What the rule is (and is not)
Every budget faces the same pull. The channels that already deliver can always absorb more money, and the case for them is easy to make because their results are visible. New channels, new audiences and new creative approaches have no results yet, so they lose every comparison with the proven work, and a budget left to that logic stops learning. The rule is a simple guard against that drift. It sets aside a share for activity that is not yet proven, and another for activity on its way to being proven, before the comparisons begin.
The rule is widely used and easy to explain, which is much of its appeal. A finance director can understand it in a sentence, and a team can defend its test budget by pointing to an agreed convention rather than arguing for each experiment from scratch.
It is worth being clear about what the rule is not. It is not a method for setting the total budget; the guide to marketing budgeting methods compares the ways of doing that, and the guide to building a marketing budget from the bottom up turns objectives into a costed plan. It is not a method for splitting money across channels, which is covered in the guide to budget allocation across channels. And its shares are not a finding. They describe a habit, not an optimum, and a team that treats them as a law has missed the point.
The three tiers
| Tier | What belongs | How it is judged | When money moves |
|---|---|---|---|
| Proven | Activity with a track record of incremental results at a known cost: core search, established social campaigns, retention email, channels with stable returns | Efficiency at the margin: what the next unit of spend returns against the target | Monthly or on a fixed rhythm, towards whichever proven item has the best marginal return, and away from items whose return has fallen below target |
| Promising | Activity that has passed a test but is not yet proven at scale: a new audience, a format or a channel showing early results | Whether results hold as spend rises, and whether an incrementality check confirms the effect | Up to proven when results hold at scale for an agreed period; out when they fail to hold after an agreed number of reviews |
| Experimental | Untested ideas: a new channel, a new market, a new creative approach, an incrementality test on an existing channel | Learning: whether the test produced a readable answer against a decision rule written before it started | At the end of each test: promote to promising, repeat with changes, or stop |
The proven tier is judged on efficiency, and its internal split should follow marginal returns rather than any fixed ratio. Every channel shows Diminishing ReturnsEach extra unit of spend or effort buys less extra result than the one before it.: each extra unit of spend buys a little less than the one before. The guide to marginal returns and spend forecasting shows how to estimate where that curve sits for each channel, and Marginal ROASExtra revenue returned by the last increment of ad spend. is the measure that should move money within this tier.
The promising tier is where most mistakes happen, because activity that looked good at small scale often weakens as spend grows. The question for this tier is not whether the activity worked once but whether it keeps working as more money goes in, and whether the results are genuinely caused by it. An IncrementalityConversions caused by marketing. test is the cleanest way to answer the second question before an item is promoted.
The experimental tier is judged on learning, not on efficiency. A test that fails clearly has done its job, as long as it answered the question it was designed to answer. Each experiment needs a hypothesis, a budget large enough to give a readable result, a duration and a decision rule, all agreed before it starts.
Adapting the shares to the business
The shares should follow the business, not the convention. A business with a long record in stable channels may need only a small experimental share, because most of what it spends has been tested many times. A young business, or one whose main channels are changing under it, may need far more, because little of what it does is proven. A business entering new markets is, for those markets, almost entirely in the experimental and promising tiers.
Budget size matters as much as maturity. A test needs a minimum spend to produce a result that can be read, and that minimum does not shrink because the total budget is small. A cleaner way to size the experimental tier is to work backwards: list the tests worth running this period, estimate what each needs to give a readable answer, and add them up. If the total is larger than the business can afford to risk, run fewer tests rather than thinner ones.
Measurement capacity sets a limit too. Each experiment needs someone to design it, watch it and read it, so a team that can properly run only a few tests at once should not fund more. Timing matters as well: experiments are best run away from peak trading, when the proven tier needs every unit of budget and results are hardest to read.
Whatever shares are chosen, they should be written into the plan with the reasoning, and protected. The guide to cross-channel performance planning shows how a test reserve sits in a one-page plan, and why it is the first thing raided when core channels look hungry.
Graduating and retiring activity
The tiers only work if activity moves between them under rules written in advance. Without rules, experiments linger because nobody wants to call them failures, promising items stay promising for years, and proven channels keep their place long after their returns have faded.
An experiment graduates to promising when it meets the decision rule set before it started, measured against the KPIA metric tied directly to an objective. the plan uses. It stops when it misses that rule, and it is repeated only when there is a clear reason to think a changed version would answer differently. Every experiment should have an end date.
A promising item graduates to proven when its results hold at a spend level close to where it would run as part of the core, over an agreed period, and when an incrementality check supports the effect. It drops back or stops if its results fade as spend rises. A cap on how long anything can stay in the promising tier stops it becoming a shelter for activity nobody wants to judge.
Proven activity can also move down. When a channel's marginal return falls below target, money comes out of it, and if the decline persists the channel's role should be reconsidered. Retiring proven activity is as much a part of the rule as promoting experiments, and it is the part teams most often skip.
Worked example (illustrative)
A bakery chain sorts its media into tiers
Barrowmint Bakeries, an invented chain of 22 bakeries with a click-and-collect site, has a media budget of £360,000 for the year. It sorts every line into a tier. Proven, at 72%: search ads on its own name and on "bakery near me" terms, and local social ads around each shop, both with two years of stable cost per order. Promising, at 18%: a placement on a food-delivery app that produced orders at £6.40 each in a three-month trial in four shops. Experimental, at 10%: a six-week sponsorship of a local food podcast, with a discount code and a before-and-after survey of brand recall in the two towns it covers. The rules are written into the plan. The delivery app graduates if cost per order stays under £8 when it runs in all 22 shops for a full quarter, and is cut if it exceeds £10 for two months running. The podcast continues only if recall rises by at least 5 points and the code brings in 300 orders. Proven spend is reviewed monthly and moved to whichever proven line has the better marginal cost per order.
Limits and criticisms
The shares are arbitrary. Nothing about the conventional split is derived from evidence, and there is no reason to think it is right for any particular business. The rule should be presented as a convention for protecting learning, never as a recipe that improves results.
A small budget cannot fund a meaningful test inside a small slice. Applying the shares mechanically to a modest total produces experiments too small to read, which waste money while appearing disciplined. Small budgets are better served by running fewer, adequately funded tests in sequence.
Tier labels can shelter weak activity. A channel called "promising" can escape scrutiny for years, and an experiment that never ends becomes a permanent line in the budget. Without written graduation and retirement rules, the tiers describe the budget without managing it.
The rule ignores marginal returns, which should decide the core split. Two proven channels can sit in the same tier while one is far past the point where the next unit of spend pays back and the other is still efficient. The rule says nothing about that; marginal analysis does, and it should override any fixed ratio inside the proven tier.
Related frameworks
The experimental tier runs best as a cycle of hypotheses, tests and decisions, which is what the experimentation loop describes, including how to score and order test ideas. Once the tiers are set, the media planning framework turns the proven and promising lines into a schedule of reach, frequency and flighting across the year.
Common mistakes
- Treating the shares as evidence. The split is a convention for protecting learning, not a proven optimum.
- Funding tests too thinly. A test that cannot produce a readable answer wastes its budget, however small.
- No graduation rules. Without written rules for promotion and retirement, items stay in their tier forever.
- Splitting the core by habit. Marginal returns, not last year's ratio, should move money within the proven tier.
- Raiding the test budget. When core channels look hungry, the experimental share is the first to go unless it is protected in the plan.
- Judging experiments on efficiency. An experiment is judged on what it taught, against a rule set before it started.
