A Meta ads provider can lower your CPA and still leave you writing six of eight ad concepts. If your calendar barely changes, what exactly did you hire them to do?
Our ranking of 8 Madgicx Alternatives, Ranked by the Hours They Leave on Your Desk can help you shortlist the kind of help you want. This is the second look: every finalist will say it saves you time, so run a 30-day test in your own account before you sign a longer commitment. Measure the hours left on your desk, the creative that actually ships, and acquisition cost against a baseline you record first. Then check whether the fee still makes sense if spend doubles.
I used to tell clients to judge a hire by its first two weeks of CPA. I was wrong. Early CPA can wobble while an account learns; hours are easier to count, and fee math decides whether an apparent win stays one. Here is the protocol I would put in front of a finalist.
Before day one: record what the job costs you now
For seven days before the provider touches the account, time every minute you spend on Meta ads. Keep four buckets:
- Work inside Ads Manager.
- Making or briefing creative.
- Reading reports.
- Answering Slack messages or email about ads.
Use a timer and round to 15 minutes. Owners spending around $5k a month may find they put in 5 to 8 hours a week once they count all four buckets. Your own total, not that estimate, is the control. Keep the same timer and categories throughout the trial. Otherwise, a finalist can claim it saved you hours you never measured.
Next, make sure both sides will read the same outcome. Confirm that your pixel and Conversions API are firing in Events Manager before spend, with deduplication through a shared event_id and Event Match Quality at or above 7. Meta cites 17.8% lower CPA for CAPI versus pixel-only setups, but this protocol has a simpler reason to check tracking: you cannot compare two periods if the counting changed between them.
Agree in writing on one purchase event, one definition, and one 7-day click window. Pull the last 30 to 60 days of CPA or ROAS, plus qualified leads or revenue rather than clicks. Record the spend behind those outcomes. Baseline hours and baseline outcomes belong on the same sheet; neither tells the full story alone.
Give the finalist a test it can actually finish
Set the terms before anyone spends a dollar: 30 days, your ad account, your pixel. Keep both in your Business Manager and grant partner access instead of sharing passwords or using an agency-owned account. Good early results are much less useful if you cannot keep the account they were built in.
Put these terms on one page:
- Access and limits: your Business Manager, partner access, verified Pixel and Conversions API, a 30-day term, and a spend cap.
- Creative responsibility: who writes hooks, who edits, and how many concepts the finalist will produce each week. Name what you still have to supply.
- Money and exit: a flat or capped test fee, no setup fee or a refundable one, and a written cancellation clause that leaves the account and data with you. Do not trade a guaranteed ROAS multiple for a lock-in.
- The read rule: the day-30 targets and the sales, leads, CPA, or ROAS used to judge them, not clicks or impressions.
If a provider will not work in an account you own or write down what it is taking off your plate, you have learned something before paying for a month of it.
Each week: count the work that actually moved
Keep logging your four hour buckets. Owner hours are the main column because that is the promise you are testing. If your baseline is six hours and week three still takes five and a half, little has been handed off, whatever the report says. On a $3k to $10k account, I would expect a useful provider to cut owner hours in half by week four: launches, kill decisions, and budget moves are now its work.
Look inside the total, too. Fewer hours in Ads Manager will not impress me if you spend the difference writing hooks and cutting variants the provider promised to make. Creative work you still do is still your time. Record concepts shipped by the finalist and concepts made by you separately, along with which ones reached spend. That makes a production gap visible instead of letting it disappear inside a lower weekly total.
Now check the work trail. Meta records changes under Account Overview > Change History and in the per-object activity history behind the clock icon; that record lasts about 90 days and has no push alerts. It tells you what moved, not why. Require a separate dated log naming who changed what and when, the reason, and the expected outcome.
The reason is not paperwork for its own sake. Significant edits can restart learning. Without a log, three edits in two days can look like unexplained performance wobble. With one, you can ask whether the change was deliberate and what spend followed it. A polished account review cannot substitute for that trail.
Creative and CPA: decide how you will read them before launch
Count concepts, not a pile of minor variants. For a small operation, I would use 4 new concepts a month as a floor; 8 is better; 12 is a machine. Note which reached spend, then wait until a concept has spent 2–3 times target CPA and had 3–4 days to deliver before judging it. On a thin budget, promising 12 concepts is less useful than getting a smaller number enough spend to read.
Have the finalist write down the creative read rule before launch. Will a concept win on CPA, ROAS, or hook rate? What will cause it to keep spending or stop? If the rule appears only after the results, every losing test can acquire a new explanation. The purpose is not to make creative mechanical. It is to stop a post-hoc story from passing as a test.
Give CPA a different timetable. Meta’s system needs approximately 50 optimization events within a 7-day window to exit learning; multiple industry sources report CPAs 20% to 50% above the eventual baseline during that period. Day-three CPA on a fresh test therefore tells you little. Compare weeks three and four with your recorded baseline, and note Learning Limited if the budget cannot buy 50 events a week. Keep logging hours and creative while you wait. They do not require you to pretend the CPA read is ready.
Finally, put the fee beside the media budget. Many Meta agencies charge 10% to 20% of spend with a minimum; a $975-or-15% arrangement costs $1,500 at $10k spend. groas runs Meta Ads for a flat $1,499 a month regardless of spend, working on audiences, budgets, and ads around the clock while testing new ads and scaling winners. At double the spend, a percentage fee rises with it; a flat fee does not. Neither fee structure excuses a weak result, but the comparison matters when you price a result you want to keep.
Day 30: pass, renegotiate, or walk away
Read the sheet, not the finalist’s presentation. For a straightforward pass, I want all three:
- Owner hours fall by at least half against the seven-day baseline, with promised creative work genuinely transferred.
- CPA or ROAS is within 20% of baseline when weeks three and four give you a usable read.
- Every material change has a dated reason in the provider’s log, so you can connect decisions with the spend and outcomes that followed.
That last condition is easy to skip when a report looks good. A monthly PDF of reach and likes without bottom-line numbers tied to actions is a report, not management. A log will not prove that every change caused an outcome. It will let you inspect what the provider did instead of crediting it for whatever happened that month.
Renegotiate if buying improves but creative still lives with you. Say your week drops from seven hours to three, yet you make six of eight concepts that month. The handoff worked in one part of the job and failed in another. Price that gap before you sign: creative billed separately can exceed the management fee, and a $15 to $40 an hour freelancer at five hours a week runs $330 to $880 before retainers. Ask the finalist to take on production or pay only for the narrower job it actually did.
Walk away if your calendar looks the same with a new logo on the invoice: hours stay flat, CPA drifts without a logged cause, or the provider presses you to raise spend quickly instead of showing what its tests found. Same hours, new logo is a failed time-saving test.
Under $5k a month: let hours lead the decision
Below $5k a month, CPA may stay noisy longer than the contract. At, say, $80 a day and a $40 target CPA, you buy about two conversions a day. Reaching 50 optimization events takes almost a month, and the account may sit in Learning Limited while it gets there. Do not call a concept a loser before it has spent enough to read, and do not treat a hot learning week as the final cost of the hire.
Here, give hours more weight. A provider that takes you from seven hours to two has delivered a measurable part of the promise even if week-four CPA remains 15% above baseline. The hours are cleanly counted; the CPA read may still be noisy. That is a reason to separate a proven time saving from an unproven performance result, not to declare a full pass or a full failure on insufficient conversions.
Watch fees just as closely. Below about $10k to $20k in spend, a management fee can take a large bite relative to media: $1,500 in ads plus a $500 part-time retainer is $2,000 out the door for $1,500 in media, with the fee equal to 33% of spend. Run fewer horses on a $3k budget: two concepts, one landing check, one budget rule, not six simultaneous tests. Splitting that budget across three audiences and four creatives in week one makes each result harder to read. A budget change beyond the 20% threshold can also reset learning, which makes casual adjustments costly on a thin account.
Two finalists: choose a clean comparison over a fast one
You can test two finalists at once only if each has a clean room. Use separate ad accounts with identical budgets and the same offer, or split geographies with similar history. Do not give two providers editing access to one test account. They can restart learning on each other’s work, leaving you to argue in week four about whose change caused what.
For a parallel test, split the budget evenly and hold the offer and landing page steady for 30 days. Log anything you cannot keep fixed. When a second variable changes mid-test, you lose the clean comparison you set up to buy.
Sequential testing is slower and, under $5k a month, the cleaner choice. Run finalist A for 30 days; record hours, creative, CPA, and the change log. Then reset to baseline settings and run finalist B for the next 30 on the same budget and conversion definition. Yes, that takes 60 days. Splitting a $3k budget into two $1.5k learning curves may leave neither side enough events for a useful read. Above $10k a month, a split may be workable if each side can still buy enough conversions; spend alone does not guarantee it.
This will not suit every account. But a finalist that refuses a contained test, partner access, and a written kill rule has already given you information you can use.
Copy this scorecard before the first edit
Fill this from your timer, the account, and the dated change log. Leave the thresholds visible so they cannot quietly move at day 30.
| Period | Owner hours: Ads / creative / reports / messages | Concepts by finalist vs. you | Changes logged with reasons | Spend, CPA/ROAS, qualified leads | Fee now vs. at 2× spend |
|---|---|---|---|---|---|
| Baseline, before test | Your seven-day total | Your normal output | — | Your 30–60-day outcomes and spend | Current cost |
| Weeks 1–2, learning | Log all four buckets | Count concepts and record the read rule | Check who, what, when, and why | Record results; do not judge early CPA as final | Confirm flat or percentage pricing |
| Weeks 3–4, read | Compare with baseline; target at least 50% fewer hours | Note what shipped and what reached spend | Check each material edit against the log | Compare usable CPA/ROAS with baseline; target within 20% | Check whether the fee still leaves room for media |
If no finalist passes, that is a result. The job as scoped may still leave you acting as the operator. Pay for a narrower promise you can verify, such as creative production alone, or test a model built around taking the work off your desk: a flat fee, work in your account within 24 hours, logged changes with reasons, and one human who owns the outcome.
I have spent enough late nights in accounts to distrust the sentence ‘We’ll save you time’ on its own. A good pitch cannot tell you who will do next week’s work. Your timer, creative count, and account log can. Run the 30 days, then sign only for the work and results your numbers show.

