Don’t ask for Meta agency case studies. They are the weakest evidence in the hiring process, and most owners treat them like the strongest.

I used to send them myself. I had a deck with a 4x ROAS slide from a fashion client: clean screenshot, nice logo, dates cropped out because no one ever asked for them. That account spent $30k a month during a clearance push when past buyers flooded back on their own. Meta took credit for the sales. I took credit for Meta. If you had hired me off that slide, you would have bought a two-week spike that had little to do with how I would run your account. Judge a manager on 30 days of work inside your ad account, against a number you write down before they touch anything.

The screenshot checklist rewards the best screenshot

The standard hiring checklist sounds responsible. Demand campaign dates, ROAS and CPA numbers, at least one raw Ads Manager screenshot, and a callable reference; mark it a red flag if the agency cannot share dates or concrete numbers. I followed that script as a buyer once. I collected four decks, picked the highest ROAS, and learned nothing about who could run my account.

Here is why the filter fails: every surviving agency has a slide. Fashion and ecommerce case studies often feature a curated short-window spike, a Black Friday or clearance sale where past purchasers flood the site and the pixel takes credit, packaged as evergreen scaling. One breakdown showed a $4M streetwear brand posting 6.5x ROAS on clearance with a negative 4% net margin once the Shopify backend was checked. The screenshot was real. The profit was not.

Dates and screenshots can catch a bad sales pitch. They cannot turn someone else’s account into yours. A documented win is still a selected win.

Their 8x account is not your account

The first problem is survivor selection. You see the wins, never the cancelled accounts. An agency that ran 40 accounts last year and kept six will show you the best two of the six. The other 34 do not make the deck. I know because I lived it. My losing accounts did not get case-studied; they got a polite offboarding email and a line about seasonality.

The second problem is the ROAS number itself. It is quoted in platform currency. The default 7-day click plus 1-day view setting can make screenshots look bigger by counting view-through sales. One owner watched a reported 4x ROAS fall to 1.8x overnight after switching to 7-day click only; Meta had been claiming welcome-flow email sales. The wider gap is arithmetic, not necessarily lying: Meta counts view-through sales with no click, attributes sales to interaction date instead of order date, adds modeled iOS estimates, and does not account for returns.

The third problem is the customer mix. That 8x prospecting case can conceal this math: $10k in spend for $80k in revenue is 8x, but if $65k came from a VIP email segment, the cold-customer figure is 1.5x. Strip existing customers out and the same breakdown puts sustainable cold prospecting in fashion at 1.8 to 2.5x, not 8x. Different budget, different offer, different creative supply. None of it transfers neatly when your catalog, margins, and customer list are different.

So stop grading decks. Grade the work. The hiring question is not, “What happened for your best client?” It is, “What will you do inside my account in the first 30 days, and how will I know whether it helped?” A case study cannot answer that. Work in your account can. I learned this after losing a pitch to an agency with a prettier slide, then watching that client come back six months later with bloated spend and the same creative I had warned them was tired. The slide won. The account lost.

Run the first 30 days as a test, not a honeymoon

Question: Can this manager move my one number without breaking my one guardrail in 30 days? Write both down before they get access. Do not let the first report choose the metric after the fact. That turns a test into a search for whichever number improved.

Setup and control: Keep ownership of the ad account, Business Portfolio, Pixel, and dataset. Give the agency removable partner access and bill spend straight to your card. If they want to run campaigns through their own ad account and Pixel, walk away. You lose the audit trail, and leaving means rebuilding on your own account. Ownership is not paperwork to sort out later. It is what lets you inspect the work now and keep it afterward.

Baseline: Before handover, record the last 30 days of spend, cost per purchase or cost per lead, and total Shopify or CRM revenue net of returns. Calculate blended ROAS: total revenue net of returns divided by total ad spend across all platforms. Record contribution margin per order and new-customer CAC separately. Do not quietly substitute the Ads Manager ROAS figure when the backend number is less flattering; platform totals can double-count sales that make blended ROAS look 30 to 50% lower. I used to report platform ROAS to clients. I was wrong. The bank account runs on blended.

Judging metric and guardrail: Pick one of each, in writing. For ecommerce, I use new-customer CPA under a cap or blended ROAS above break-even. For lead gen, I use cost per qualified booked call, not cost per form fill. The guardrail covers spend and margin: do not exceed the agreed spend without sign-off; do not let contribution margin per order go negative. Judge a Meta agency on enquiries, bookings, and sales traced to ads, never reach or likes. One number to win, one line not to cross. That is the scorecard.

What to measure at day 30: Compare the judging metric with the baseline, check the guardrail, then read the change log. Ask what the manager changed and why, not merely whether a line on a chart went up. Thirty days is short; a lucky week can flatter a weak manager. The point is to make the work visible enough that you can distinguish a plausible improvement from a well-presented accident.

My expectation is not that every account produces a clean lift in a month. It is that a competent manager can establish a trustworthy baseline, make deliberate changes, and show you what happened without hiding behind a screenshot. If the number improves but the guardrail breaks, the test fails. If the number does not improve but the tracking and work are sound, you have a harder decision and better evidence than any case-study deck could give you.

By day 30, point to the work

Forget promises about month three. By day 30, you should be able to point to three things:

  1. Tracking verified. The first takeover job should be a tracking audit of Pixel, Conversions API, and deduplication, not a rush to launch ads. Check Meta’s reported sales against the backend as well. If welcome-flow email sales are inflating the platform number, you need to know before anyone calls it a campaign win.
  2. Creative shipped. Ask how many new creative concepts per month are included. That answer tells you more about month six than any case study. I want to see at least two to four new concepts live by day 30, with losers cut and a winner given more spend. A manager who only rearranges the existing ads has not shown much about what happens when those ads tire.
  3. A written change log. Not a monthly report with stock photos. A dated list of what changed, why, and what it cost. If you want a model for the questions that log should answer, read what to ask for instead of another screenshot.

No log by day 30 means no visible management. Reports describe; logs show the decisions behind them. If the manager cannot show you the five to ten moves they made and what each move was supposed to fix, you hired a spectator with an invoice. No change log, no renewal. That rule alone would have saved most of the owners I have talked to.

At $3k a month, judge discipline before lift

Say you spend $3,000 a month. That is about $750 a week. At $30 per purchase, you get roughly 25 conversions a week across the whole account. You will not prove a ROAS lift in 30 days on that volume, and anyone who promises proof is selling you math they cannot do. You can judge direction and discipline: did blended CPA move the right way, did new creative get tested and cut, and did the tracking get cleaner?

Small accounts also have little room to scatter spend. Meta’s peak-season post found that ad sets under about 50 conversions per week cost about 28% more per purchase. At 25 purchases a week account-wide, splitting the budget into six ad sets does not create more chances to learn. It spreads the same purchases thinner. I would ask the manager to explain every split and keep the structure as consolidated as the account allows.

The 50-conversion figure is a useful check on a plan, not a promise that consolidation fixes the account. At this spend, a pass is coherent work and an honest reading of a noisy result, not a victory lap over a tiny ROAS swing. One judging metric. New creative moving through the account. No invented certainty.

The contract can fail the test before the ads do

You cannot run a meaningful 30-day hiring test inside a six-month contract. That is the point of the six-month contract. Meta agencies may ask for 3 to 6 month commitments plus $500 to $2,000 in setup fees; a 2026 pricing survey describes 15 to 25% of spend with $2,000 minimums, $1,500 to $5,000 setup fees, and 6 to 12 month minimums with 30 to 90 days’ notice. I paid one of those setup fees early in my career. It bought a questionnaire and a logo slide. The work started in week five.

Percentage pricing adds another problem. When the fee is 10 to 20% of spend with minimums, agency income rises with spend, not results. Even an agency-written pricing guide calls that a conflict of interest. Recommending a higher budget can also mean recommending a raise for the agency. It does not make every budget recommendation wrong. It means the contract gives you one more reason to demand a clear metric and a clear account of what changed.

I filter for the opposite paperwork: month-to-month or 90 days maximum to start, 30-day written cancellation, client-owned accounts and data, and spend billed to the client card. A vetting scorecard flags 12-month minimums, pay-the-rest-of-term clauses, and agency-owned accounts as walk-away signs. Read the exit terms before the kickoff call. If the manager will not let you judge the first 30 days, the deck is doing the selling because the work cannot.

The one account too small for this test

If you spend $500 a month on Meta, skip the 30-day manager test. One good weekend or one bad creative can swing the whole month, and no manager can fix tracking, ship creative, and exit learning on that budget in four weeks. Pick from the options ranked by budget and choose the lightest one that fits. Then ask a reference one question: did they answer fast and leave the account cleaner than they found it? If you are buying pure creative production, judge the portfolio, not a media test.

That exception is not permission for everyone else to go back to collecting trophies. Keep hiring from case studies and you will hire the best storyteller in the pile. The 4.2x slide will look great in your inbox while your blended number sits at 1.8x and your margin pays for the difference. I kept my old deck for years because it worked. It stopped working the day owners started asking for the log instead of the logo slide. Make them earn it on your account, on your metric, in 30 days, or hire someone else.