engineer | A software engineer building product loops: sign-up, activation, retention | Does the product grow itself? |
Data analyst | Explains what happened | What do the numbers say? |
Marketing engineer | Builds the system that makes and executes marketing decisions, and owns whether they drove growth | Is the system making the right calls? |
The difference that matters: ops keeps the machine running, the marketing engineer makes sure the machine is right. That's also why the role needs measurement skills that most "AI automation" job descriptions forget. An agent that moves budget based on what Meta reports, instead of what actually caused sales, is just a faster way to make the same mistake. Knowing the difference between attribution and incrementality is part of the job.
What skills does a marketing engineer need?
Nobody starts with all of them. Most marketing engineers are strong in two or three and good enough in the rest.
Performance marketing: how Meta and Google bid, budget and report, and what moves results in practice.
Measurement: marketing mix modeling, incrementality testing, diminishing returns and the limits of attribution.
Data: connecting sources, checking that totals match, enough SQL to not be afraid of it.
AI agents: writing instructions and skills, choosing tools, setting permissions, evaluating output.
Experiment design: control groups, duration, what "inconclusive" means and how to say it.
Judgment and communication: explaining a budget move to a CFO in two sentences, with the evidence attached.
Do you need to code? Less than you think. The best marketing engineers I know came from performance marketing or analytics and learned to build with AI. Writing a clear skill is closer to writing a great brief than to writing software.
How to become a marketing engineer: a 30-day path for marketers
The search results for this question are full of advice for the old meaning of the title (an engineering degree plus sales). If you're a marketer today, this is the path I'd follow.
Week 1: connect your data. Connect Meta Ads, Google Ads and GA4 to Claude or ChatGPT, read-only. Ask it ten questions you'd normally answer with a spreadsheet, and check every number against the platform.
Week 2: automate one loop you repeat every week. The weekly report is the best first candidate: build the dashboard and the Monday report, with progress to goal, what changed, why and what to do next.
Week 3: write your first skill. Take a decision you make often and write down how you make it. Start from an existing one, like the demand map skill, and adapt it to your brand.
Week 4: run your first approved test end to end. Let the system observe, hypothesize and design. You approve the design. It runs. You approve the learning. Measure how many minutes of your time it took.
After 30 days you won't have a job title. You'll have something better: proof, in your own company, that the loop works.
How do you measure a marketing engineer?
Not by hours saved. Hours saved is the vanity metric of AI. Measure the system they built by what it does to growth:
Learning speed: days per learning loop (target: from about 19 to about 7) and tests per month.
Decision quality: how often the system's forecasts land, checked against what really happened.
Incremental growth: revenue the changes actually caused, measured with tests or a marketing mix model, not with platform-reported numbers.
Safety: changes made without approval (target: zero) and wrong learnings caught before they spread.
Is the marketing engineer a passing trend?
I don't think so, and the reason is the sequence. Data got its engineer in 2012 and data engineering never went away. Sales got its engineer in 2023 and the postings keep growing. Marketing is the function with the most money, the most repetitive decisions and the slowest learning loops. It was always going to be next.
What will change is the name. In a few years "marketing engineer" may just be what a good marketer is: someone who builds the system and owns the outcome.
Frequently asked questions
What does a marketing engineer do?
They build and run the system that makes and executes marketing decisions: connected data, a brain with the brand's context, a harness where AI agents work under rules, and tools that act in the ad accounts with approval. They are responsible for the outcome, not for individual tasks.
How much does a marketing engineer earn?
Public job postings in 2026 show base salaries from $127,000 to $400,000 a year at companies like Figma, ClickHouse, Legora, Harvey and Profound. Generic salary sites show much lower averages (PayScale lists $74,614 based on 14 profiles), because they mix in the older "technical marketing engineer" meaning of the title.
Is a marketing engineer a marketing role or an engineering role?
A marketing role that builds. The core skill is knowing what drives growth and how marketing decisions should be made. The building part is increasingly done with AI agents, skills and connectors, not with traditional software development.
What's the difference between a marketing engineer and a GTM engineer?
A GTM engineer automates the sales and RevOps stack: enrichment, outbound, CRM. A marketing engineer builds the system for marketing decisions: budgets, creatives, tests, measurement and reporting. Growth Method has a good three-way comparison with the growth engineer.
Do marketing engineers need to code?
Not necessarily. Many come from performance marketing or analytics. What they need is to write precise instructions, connect data, set permissions and judge whether the system's output is right.
Will AI agents replace marketers?
AI agents replace tasks, not the responsibility for the outcome. When the system runs the operations, someone has to define how the work should happen, approve what changes and check what was learned. That person is the marketing engineer.
Key takeaways
A marketing engineer builds the system marketing runs on. The output is the machine, not the campaign.
Companies are posting base salaries from $127K to $400K, because one person with a system replaces a team's throughput.
The system has four layers: data, brain, harness, tools. Nothing above works without the layer under it.
The real unlock is learning speed: from 19 days and 6 people per learning to about 7 days and 1 person, and up to 25 times the tests.
Two approvals are not optional: the design of every change, and every learning.
Measure the role by learning speed and incremental growth, never by hours saved.
Build it without starting from zero
Marketing decisions should not rely on one data source, and a marketing engineer shouldn't have to build the whole system from scratch. Cassandra is the workbench for the marketing engineer: it connects your ad accounts, analytics, sales, products and customer reviews into one brain, runs the learning loop with incrementality tests and marketing mix modeling, and does the work in your ad accounts within the approvals you set. You can start for free with your website and data.
Read next
A marketing engineer is the person who builds the system marketing runs on. Instead of pulling reports, writing briefs and moving budgets by hand, they build the machine that does it: connected data, a brain that holds the brand's context and learns, a harness where AI agents work under rules, and tools that act in the ad accounts with approval.
What you'll find in this guide
This guide is based on my talk at GA Summit 2026, "The AI marketing system that learns what works". You can watch it here, or read on.
Companies like Figma, ClickHouse, Legora, Harvey and Profound are hiring for a role most marketers had never heard of a year ago. The posted base salaries go from $127,000 to $400,000 a year. That's roughly double what most marketing roles pay.
So the obvious question: is AI about to take our jobs, or is it creating new ones? Look at those job postings and the answer is already written. The job isn't disappearing. It's moving up one level, from doing the work to building the thing that does the work.
What is a marketing engineer?
A marketing engineer builds and runs the system that makes marketing decisions and carries them out. Their output is not a campaign or a report. Their output is the machine that produces campaigns, reports, tests and budget moves, every day, in parallel.
The easiest way to see the difference is to look at a normal week.
The marketer as executor | The marketing engineer as system builder | |
|---|---|---|
What they do | Pull the report, write the brief, shift the budget, manage campaigns in Google Ads and Meta | Define how the work should happen, once, and let the system repeat it |
Limit | The hours in their day | The quality of the system they built |
Work pattern | One task at a time, every week again | Many tasks in parallel, every day |
Responsible for | Their channels and their tasks | The outcome: growth |
Here's the shortest way I know to say it: the marketer defines how the work should happen. The system repeats it at speed.
The title itself is not brand new. Scott Brinker already listed "Marketing Engineer" as a typical title of the marketing technologist "maker" archetype in 2020. And the term marketing engineering goes back to 1998, when Gary Lilien and Arvind Rangaswamy used it for computer models that support marketing decisions: response models, segmentation, forecasting, marketing mix. What changed in 2026 is that the models can now act, not only advise.
Why are companies paying up to $400,000 for marketing engineers?
Because one person with the right system produces what used to take a team.
When a marketer delegates tasks to AI agents (in Claude Code, for example), output goes up about five times. Nothing was outsourced and nobody was hired. The work simply stopped depending on one person's hours.
But five times is the small number. The big number comes when you stop delegating tasks and start building a system that understands what to do next. More on that below: in our work it moves a team from about 1 test a month to 30.
That changes who is responsible for what. When the system runs the operations, the marketing engineer becomes responsible for the outcome of the whole growth engine. In the past that responsibility was spread across a team of six or seven people. Now it sits with one. That's why the salary is where it is.
Where did the marketing engineer come from?
Every business function went through the same sequence: first it gets data, then it gets an engineer.
2012, data engineering. Paid social took off and companies realized better decisions needed better data. Data engineering split from business intelligence at companies like Facebook and Airbnb.
2023, GTM engineering. With AI, sales operations got its engineer: the go-to-market engineer, who solves sales ops problems with automation and AI. The title is now on the order of a hundred job listings a month.
2026, marketing engineering. The same sequence reaches marketing operations. The postings already exist, often under adjacent names.
Searches tell the same story. On Google Trends (worldwide, November 2023 to July 2026), interest in "marketing engineering" stayed flat for two years and then broke away: the July 2026 peak is about five times the level of a year earlier and ten times the 2024 low.
Marketing is simply the next function to get one.
What does a marketing engineer build? The four layers
A marketing system has four layers. Think of them as a body: data is what it sees, the brain is how it thinks, the harness is how it works, the tools are its hands. Nothing above works without the layer under it.
Layer | Job | What the marketing engineer builds |
|---|---|---|
1. Data | Shows what is happening | One version of reality: ad platforms, revenue, CRM and goals connected and refreshed |
2. Brain | Decides what to do about it | The brand's context plus a learning loop that gets better every week |
3. Harness | Controls how the work runs | The place where agents work, with memory, rules, evaluation and approvals |
4. Tools | Let the system act | Permissions to read results, draft assets, change budgets, segment audiences |
Layer 1: data. Our data is still a mess
Nothing changed here. Ads live in one platform, revenue in another, goals in a spreadsheet nobody updated. The more data you have, the better decisions you can make, and the same is true for an AI agent. The first job of a marketing engineer is to give every agent one version of reality: the same numbers, updated often.
If you're starting from zero, connect the ad accounts and analytics to the AI you already use: Meta Ads, Google Ads and GA4 each take about 10 minutes.
Layer 2: the brain. Your context, and a loop that keeps improving it
In a world where intelligence can be bought per token, why are humans still relevant? Because our brain holds a much wider context than any AI agent: the brand, the competitors, the industry, every decision we made and what happened after. That's what we call judgment.
The brain layer is the attempt to write that context down so a machine can use it. Brand, competitors, industry and past decisions go in. Then a learning loop runs on top: decision, hypothesis, result, learning, and the learning feeds back into the context every day. The more learnings it has, the better its next decision. We wrote a full explainer on what a marketing brain is.
Layer 3: the harness. Where the work actually gets done
If the brain is how the system thinks, the harness is how it works. It's the environment where AI agents live: they use the brain, keep a memory, follow rules, get evaluated over time and ask for your approval before they do certain things. You delegate, the agent learns the skill, it executes. We explain this layer in what is a marketing harness.
Layer 4: the tools. Let the system act
If we stopped at the harness, we'd have a very advanced system that tells us what to do. And then we'd do the dirty work ourselves, as the servant of our own AI. The point is to close the loop: the system changes things in the real world. Tools are what let it do that.
Tool | What it does | Typical permission |
|---|---|---|
Ads | Change a budget | Needs approval |
Analytics | Read campaign results | Read only |
Creative | Draft new assets and briefs | Draft only |
CRM | Segment an audience | Read only |
Prepare a send | Needs approval |
Here's the rule: you choose what the system can touch, and what waits for you. Start everything that spends money on "needs approval". If you want the exact settings in Claude, ChatGPT, Meta and Google Ads, read how to stop an agent from changing your budget.
Why are today's marketing teams so slow?
Not because people are slow. Because every learning passes through six people and five handoffs.
Here's how long one learning loop takes in a fast team today:
An analyst collects the data. Two days later there's a meeting.
The analyst analyses it. Three days to understand it and bring it to the decision maker.
A manager decides. Two days to get the executive team aligned.
A specialist executes the change.
The market responds. Seven days to see results.
Someone analyses the results. Five more days before the CMO learns something.
19 days to learn one thing. Six owners, five handoffs, and context lost at every one. And that's a fast team.
How does a marketing engineer turn 19 days into 7?
By removing the handoffs, not by making people work faster.
Step | Today | With the system |
|---|---|---|
Data | 2 days | 2 minutes |
Analysis | 3 days | 5 minutes |
Decision | 2 days | 20 minutes (you) |
Execution | days | minutes |
Market results | 7 days | 7 days |
Learning | 5 days | 5 minutes |
Total | 19 days, 6 people | about 7 days, 1 person |
Everything before the market goes from days to minutes. The only wait left is the market itself, not your team. And because there are no handoffs, the system can run many loops in parallel instead of one at a time.
Here's what one test looks like end to end:
Observe what changed: 2 minutes.
Hypothesize why it changed: 5 minutes.
Design the experiment: 20 minutes. You approve.
Launch it into the market: 7 minutes.
Measure the lift: 5 days (that's the market, not you).
Validate the learning and decide the next step: 5 minutes. You approve.
That's 39 minutes of your time per test.
Where does the marketing engineer still have to approve?
In two places, and they are not optional.
First, the design. Before anything changes in the real world (a budget, an ad, a campaign) you approve what the system wants to do. You're responsible for the outcome, so you check that every change is the highest quality it can be.
Second, the learning. After the market responds, you check that what the system concluded matches what actually happened. This one matters more than people think: the system decides its next move from its learnings. Approve a wrong learning and every decision after it is built on sand.
A useful way to think about it: the marketing engineer is not the operator of the machine. They're its editor. They don't write every line, but nothing gets published without them.
What changes when it works: 25 times the tests
A human-paced team runs about 1.2 tests a month with 10 people involved. Mapped into a system, the same company can run up to 30 tests a month with one person approving. That's 25 times the tests with a tenth of the team.
Tests are how a marketing team learns what actually drives growth, so this is not a productivity number. It's a learning-speed number. The team that learns 25 times faster doesn't just save time, it makes better decisions every month.
What can a marketing engineer's system actually run?
Most people start with the obvious two: managing ads and designing creatives. But that's not what a marketing team does all day. Then the system takes on analytics, conversion rate optimization, competitive intelligence and reporting. And then it runs the hard ones.
Starts with | Then | Then the hard ones |
|---|---|---|
Managing ads | Analytics | Marketing mix modeling |
Designing creatives | Conversion rate optimization | Incrementality testing |
Competitive intelligence | Conversion lift | |
Forecasting | ||
Budget allocation | ||
Strategic decisions |
Think about how a company buys a new capability today, say marketing mix modeling. Vendor selection, procurement, legal, two months of negotiation, a signature from your boss, another month of onboarding. In a system, a new capability is a new skill: you add it, and it runs on the data and context the system already has. We show one in practice in building your messaging from customer words with Claude Code.
And because every company decides differently, the last piece is the custom workflow: you write down the logic you use to make a specific decision, toward a specific objective, and the system repeats it at scale.
The limit stops being your team. It becomes your imagination.
Marketing engineer vs marketing ops, GTM engineer and growth engineer
Role | Focus | Main question |
|---|---|---|
Marketing operations | Keeps the stack running: tools, integrations, lead routing, campaign setup | Is the system working? |
GTM engineer | Builds automation for the sales and RevOps stack: enrichment, outbound, CRM | Does pipeline get built without manual work? |
Growth engineer | A software engineer building product loops: sign-up, activation, retention | Does the product grow itself? |
Data analyst | Explains what happened | What do the numbers say? |
Marketing engineer | Builds the system that makes and executes marketing decisions, and owns whether they drove growth | Is the system making the right calls? |
The difference that matters: ops keeps the machine running, the marketing engineer makes sure the machine is right. That's also why the role needs measurement skills that most "AI automation" job descriptions forget. An agent that moves budget based on what Meta reports, instead of what actually caused sales, is just a faster way to make the same mistake. Knowing the difference between attribution and incrementality is part of the job.
What skills does a marketing engineer need?
Nobody starts with all of them. Most marketing engineers are strong in two or three and good enough in the rest.
Performance marketing: how Meta and Google bid, budget and report, and what moves results in practice.
Measurement: marketing mix modeling, incrementality testing, diminishing returns and the limits of attribution.
Data: connecting sources, checking that totals match, enough SQL to not be afraid of it.
AI agents: writing instructions and skills, choosing tools, setting permissions, evaluating output.
Experiment design: control groups, duration, what "inconclusive" means and how to say it.
Judgment and communication: explaining a budget move to a CFO in two sentences, with the evidence attached.
Do you need to code? Less than you think. The best marketing engineers I know came from performance marketing or analytics and learned to build with AI. Writing a clear skill is closer to writing a great brief than to writing software.
How to become a marketing engineer: a 30-day path for marketers
The search results for this question are full of advice for the old meaning of the title (an engineering degree plus sales). If you're a marketer today, this is the path I'd follow.
Week 1: connect your data. Connect Meta Ads, Google Ads and GA4 to Claude or ChatGPT, read-only. Ask it ten questions you'd normally answer with a spreadsheet, and check every number against the platform.
Week 2: automate one loop you repeat every week. The weekly report is the best first candidate: build the dashboard and the Monday report, with progress to goal, what changed, why and what to do next.
Week 3: write your first skill. Take a decision you make often and write down how you make it. Start from an existing one, like the demand map skill, and adapt it to your brand.
Week 4: run your first approved test end to end. Let the system observe, hypothesize and design. You approve the design. It runs. You approve the learning. Measure how many minutes of your time it took.
After 30 days you won't have a job title. You'll have something better: proof, in your own company, that the loop works.
How do you measure a marketing engineer?
Not by hours saved. Hours saved is the vanity metric of AI. Measure the system they built by what it does to growth:
Learning speed: days per learning loop (target: from about 19 to about 7) and tests per month.
Decision quality: how often the system's forecasts land, checked against what really happened.
Incremental growth: revenue the changes actually caused, measured with tests or a marketing mix model, not with platform-reported numbers.
Safety: changes made without approval (target: zero) and wrong learnings caught before they spread.
Is the marketing engineer a passing trend?
I don't think so, and the reason is the sequence. Data got its engineer in 2012 and data engineering never went away. Sales got its engineer in 2023 and the postings keep growing. Marketing is the function with the most money, the most repetitive decisions and the slowest learning loops. It was always going to be next.
What will change is the name. In a few years "marketing engineer" may just be what a good marketer is: someone who builds the system and owns the outcome.
Frequently asked questions
What does a marketing engineer do?
They build and run the system that makes and executes marketing decisions: connected data, a brain with the brand's context, a harness where AI agents work under rules, and tools that act in the ad accounts with approval. They are responsible for the outcome, not for individual tasks.
How much does a marketing engineer earn?
Public job postings in 2026 show base salaries from $127,000 to $400,000 a year at companies like Figma, ClickHouse, Legora, Harvey and Profound. Generic salary sites show much lower averages (PayScale lists $74,614 based on 14 profiles), because they mix in the older "technical marketing engineer" meaning of the title.
Is a marketing engineer a marketing role or an engineering role?
A marketing role that builds. The core skill is knowing what drives growth and how marketing decisions should be made. The building part is increasingly done with AI agents, skills and connectors, not with traditional software development.
What's the difference between a marketing engineer and a GTM engineer?
A GTM engineer automates the sales and RevOps stack: enrichment, outbound, CRM. A marketing engineer builds the system for marketing decisions: budgets, creatives, tests, measurement and reporting. Growth Method has a good three-way comparison with the growth engineer.
Do marketing engineers need to code?
Not necessarily. Many come from performance marketing or analytics. What they need is to write precise instructions, connect data, set permissions and judge whether the system's output is right.
Will AI agents replace marketers?
AI agents replace tasks, not the responsibility for the outcome. When the system runs the operations, someone has to define how the work should happen, approve what changes and check what was learned. That person is the marketing engineer.
Key takeaways
A marketing engineer builds the system marketing runs on. The output is the machine, not the campaign.
Companies are posting base salaries from $127K to $400K, because one person with a system replaces a team's throughput.
The system has four layers: data, brain, harness, tools. Nothing above works without the layer under it.
The real unlock is learning speed: from 19 days and 6 people per learning to about 7 days and 1 person, and up to 25 times the tests.
Two approvals are not optional: the design of every change, and every learning.
Measure the role by learning speed and incremental growth, never by hours saved.
Build it without starting from zero
Marketing decisions should not rely on one data source, and a marketing engineer shouldn't have to build the whole system from scratch. Cassandra is the workbench for the marketing engineer: it connects your ad accounts, analytics, sales, products and customer reviews into one brain, runs the learning loop with incrementality tests and marketing mix modeling, and does the work in your ad accounts within the approvals you set. You can start for free with your website and data.
Read next
Your marketing brain.
Cassandra is the AI marketing brain for growth teams. It learns what drives growth with marketing mix modeling and incrementality tests, then does the work in your ad accounts, within your rules.
Product
Company
Copyright © 2026 – All Rights Reserved
Your marketing brain.
Cassandra is the AI marketing brain for growth teams. It learns what drives growth with marketing mix modeling and incrementality tests, then does the work in your ad accounts, within your rules.
Product
Company
Copyright © 2026 – All Rights Reserved