Six questions that separate a freelancer from the person who owns the number

The freelancer-versus-engineer question is usually asked as a price question. It is not. A freelancer sells hours in a craft you can judge. A marketing engineer sells ownership of a number you can read, which includes the channel work, the tracking and the report around it.

Those are different purchases, and paying more for the wrong one is how a company ends up with beautiful deliverables and no idea whether anything worked.

Here are six questions you can ask either kind of hire. Every one of them has an answer you can check without trusting the person.

1. "What number will this move, and what was it before?"

A freelancer answers with the deliverable: twelve pages, a design system, a video edit. That is a fine answer when the craft is what you are buying.

An engineer answers with the baseline and the event. Not "we will improve conversion" — the number, its current value, and the exact event that counts as a result. If nobody can name the baseline, there is no way to tell afterwards whether the money was well spent, and no amount of additional hours fixes that.

2. "What does the first read look like, and on what date?"

If the answer is "after the campaign", there is no measurement plan — only a hope with a budget attached.

A read date is a commitment. It also forces the uncomfortable arithmetic: a strict two-variant test at a 3% base rate needs 13,914 contacts per arm, 27,828 in total, at 80% power and alpha 0.05. At 500 contacts a day that is about 56 days. A monthly report that claims three significant wins at that volume is worth reading twice.

Compute the floor before the work starts, not after. The email test planner uses the same two-proportion formula you can run yourself, so the number on the slide and the number in the terminal agree.

3. "What would make you kill this?"

A written stop rule is the cheapest line item on any invoice. Without one, the test never ends — it just gets quieter.

For a paid test on the same offer the rule is blunt: 1–3× the target cost per order with no readable return in 48–72 hours. For owned channels it is the read date plus a threshold. Either way it is written before the first contact, and it names the person who is allowed to switch the channel off. A channel that is never allowed to be switched off is not a test, it is a subscription.

The kill-rule calculator turns those decisions into the two lines you paste into the plan.

4. "What do I keep if we stop working together?"

This is the question that separates the two purchases cleanly.

An hourly engagement ends when the invoice stops. That is honest and often correct — as long as the artifact is the point. A twelve-page set of landing pages is yours; it does not need the person to keep existing.

Ownership of a number ends differently. What should remain: the pages, the tracking, the automation, the credentials, the definition of a lead, and the written record of what was killed and why. The test of a good handover is that the next person can pick it up without a call. What the scope covers is written out for exactly that reason.

5. "Where can I see your work without asking you?"

Public, dated artifacts on a domain the person controls. Not a slide with a logo wall, not "happy to share under NDA".

This is also the cheapest filter on the buyer's side: if the artifact list is empty, the reported results are unverifiable. Mine is openable — live pages, working tools, public repositories, dates on everything.

6. "Who writes the copy, the schema and the tracking?"

One accountable builder. Four vendors for one funnel is how the number gets lost: the copywriter optimises the paragraph, the analytics person tracks the click, nobody owns the event the sales team actually acts on.

At small volume one person can genuinely own the channel, the measurement stack and the report. The failure mode is not capacity — it is hiring someone who cannot wire an API, read a sample size, or say no to a test that cannot possibly finish.

So which one should you hire

Hire the freelancer when: the job is one craft and it is finite; you already own the system; the measurement question is answered and you can read the number the same day; you are buying output, not a feedback loop.

Hire the owner of the number when: reports exist but nobody can say what a result cost; the channel keeps restarting with a new tactic every month and no baseline; the handoff itself is the deliverable.

What it costs, in public

Numbers beat adjectives here as everywhere. Fixed and published: Sprint $900 for one focused build in five working days, Engine $1,900 a month for a running queue of experiments with a monthly read-out, Full Build $2,900 for site architecture, the page set, the machine-readable layer, tooling, distribution and handover. Compare any offer the same way — the cost breakdown lists what lands and what does not, and how this differs from an agency if you are weighing a team instead.

Ask the six questions in the first call. The answers decide the purchase faster than any proposal.

What the work includes, and how a month is read.