Paid Social Without Waste: Audiences, Creative Testing & Ad Fatigue

Paid Media
Written By wevisualise designs

Paid social rarely fails in one obvious place. It leaks. Audiences bid against each other, ad sets are too small to learn anything, creative tests produce results that cannot be repeated, and frequency climbs while performance falls. This is where the budget actually goes, and how to close each gap without spending more.

The scale of the spend

IAB UK reported on 03 March 2026 that UK social media advertising reached £11.5 billion in 2025, growing 21 per cent year on year and accounting for 28 per cent of a £40.5 billion digital advertising market. Social video now makes up 59 per cent of social investment, according to the same research, conducted with Oliver Wyman.

Those numbers describe volume, not effectiveness. A growing category attracts more money per advertiser, and more money per advertiser hides structural waste that a smaller budget would have exposed. Most of the waste in a paid social account is not the result of choosing the wrong interest. It is the result of account structure, testing method and creative supply.

Audience construction & the overlap problem

Start with the seed data, because everything downstream depends on it. Meta's marketing documentation states that a Custom Audience needs at least 100 people before a lookalike audience can be built from it, and recommends at least 100 unique conversions, with 200 or more converters producing better results. It also notes that the prediction model can draw on up to 180 days of past conversion data.

Read that carefully. A lookalike built from a list of 150 mixed enquiries is technically valid and practically thin. A lookalike built from 400 paying customers, or better still from your highest value customers, is a different asset entirely. The quality of the seed sets the ceiling. Feeding the system every form fill you have ever received is the most common self inflicted wound in a paid social account.

Sizing is the second decision. Meta documents two presets: similarity, which takes the top 1 per cent of people in the selected country who most closely match the seed, and greater reach, which takes the top 5 per cent with a less precise match. Ratios can also be set manually from 1 to 20 per cent in single percentage point steps.

Here is the trap. A 1 per cent lookalike is contained inside the 5 per cent lookalike. If you run both in separate ad sets, plus an interest stack that draws from the same population, you are not covering three audiences. You are covering one audience three times and paying to compete with yourself. Meta's own optimisation guidance recommends consolidating audiences with similar targeting parameters into single ad sets rather than fragmenting them across multiple sets, precisely because fragmentation splits the data and prevents any one ad set collecting enough signal.

Why broad targeting often beats narrow

Meta's guidance is direct about narrow targeting: it "can increase your costs and lead to creative fatigue". Both halves of that sentence matter and they compound.

Cost rises because a smaller eligible pool means less supply in the auction for the same demand. Fatigue arrives faster because a small pool means the same people see the same advert sooner. You then respond to rising cost by producing new creative, which is the right response to the wrong problem, and the cost stays high because the pool is still small.

Broad targeting does not mean no targeting. Geography, language and any age or category restriction that applies to your product still belong in the ad set. Customer exclusions still belong there, so you stop paying to acquire people you already have. What broad targeting means is that you stop trying to describe your buyer with interest checkboxes and start describing them with conversion data, which the delivery system reads far more accurately than you can guess.

The learning phase, in plain arithmetic

Meta's documentation is specific: an ad set needs a minimum of 50 optimisation events over a seven day period for delivery to stabilise, and significant changes restart that process, which is why Meta recommends applying multiple edits at once rather than one at a time.

Put a real budget through that rule. An account spending £6,000 a month at a £25 cost per acquisition generates roughly 55 conversions a week in total. Spread across 12 ad sets, that is under five conversions per ad set per week. Every ad set in the account sits permanently below the threshold, delivery never settles, and the weekly cost per acquisition swings wildly for reasons no one can explain.

Consolidate the same budget into three ad sets and each one earns roughly 18 conversions a week. Consolidate into one and it clears 55. Nothing about the audience, the creative or the offer has changed. The only change is that the system now has enough events in one place to learn from.

The instinct that fights this is the desire to see performance broken out by audience. That instinct costs more than the reporting is worth. If you genuinely need audience level reads, run them as a deliberate test for a fixed period, then collapse the structure again.

How to structure a creative test that means something

Most creative tests fail before they launch, because they change several things at once and then run at a volume too small to separate the result from ordinary week to week variation.

Five rules make a test worth running.

  1. Test genuinely different concepts, not variations of one. TikTok's creative guidance for performance advertisers recommends using "creatives with big differences, especially when testing", particularly in the early exploration stage. Two hooks that differ by a word tell you nothing.
  2. Cap the number of variants against the budget. TikTok recommends 3 to 5 different creatives per ad group and 3 to 5 diversified ad groups per campaign. Any more and the arithmetic from the previous section takes over.
  3. Set the decision volume before you launch. Decide what result would change your mind. If creative A produces 14 conversions and creative B produces 11, that difference sits well inside normal weekly variation and should not be treated as a winner. If the same test runs to 90 against 60, act on it.
  4. Run at least one complete week. Weekday and weekend behaviour differ enough that a four day test flatters whichever creative happened to catch the better days.
  5. Write down the result and the reason. A test that is not recorded gets repeated within six months by someone who was not in the room.

Reading frequency & diagnosing ad fatigue

Frequency on its own is not a diagnosis. It only becomes one when read next to unique reach and click through rate. Three patterns cover almost every case.

Frequency rising, unique reach flat, click through rate falling: this is fatigue. You have exhausted the addressable pool and you are re-serving the same people.

Frequency rising, unique reach also rising, click through rate stable: this is normal. The campaign is scaling and frequency is rising because delivery is increasing, not because supply has run out.

Frequency low, click through rate falling: this is not fatigue. It is a creative, offer or audience mismatch, and adding new creative to the same concept will not fix it.

Where the platform allows it, a frequency ceiling is the cheapest control available. TikTok documents frequency caps as a way to "set the maximum number of times a user will be shown your ad within a set period of time", and target frequency as a weekly setting for how often you want your audience to see your ads. TikTok states that once a user's cap is met, impressions are redistributed to reach more unique users, which reduces "wasted impressions on users who have already seen your ad multiple times", and advises advertisers to avoid overexposing audiences to the same creative within short timeframes.

The limitation is worth knowing before you go looking for the setting. TikTok lists frequency caps as available on reach objectives, Reach & Frequency buying, Pulse Suite and video views with six or fifteen second focused view optimisation. On a standard conversion campaign, your frequency control is creative supply, not a cap.

On that point, TikTok's guidance is to refresh creative when "delivery results exhibit a consistently declining trend, or when daily new users are low", and to add fresh content to existing ad groups rather than building new ones. The reason is the learning phase again: a new ad group starts from zero.

Six leaks, their symptoms & their fixes

Leak Symptom Fix
Overlapping audiences across ad sets Cost per thousand impressions rising, one ad set stalls while a near identical one spends Consolidate ad sets with similar targeting into one
Ad sets below the event threshold Fewer than 50 conversions per ad set per week, cost per acquisition swinging Fewer ad sets, more budget in each
Narrow interest stacking High cost per thousand impressions, frequency climbing within days of launch Widen the audience, keep only necessary exclusions
Thin lookalike seeds Lookalikes perform no better than broad Rebuild seeds from paying customers, not all enquiries
Multi-variable creative tests A winner that does not repeat next month One decision per test, genuinely different concepts, a pre-set volume
Constant mid-flight edits Delivery never settles Batch edits into one change, then leave the ad set alone

What to do next

  1. Open the account and count the ad sets. Divide last month's conversions by that number, then by four. If the result is under 50 per ad set per week, consolidation is the highest value change available to you.
  2. Rebuild your primary lookalike seed from customers who paid, not everyone who filled in a form. Check the seed clears the 100 person minimum comfortably, and aim for 200 or more converters.
  3. Pull the last 30 days of frequency, unique reach and click through rate onto one chart per campaign. Diagnose against the three patterns above before you commission any new creative.
  4. Write your next creative test on one page before launching it: the single question, the variants, the volume at which you will call it, and the date you will stop.
  5. Set a rule that edits are batched weekly rather than made daily, and hold to it for a month.

Waste in paid social is structural, which is the good news. Structure is something you can change on a Tuesday afternoon without approving another penny of budget. If you want a second pair of eyes on account structure, testing method and creative supply, that is the work our paid media team does.

Sources

wevisualise designs https://wevisualise.co.uk
Next

Google Ads or Meta Ads: Choosing by Job, Not by Preference