What is a marketing engineer? The four things the title actually promises

Most job titles are promises nobody checks. "Growth" promises a number. "Marketing" promises activity. "Marketing engineer" promises something narrower and more falsifiable: that the work ends at an artifact you can open, and that the numbers on the invoice are read from something you own.

Here is the definition I work from, and the four things it commits to. It is the same on the site as it is in a contract, which is the point.

The definition, in one sentence

A marketing engineer runs acquisition the way a software team runs a product: scope is written down before the work starts, every task ends as an artifact at a public URL, a test states in advance how much volume it needs before a result means anything, and everything is built inside the client's own accounts.

Marketing is the judgement. Engineering is what makes the judgement checkable.

1. Scope written before the first send

The usual failure of an outbound or content engagement is not bad work. It is work nobody wrote down. The list was "included". The verification was "obviously included". The tracking was "the vendor's job". By week three both sides are reading the same invoice and disagreeing about what the month contains.

So the first deliverable is a scope page, not a campaign. It names: who owns the contact file, what happens to a catch-all domain, what the bounce ceiling is before the campaign pauses, which events count and in what order, the date the first verdict is due, the threshold that makes it a verdict, and — the line people skip — what is deliberately not included.

A scope that never says no is a scope that gets renegotiated.

2. Artifacts at URLs, not screenshots

The second promise: the work is inspectable without a call. Pages that load, an llms.txt that lists what is machine-readable, a published price list, a proof page that links every shipped artifact, a tool you can run in the browser rather than a claim about a tool.

This is not a portfolio preference, it is an incentive structure. If a deliverable cannot be given a URL, it is a status update, and status updates are the cheapest thing in a retainer.

3. Volume stated before the result, or the result is noise

The most expensive lie in marketing is a verdict drawn from too little data. "Subject line B won" after 400 sends of a cold campaign reads real and is not: at a 3% base rate a strict two-variant test at 80% power and alpha 0.05 needs 13,914 contacts per arm — 27,828 in total. That is not a conservative number pulled out to justify a longer contract. It is arithmetic, and it runs the same way in your spreadsheet as in mine.

What small volume is good for: a diagnostic. A 300–500 contact pass tells you whether the offer, the subject line and the segment are alive — whether anything is worth scaling. What it cannot tell you is which of two variants is better. Engagements that confuse the two end on day 12 with a "winner" that does not reproduce in month two.

4. Ownership from day one, and a handover you can read

Sending domains with SPF/DKIM/DMARC, the CRM, the automation, the tracking, the contact file, the definitions. Built under the client's logins, not the vendor's. At the end of the engagement, the test of the work is simple: can the next person pick it up without calling me?

If the answer is no, what was bought was access, not an asset — and access is the thing that expires when the contract does.

What it is not

• Not an agency retainer. Retainers rent attention and activity. Accounts and files stay with the agency, so the asset leaves when the agreement ends. Nothing wrong with that model; it is simply a different deliverable.
• Not hourly freelancing. Cheap per hour, unpriced per outcome. Scope drifts, and nothing forces the work to end at something you can open.
• Not "AI content". Automation in this job is capture, routing, suppression, alerts and kill rules — the machinery around the marketing, with a human still answering for the number.

What it costs, published

Being specific is cheaper than being vague, so the prices sit on the site next to the scope:

• Sprint — $900, one-time. One focused build in five working days: the decision, the artifact, the read date.
• Engine — $1,900/month. A running queue of experiments with a monthly read-out and a kill rule per channel.
• Full Build — $2,900, one-time. Site architecture, the page set, the machine-readable layer, tooling, distribution and handover.

If a vendor's price is not published, the first conversation is about price. If it is published, the first conversation can be about the bottleneck — which is the only conversation that decides whether you need this at all.

The test to apply to any hire, including me

Ask for four things and see what you get:

1. the scope page, with exclusions;
2. three live artifacts you can open right now;
3. the volume your test needs before a difference means anything, written before the work starts;
4. the handover list, written on day one.

Four answers, and you know what you are buying. Fewer than four, and you are buying a feeling with an invoice attached.

See how the work runs, and what a month is read from — packages, the scope page, and the artifacts, all with open links.