Most hiring decisions in growth are made on vibes: a call, a deck, a promise about "compounding". Here is the boring checklist I would use as a buyer, written by someone who sells this work.
1. Ask for the last thing they shipped, as a URL
Not a case study PDF. A live URL, on a client's domain or their own, that loads right now. Then ask which number it moved and where that number lives.
If the answer is "our clients' accounts are confidential", you learned something. Confidentiality is real, which is exactly why the second-best answer matters: the same person should be able to point at their own site, their own tool, their own published pages.
2. Separate the judgement from the engineering
The job has two halves and most sellers have one of them.
- Judgement: what to sell, to whom, at what price, what to exclude.
- Engineering: whether the page is crawlable, whether the sending domain passes a deliverability checklist, whether the test can be read at all, whether the click is recorded in analytics you own.
Agencies usually have the first half and subcontract the second. Developers have the second half and guess at the first. The combination is the thing you are paying for, and it is checkable: ask what they will do about robots.txt, structured data, and how the test volume is calculated. A marketer who cannot answer the volume question has never killed a channel on evidence.
3. Demand the kill rule before the test starts
This is the single highest-value question in the whole process: how much volume does this test need before we are allowed to judge it?
Published floors for reference (they do not move much across industries):
| Test | Volume before a verdict means anything |
|---|---|
| Cold email reply rate | ~1,500-2,000 sends per variant |
| Landing page smoke test | 100-200 targeted visitors |
| Strict A/B test on conversion | ~10,000 visitors per variation |
| Paid ads | spend gate at 1-3x target CPA, 48-72h |
Anyone who draws a conclusion at 200 cold emails is selling you a story with a number attached. The rule is: write the floor down, then decide at the floor — not when the chart looks nice.
4. Check who owns the assets
Ask three questions:
- Whose domain do the pages live on?
- Whose analytics property holds the events?
- If we stop tomorrow, what do I keep?
Good answers are short: your domain, your property, everything. If the pages live on the agency's site "for speed", you are renting your own acquisition layer.
5. Price it against the alternatives, not against zero
Two comparison points that actually exist:
- Agency retainer: typically $3k-10k/month, accounts held by the agency, work reported in their dashboard.
- Hourly freelancer: $50-150/hour, scope drifts, and you carry the risk of the hours.
A fixed package with a published scope page is the third option: you buy a defined outcome, and the vendor's incentive is to ship rather than to stretch the calendar. Reference pricing for this kind of fixed-scope work, published openly so it can be compared: Sprint $900 one-time, Engine $1,900/month, Full build $2,900 one-time.
6. Score the process, not the personality
A short scorecard, one point each:
- Live URL from previous work, with a number attached.
- Named deliverable list, in writing, before the invoice.
- Pre-declared test volume and a kill rule.
- Assets in your accounts from day one.
- The last thing that got killed, and why.
Five out of five is rare. Three is workable. Zero to two means you are buying activity, and activity is the most expensive thing in marketing.
What "good" looks like in practice
The standard I hold myself to is dull on purpose: every week ends with a file on the internet that returns 200. A page with structured data. A calculator worth linking to. An event firing in your analytics. A test with its floor written down. If a month passes without a file, the month should be free — that term belongs on the offer page, not in a private negotiation.
Ask any candidate what their equivalent sentence is. Then check whether their own site proves it.
Scope and prices, published: https://axelfreeman.com/marketing-engineer.html
Free test-size calculator (browser, no signup): https://axelfreeman.github.io/marketing-mindset/tools/email-test-planner.html
Top comments (0)