When you’re running paid media for a real business — not a sandbox account, not a demo — every euro carries weight. The question isn’t just where to spend, it’s how to divide your budget between what already works, what might work, and what probably won’t but needs a fair test. At Choco Media, we use a simple paid media budget allocation framework: 70% on proven campaigns, 20% on structured experiments, and 10% on long-shot tests. This post explains the logic behind that split, how we apply it across different budget sizes, and when to break the rules.
This is for brands and in-house teams running ongoing paid media, typically on Meta, Google, or TikTok. If you’re still on your first campaign and haven’t found anything that works yet, this framework doesn’t apply — come back when you have at least one campaign with 60+ days of stable data. If you’re already past that threshold, read on.
Why a fixed budget split beats gut feel
Most paid media accounts we audit fall into one of two failure modes. The first: everything goes into proven campaigns and nothing ever changes, so results slowly decay as creative fatigues and audiences saturate. The second: the account is in constant churn — new campaigns every week, no clear winners, budget spread so thin that nothing gets enough data to learn from.
The 70/20/10 split is a forcing function. It makes you protect what works (the 70), allocate real budget to structured tests (the 20), and preserve space for category-level experiments (the 10). None of those decisions get made reactively. They’re baked into how you plan each month.
- 70% — proven campaigns: campaigns with stable CPA or ROAS over at least 60 days, with creative that hasn’t fatigued
- 20% — structured experiments: new audience segments, new creative formats, or new offers tested against known benchmarks
- 10% — long-shot tests: new channels, formats you’ve never run, or high-risk high-reward creative hypotheses
The percentages aren’t sacred. What matters is that all three buckets exist and that budget is allocated deliberately, not by default.
The 70: protecting your floor
The 70% bucket is your revenue engine. These are the campaigns you’d be genuinely worried about if you had to pause them tomorrow. Treat this budget defensively: don’t tinker with it to satisfy curiosity, don’t reallocate it mid-month because something new looks promising, and don’t let it degrade quietly while you’re distracted by tests.
What belongs in the 70
A campaign belongs in the 70% bucket if it meets all three of these criteria:
- It has been running for at least 60 days with consistent targeting
- CPA or ROAS is within 15% of your target over the trailing 30 days
- Creative hasn’t fatigued — frequency is below the point where cost-per-click starts climbing
When creative in a 70% campaign starts to fatigue, refresh it within the campaign — don’t move budget. The campaign structure stays; the creative rotates. In client work we’ve found that teams who conflate creative refresh with campaign rebuild end up resetting their learning phases unnecessarily, which costs both time and money.
The 70% bucket isn’t about being conservative. It’s about protecting the compounding effect of a campaign that has earned trust from the platform’s algorithm and found its footing with an audience. Disrupting it has a real cost.
The 20: structured experiments
This is where learning happens. The 20% bucket funds tests against a known baseline — you’re not guessing, you’re comparing. Every experiment in this bucket should have a clear hypothesis, a defined success metric, and a minimum runtime before you draw conclusions.
What makes a good 20% experiment
- Hypothesis-first: “We believe X audience will convert at a lower CPA than our current segment because Y” — not “let’s try this and see”
- Defined runtime: most experiments need at least 7–14 days and a minimum of 50 conversion events before the data is directional
- Single variable: if you change audience, creative, and offer simultaneously, you learn nothing about which variable drove the result
- Clear threshold: decide before you launch what “winning” looks like — is it 10% lower CPA? 15% higher ROAS? Pick the number before you see the data
Typical 20% experiments include: new lookalike audiences built on different seed lists, creative format tests (static vs. video, UGC vs. studio), landing page variants (linked from conversion rate optimisation work), and new offer structures (free trial vs. demo, discount vs. bonus).
When to graduate an experiment to the 70
An experiment graduates when it has hit or beaten your target metric over a minimum of 30 days with enough conversion volume for statistical confidence. We typically require 100+ conversions over 30 days before calling a winner. If it hits the threshold, move it to the 70 at the start of the next planning cycle and reduce another campaign proportionally.
The 10: long-shot tests
The 10% bucket funds genuine exploration — things that haven’t worked for you yet and might not. New channels, formats you’ve never tested, or creative directions that are far enough outside your current playbook that you can’t predict the outcome.
What belongs in the 10
- A channel you’ve never run before (e.g. Pinterest, Snap, YouTube pre-roll)
- A creative format that’s structurally different from your current mix (e.g. long-form testimonial video when you’ve only ever run 15-second clips)
- A new audience category that isn’t adjacent to your existing buyers
- An offer that hasn’t been tested in paid at all
The 10% bucket is explicitly protected from results pressure. If you let the results from this bucket influence how you feel about the whole account month-to-month, you’ll kill it. Long-shot tests fail more often than they succeed. That’s fine — the ones that do work become the 20% experiments of next quarter and the 70% campaigns of next year.
In client work we’ve found that when teams remove the 10% bucket under budget pressure, accounts stagnate within 12–18 months. The 70% decays, the 20% runs out of fresh hypotheses, and there’s nothing coming up from behind. The 10% is the pipeline.
Applying the split at different budget sizes
The framework scales, but the mechanics change as total budget changes.
Under €3,000/month
At this level, the 10% is only €300 — not enough to get meaningful data from a new channel. We’d suggest redirecting it: keep 70/30, with the 30 funding sequential experiments rather than parallel ones. Run one experiment at a time, let it reach conclusion, then start the next.
- €2,100 on the proven campaign
- €900 on one structured experiment
- No 10% bucket until monthly budget crosses €5,000
€5,000–€15,000/month
The full 70/20/10 works cleanly at this level. You have enough in each bucket to run meaningful tests and still fund the proven base. This is also the range where paid media strategy decisions compound fastest — a well-structured account at €10k/month, run for 12 months with disciplined experimentation, looks very different from one that was just optimised month-to-month.
- €7,000–€10,500 on proven campaigns
- €2,000–€3,000 on structured experiments (typically 1–2 running simultaneously)
- €500–€1,500 on long-shot tests
€15,000+/month
At higher budgets, the 20% and 10% buckets become large enough to run real parallel experiments. The discipline shifts from “can we afford to test?” to “are we testing the right things?” At this scale, maintaining a test backlog — a prioritised list of hypotheses waiting for budget — becomes essential. Without it, the 20% and 10% buckets drift toward opportunistic spending rather than structured learning.
The planning cadence that makes this work
The 70/20/10 split is a monthly planning decision, not a weekly one. Here’s the cadence we use:
- Monthly: review what’s in each bucket, graduate or retire experiments, decide what the 20% and 10% funds this month
- Weekly: check for creative fatigue in the 70%, confirm experiments are on track to reach conclusion, flag any budget pacing issues
- Daily (or platform-automated): budget pacing, bid adjustments, pause any ads with obvious performance issues — no structural changes
The mistake we see most often is making structural decisions — campaign pauses, budget reallocations, audience changes — at the daily or weekly review. Daily variance in paid media results is noise. Reacting to it with structural changes burns learning data and makes it impossible to know what caused what.
When to break the rules
The 70/20/10 framework assumes a degree of stability — a consistent offer, a known audience, a business that isn’t in the middle of a major pivot. There are situations where it doesn’t apply:
- New product launch: when you’re launching something genuinely new, there’s no proven 70% bucket. Run at 50/50 (structured tests across channels and audiences) until you have 60 days of data, then shift to the standard split.
- Severe underperformance: if the 70% campaigns are underperforming badly (CPA 40%+ above target for 30+ days), fix them before structuring the rest of the budget. There’s no point running experiments on top of a broken base.
- Seasonal peaks: Black Friday, summer peaks, or product-specific seasonality may warrant temporarily increasing the 70% at the expense of the 10%, to protect performance during high-value windows.
Connecting budget structure to creative strategy
Budget allocation and creative strategy aren’t separate decisions. The 70% bucket needs a creative refresh process — you can’t protect it if the creative is burning out and you have no plan for replacing it. The 20% bucket often exists specifically to test creative variables. And the 10% bucket frequently funds creative experiments that couldn’t happen inside the proven campaign structure.
If you’re building the creative side of this infrastructure, our post on AI content creation for paid media covers how we use AI tools to produce creative volume without proportionally scaling production cost.
The budget split only works if creative supply keeps pace with it. A 20% experiment budget with nothing worth testing is just waste.
What good looks like after 12 months
A paid media account that has been run on a disciplined 70/20/10 framework for 12 months should look like this:
- The 70% campaigns are on their third or fourth creative rotation — still performing because creative has been maintained, not because you got lucky
- The 20% bucket has graduated at least 2–3 experiments into the 70%, expanding the proven base
- The 10% has produced at least one channel or format that wasn’t in the account 12 months ago
- CPA trend is flat or declining over the period, even as monthly budget has grown
That outcome doesn’t happen by accident. It’s the result of a planning process that treats budget allocation as a discipline, not a monthly negotiation.
If you’re building or auditing a paid media account and want an outside read on how your budget is structured, reach out — we’re happy to take a look.