If everybody knows ads have to be tested, why does almost nobody do it? It is not laziness or ignorance: it is that the method, done by hand, runs into five walls that no manager —in-house or agency— can climb with the tools in the interface. Knowing them is the first step to building what is missing.
Wall 1 · It keeps no continuous history — and Google does not give you one either
Google shows you the performance of an RSA's pieces while the ad is alive: impressions per headline and a coarse label («Best», «Good», «Low»). But the moment you replace the ad, that snapshot is buried: the interface does not build the timeline of which headlines have passed through that ad group over the years, when, with what result and against whom they competed.
Rebuilding it by hand would mean capturing each piece's data periodically, in every ad group, before every change: a living archive you would have to feed every week, forever. Nobody does it. The result: every time an ad is rewritten, it is rewritten from scratch and blind, with no access to years of evidence the account itself generated and nobody kept.
Wall 2 · It does not remember mistakes — and repeats them
A direct consequence: the account does not learn. The headline that failed in 2024 looks like a good idea again in 2026, because no record of its failure survives. The price angle that has never worked in this sector gets tried again every time the manager changes. With no written memory, every person who touches the ads starts their own collection of mistakes, paying for each lesson with real budget — including the lessons the account had already paid for.
Wall 3 · It does not test angles: it rewrites what was already there
When an ad finally does get refreshed, the usual move is to paraphrase the previous one: the same argument in different words. That is not a test, it is a placebo. A real test pits different angles against each other: price against trust, urgency against information, emotional benefit against technical feature. Exploring the map of angles systematically —knowing which ones have already been tried in each ad group, which are still pending and which won each duel— demands a set of books nobody keeps in their head across 30 ad groups.
Wall 4 · It does not know which headline worked: it looks at the ad, not the pieces
The manager evaluates the whole ad («this RSA is doing well»), but the RSA is a container of up to 19 assets with very different individual performances. Inside an ad "that is doing well", excellent headlines live alongside headlines Google barely serves because they perform badly. Without going down to the detail of each piece —real impressions, label, which combinations it appears in— you end up throwing out the good with the bad or, worse, keeping the bad because it is stuck to the good.
Wall 5 · It does not have the discipline — because it is human
And even if everything above were solved, the final wall remains: consistency. The method demands reviewing the data every working day, in every ad group, for years, without the next day's work piling up and without any month "slipping by". The reality for any manager is that ads compete for their attention against infinite emergencies, and they always lose: testing is the important thing that is never urgent.
The archive sheet: what solves walls 1-4
The first four walls fall with a single tool: an archive per ad group that is fed at every review. Minimum viable version in a spreadsheet:
| Column | Content |
|---|---|
| Ad group (by ID) | Where the piece competes |
| Piece | The exact text of the headline or description |
| Type | Headline / description |
| Angle | Price · urgency · trust · novelty · benefit · social proof · specialisation · proximity |
| Start date · end date | When it went in and when it was pulled |
| Cumulative impressions | At the point it was pulled (or at the last capture) |
| Final label | Best / Good / Low |
| Verdict | Won · lost · pulled with no data |
| Against whom | The champion at the time |
With this sheet: the history exists (wall 1), losers are not repeated (wall 2), tried and pending angles are visible at a glance (wall 3) and judgement happens piece by piece (wall 4). It costs 15 minutes per ad group per review. Multiply by your ad groups and your frequency and you have the real cost of doing the method by hand — and the reason for wall 5.
Periodic capture: the minimum routine
If you do it by hand: every two weeks, for each ad group with meaningful spend, open the RSA's asset report, copy each piece's impressions and label into the sheet, note the pieces being pulled and the ones going in with their angle. Never change a piece without capturing its state first: that is the data wall 1 eats.
💡 Ninja trick: the RSA Optimizer is, above all, the archive no manager has: on every run it captures each RSA's pieces with impressions and label, keeps the permanent history of what was «Best» and what was «Low» in each ad group over time, analyses the combinations Google actually serves, detects global message patterns across the whole account and logs every optimisation. Four walls, knocked down every night; the fifth, discipline, is the one a script does not have.
What you should remember
- Five walls: no history, no memory of mistakes, paraphrases instead of angles, the ad instead of the pieces, no discipline.
- The first four fall with an archive sheet per ad group (piece, angle, dates, impressions, label, verdict).
- Never change a piece without capturing its state first.
- The fifth wall only falls with consistency — or with a machine.