The best ad creative testing workflow for a marketing agency is a fixed weekly rhythm that every client runs on identically: brief on Monday, launch on Tuesday, read at 48 hours, report on Friday. Test design is the easy part. What breaks agencies is running the same discipline across eight clients at once, so the workflow has to be a calendar with named owners, not a methodology document.
Why agencies fail at testing even when they know how
Ask any media buyer how to run a creative test and you will get a sound answer: hold the body and CTA constant, change one thing, read hook rate before cost. Then look at the same buyer's accounts. One client has a test that has been "running" for five weeks. Another launched twelve variations into four ad sets on a Thursday afternoon. A third has no test live at all because the edits are still with the freelancer.
None of that is a knowledge problem. It is a scheduling problem multiplied by the number of clients. A test that slips two days on one account is an inconvenience. Two days slipping on eight accounts, every week, means half your clients are reading tests on stale data and the other half are not testing.
The fix we use is boring on purpose: every client gets the same week, the same day for each step, and one owner per step. Clients differ in budget, platform and offer. They do not differ in when things happen.
Pick the strategy per client, then lock the rhythm
Different clients need different kinds of test. A new DTC client with no proven ads needs concept tests; a mature account with a winner going stale needs iteration. Choose once per quarter per client, write it on the client sheet, and stop re-arguing it every Monday.
| Strategy | What changes | When it fits | Watch out for |
|---|---|---|---|
| Isolated variable | One element (usually the hook), everything else fixed | Client has a working ad and needs fresh openings | Changing the voiceover "slightly" while testing hooks |
| Concept test | The whole idea: angle, structure, footage | New client, no proven ad, or a relaunch | Calling a concept dead after one hook |
| Iterative refinement | Small edits to a proven winner, one round at a time | Mature account, a winner starting to fatigue | Iterating a winner that is already past its peak |
| Dynamic creative | Platform mixes assets you supply | High budget, many assets, client wants volume over insight | Expecting it to tell you why something won |
If you want the longer argument for when the platform's mixing beats a structured test, creative testing versus dynamic creative covers it.
The week, day by day
This is the rhythm. Each step has one owner, and the owner is the same person across every client they hold.
| Day | Step | Owner | Output |
|---|---|---|---|
| Monday | Brief | Strategist | One-page brief per client: question being tested, what stays fixed, number of variations |
| Monday to Tuesday morning | Production | Editor or producer | 5 to 8 variations per test, named consistently |
| Tuesday | Launch | Media buyer | All variations in one ad set per client, equal starting conditions |
| Thursday (48 hours) | First read | Media buyer | Hook rate per variation; kill clear losers |
| Friday | Report and log | Strategist | Client update plus one line per test in the shared learnings log |
The 48 hour read is hook rate only. Typical ranges from accounts we run are 15 to 25 percent on Meta feed and 20 to 30 percent on TikTok in-feed, but judge against the client's own median, not an industry figure. Cost per result comes later, on the following week's read, once there is enough conversion volume to mean something. Reading creative test results goes through the read order in detail.
Two rules hold the calendar together. A test that misses Tuesday launches next Tuesday, not Wednesday; a late launch shortens the read window and the Friday report ends up describing noise. And production that is not ready on Tuesday morning is a capacity problem to fix, not a reason to bend the week.
Per-client test budget arithmetic
Every client's test budget is a sum, not a negotiation. Take the impressions each variation needs before a hook read is worth anything, multiply by the number of variations, and price it at the client's CPM.
A worked example, using an assumption of about 2,500 impressions per variation for a hook read:
- 6 variations x 2,500 impressions = 15,000 impressions
- At a CPM of €12, that is 15 x €12 = about €180 to reach the 48 hour read
Then add the conversion phase for the two or three survivors, which is where most of the money goes. If a client's monthly budget cannot cover one test cycle a week, test fortnightly with fewer variations rather than starving every variation of impressions. The creative testing calculator will do this sum per client so nobody does it on a napkin.
| Client monthly spend | Variations per test | Test cadence |
|---|---|---|
| Small (a test cycle uses most of a week's budget) | 3 to 5 | Fortnightly |
| Mid | 5 to 8 | Weekly |
| Large | 8 to 12, often two tests in parallel | Weekly |
These bands are how we run it in practice, not a published standard. The point is that cadence is set by the arithmetic, and the arithmetic is written down.
The learnings log is the agency's actual product
Eight clients each running a test a week produce more than 30 results a month. Most agencies lose all of them in Slack threads and slide decks. Keep one shared log across every client, one row per test:
| Date | Client | Vertical | Question | What won | Margin | Confidence |
|---|---|---|---|---|---|---|
| Week 1 | Client A | Skincare | Problem-first vs product-first hook | Problem-first | Hook rate clearly ahead | Medium |
| Week 1 | Client B | SaaS | Founder on camera vs screen recording | Screen recording | Small | Low, rerun |
After a quarter, filter by vertical before you brief a new client and you start from evidence rather than a blank page. That is the compounding advantage an agency has over an in-house team, and it only exists if somebody writes the row every Friday. More on running this across accounts in scaling creative across clients.
Reporting a test to a client who wanted a winner
Clients want to hear "variation C won, we are scaling it". Many weeks the honest result is "nothing separated" or "the winner on attention lost on cost". Report it anyway, in this order:
- The question the test asked, in one sentence.
- What the data says, with the confidence stated plainly.
- What you are doing next week because of it.
"No clear winner" is a finding: it means the variable you changed does not matter much for this audience, so next week tests a different one. What damages the relationship is not a null result. It is a report that dresses noise as a win, which the client discovers when the "winner" fails at scale.
Where production fits
The rhythm only works if Monday's brief becomes Tuesday's variations. For most agencies that is the step that slips. Genyad turns a client's existing footage library into variations with a fresh script, shot selection, voiceover and captions each, so a Monday brief can be eight variations by Monday evening. It does not publish to Meta or TikTok and does not predict which variation will win; the buyer still launches and reads the test. The agency creative workflow page shows how teams set that up per client.
Frequently asked questions
How many creative tests should an agency run per client each month?
Usually four, one a week, for any client whose budget can fund a week's test cycle. Smaller clients do better with two fortnightly tests that each get enough impressions to read, than with four starved ones.
Who should own creative testing in an agency?
Split it by step rather than by client: a strategist owns the brief and the report, a producer owns the variations, a media buyer owns launch and reads. One person owning everything for one client is how tests slip without anyone noticing.
How long should an agency creative test run?
Read hook rate at 48 hours and cost per result once survivors have meaningful conversion volume, usually the following week. Tests left running past three weeks start measuring fatigue rather than the variable you set out to test.
What should an agency do when a test has no winner?
Log it, tell the client what it means, and test a different variable next week. A null result on a hook test usually means the hook was not the constraint, which points at the offer or the body.