No published source gives a number for how long it takes to build a brand. Not one. Searching for it returns agency posts offering four to six months, six to twelve months, or “at least five years”, none of which cite anything.

What does exist is a single figure, repeated everywhere, that is not what people think it is: the claim that brand effects take about six months to appear. Its actual source is a 2013 report by Les Binet and Peter Field for the Institute of Practitioners in Advertising, and the sentence reads:

“long-term advertising effects on sales uplifts only start to dominate short-term effects after six months. That is to say brand building takes over as the primary driver of growth from sales activation after six months.”

Read that carefully. It says one aggregated curve crosses another at month six. It does not say anything becomes visible at month six that was invisible at month five. It is a crossover point, not a latency.

The distinction matters enormously if you are deciding whether to keep funding something.

Where the data comes from

The IPA Databank is the most cited evidence base in marketing effectiveness, and its provenance is stated plainly in the reports themselves. From Media in Focus (2017), page 8:

“the data source is the IPA Databank - the confidential data submitted alongside entries to the biennial IPA Effectiveness Awards competition”

So the population is campaigns that an agency chose to enter into a paid effectiveness awards competition. That is a selected sample by construction, and the authors do not hide it.

The second feature is less often quoted and matters more. From the same report, page 9:

“Case study authors assess these measures on a four-point scale of magnitude: only top-box scores (i.e. ‘very large’) are used to identify high performers.”

The people submitting the case grade the size of their own result, on four levels, and the analysis keeps the top level. The researchers did not measure the effects. The entrants rated them. As a partial counterweight, 58 percent of the 2014 and 2016 cases were backed by a full econometric model.

The authors defend the sample directly, and the defense is worth reading in full because it is more honest than most of what gets built on top of it:

“Although this still means that we have a biased sample of campaigns that tend to be successful, there is a wealth of variation in the effectiveness data, including some very mediocre performances. We know from comparisons of efficiency metrics with non-biased samples from other studies that our results are not greatly different, which leads us to believe that our sample is more average than many think.”

Not everyone accepts that. Byron Sharp of the Ehrenberg-Bass Institute put it bluntly in 2022: “they analysed a very weird data set, which is award submissions. If you wanted to solve this issue, you would never use that data.” A strategist at BBH London made the same point in 2018: “every one of those 996 campaigns is a success in its own right. Nobody’s entering dross into the IPA.”

The four selection filters between a real campaign and the IPA Databank datasetThe four selection filters that sit between a real marketing campaign and its appearance in the IPA Databank, the most cited evidence base in marketing effectiveness. The first filter is that the campaign must have been run by an agency that participates in the biennial Institute of Practitioners in Advertising Effectiveness Awards competition. The second filter is that the agency must have chosen to enter that specific campaign, which normally means the agency believed the campaign worked, since entries are paid submissions to an awards competition. The third filter is that the agency must have written a full case study documenting the campaign and its results. The fourth filter is that the magnitude of the effect is assessed by the case study authors themselves on a four point scale of magnitude, and only top box scores, meaning very large, are used to identify high performers, so the outcome variable is graded by the party with an interest in the grade. The reports state as a partial counterweight that fifty eight percent of the cases submitted in two thousand fourteen and two thousand sixteen were backed by a full econometric model. The authors acknowledge the resulting bias, writing that this still means they have a biased sample of campaigns that tend to be successful, while arguing that there is a wealth of variation in the effectiveness data including some very mediocre performances, and that comparisons of efficiency metrics with unbiased samples from other studies suggest their results are not greatly different.Four filters between a campaign and the dataset1. Agency entersthe competitionA paid biennialeffectiveness awardscompetition2. Agency picksthis campaignNobody submits workthey believe failed3. Agency writesthe case studyThe narrative isauthored by the partybeing evaluated4. Author gradesthe effect size”a four-point scale ofmagnitude: only top-boxscores are used”Partial counterweight stated in the report58 percent of the 2014 and 2016 cases were backed by a full econometric model.What this does not permitAny statement about campaigns that were never entered, which is nearly all of them.
Four filters sit between a campaign and the dataset. Each one selects for success, and the last one is self-scored. Source : Binet and Field, Media in Focus, IPA, 2017, pp. 8-9 (2017)

What the Databank does show, honestly read

Taken as what it is, the pattern is consistent and worth knowing. Across cases from 1998 to 2016:

  • Very large market share effects appear in 3 percent of cases evaluated over zero to six months, against 38 percent of cases evaluated over more than thirty months.
  • Cases running longer than six months generate about 4.6 times more market share growth than short cases.
  • The relationship inverts for activation: 65 percent of short cases produce very large activation effects, against 33 percent of cases of three years or more.
  • The advantage of emotional over rational campaigns on very large profit effects widens with time: 13 against 10 percent at one year, 30 against 20 at two years, 43 against 23 at three years and beyond.

Now the circularity problem, which nobody who quotes these numbers mentions. A campaign evaluated over three years is, by construction, a campaign that survived three years. Its budget was renewed. Someone judged it good enough to write up and enter. The correlation between long duration and large effect is partly a mechanical consequence of how cases arrive in the dataset, not a discovery about how brands work.

The randomized evidence says something harder

There is one large body of genuinely experimental work on this question, and it is thirty years old and rarely cited in brand strategy decks.

In 1995, Marketing Science published a summary of 55 in-market split-cable experiments. Households in the same market were matched and randomly assigned to receive different television advertising weights, for established consumer packaged goods, in real stores. This is randomized assignment, not a sample of self-reported success stories.

Two findings, and the second one is the uncomfortable one:

  1. Where a weight increase produced a significant effect in year one, the cumulative impact over the next two years was about double the year one effect. Long-term effects are real and roughly triple the first-year figure in total.
  2. Where there was no significant effect in year one, there was none in years two and three either.

The second finding directly contradicts the most common defense of an underperforming brand campaign. “Brand effects take time, be patient” is not supported by the only large randomized experiment on the subject. In that data, patience paid off exclusively where something was already measurable.

Findings of fifty five randomized split cable experiments on long term television advertising effectsFindings of fifty five in market split cable experimental estimates of the long term effect of television advertising, summarized in Marketing Science in 1995. The design randomly assigned matched households in the same market to receive different television advertising weights, for established consumer packaged goods brands, measured through real store purchasing. Two results follow. In the first case, where an increase in television weight produced a statistically significant effect on sales in year one, the cumulative impact over the following two years was approximately double the year one effect, so the total effect across three years was roughly three times the first year figure. In the second case, where there was no statistically significant effect in year one, there was no significant effect in years two and three either. The second result matters because it contradicts the common defense of an underperforming brand campaign, namely that brand effects take time and require patience. In this randomized evidence, patience was rewarded only where an effect was already measurable in the first year. This body of work is unusual in the marketing effectiveness literature because assignment was randomized rather than observed, so the sample does not consist of campaigns selected for having succeeded.55 randomized experiments, two outcomesSignificant effect in year 1about 2xcumulative effect over the next two years,relative to the year one effectTotal across three years is roughly threetimes the first year figure.No significant effect in year 1nothingin years two and three either”Be patient, brand effects take time” isnot supported here. Patience paid offonly where something was measurable.Design: matched households randomly assigned to different TV weights, established packaged goods brands,real store purchasing. Randomized assignment, not a sample of campaigns selected for having succeeded.
Long-term effects are real and roughly double the first year. But they only appeared where the first year already showed something. Source : Lodish, Abraham, Livelsberger, Lubetkin, Richardson and Stevens, Marketing Science 14(3), 1995, G133-G140 (1995)

The 60:40 rule does not survive contact with your sector

The companion claim to the six month figure is that roughly 60 percent of budget should go to brand building and 40 percent to activation. In 2018 the authors refined it to 62:38, with a footnote reading “Unlikely to be statistically significant from 60:40.”

The problem is not the precision. It is that the same report publishes a table of sector optima, and the spread destroys the single number:

SectorBrand to activation
Financial services80:20
Non-automotive durables69:31
Retail64:36
Packaged goods, non-food65:35
Packaged goods, food56:44
Durables58:42
Telecoms and internet providers58:42
Other services51:49
Perishable services48:52

From 80:20 to 48:52. And the authors add that “the IPA data cannot reliably be cut at the individual category level”, which is an unusual thing to publish next to a table of category-level numbers.

For business to business specifically, the same authors published a separate estimate in 2019: “Efficiency appears to be maximised when around 46% of the budget is allocated to brand, with around 54% allocated to activation.” Note that this is 46:54, not the 50:50 that circulates in summaries.

Their own caveat on that number is the most important sentence in the report:

“There are still relatively few B2B cases in The Databank, so sample sizes are small, at less than 50 cases. And those cases may not be typical of B2B marketing in general, because the IPA Databank is biased towards effective campaigns… Geographically, they are skewed towards the UK… They also tend to have relatively big budgets.”

Fewer than fifty self-selected cases, mostly British, mostly well funded. That is the entire evidentiary basis for the B2B budget split repeated across the industry.

Optimal brand to activation budget split by sector, with the business to business estimateOptimal brand building to activation budget splits by sector, published alongside the general sixty forty rule by the same authors, showing that the single figure conceals a very wide spread. Financial services is eighty percent brand to twenty percent activation. Non automotive durables is sixty nine to thirty one. Non food packaged goods is sixty five to thirty five. Retail is sixty four to thirty six. Food packaged goods is fifty six to forty four. Durables is fifty eight to forty two. Telecommunications and internet service providers is fifty eight to forty two. Other services is fifty one to forty nine. Perishable services is forty eight to fifty two, which is the only sector where activation outweighs brand building. Automotive is reported as having insufficient data. The authors state that the IPA data cannot reliably be cut at the individual category level, which sits awkwardly beside a published table of category level numbers. Separately, a two thousand nineteen report by the same authors for the business to business institute estimates that efficiency appears to be maximised when around forty six percent of the budget is allocated to brand with around fifty four percent allocated to activation, and the authors caution that this ratio should not be followed too precisely and that their small sample only allows a rough estimate, since there are fewer than fifty business to business cases in the databank, those cases may not be typical because the databank is biased toward effective campaigns, they are skewed toward the United Kingdom, and they tend to have relatively big budgets.One rule, ten different answersShare of budget to brand building, by sectorFinancial services80:20Non-automotive durables69:31Packaged goods, non-food65:35Retail64:36All contexts combined62:38Telecoms and internet58:42Durables58:42Packaged goods, food56:44Other services51:49Perishable services48:52Business to business46:54The B2B figure rests on fewer than fifty cases, skewed toward the UK and toward relatively big budgets,and the authors state their small sample only allows a rough estimate.
The single 60:40 number sits inside a spread from 80:20 to 48:52. The B2B figure rests on fewer than fifty cases. Source : Binet and Field, Effectiveness in Context, IPA, 2018; The 5 Principles of Growth in B2B Marketing, B2B Institute, 2019 (2018)

The number that should change the conversation

In 2020 the National Bureau of Economic Research published a study built specifically to avoid the selection problem that afflicts everything above. It covers 288 consumer packaged goods brands, uses an explicit and replicable brand selection protocol, estimates effects brand by brand under two separate identification strategies, and, critically, publishes every result regardless of its sign, size or significance.

The findings:

  • Mean long-run advertising elasticity: 0.025. Median: 0.014.
  • More than two thirds of the brand-level estimates are not statistically different from zero.
  • Median weekly return on investment: negative 79 percent, and negative for more than two thirds of brands.

And the authors’ own interpretation of why their numbers are smaller than everyone else’s:

“the magnitudes of the estimated advertising elasticities are considerably smaller compared to the results in the extant literature. This is consistent with both publication bias and over-estimated elasticities in the literature due to confounding factors.”

A peer-reviewed counterweight from 2010 puts the same point differently. Analyzing five years of advertising and scanner data across 25 categories and 70 brands, it estimated total long-run elasticities of 1.37 for product, 0.74 for distribution, 0.13 for advertising and 0.04 for promotion. What you sell and where people can buy it weigh roughly ten times more on long-run sales than what you say about it.

That is not an argument against brand building. It is an argument against expecting brand building to do the work that product and distribution have not done.

Advertising effect estimates from a study that publishes every result regardless of significanceAdvertising effect estimates from a two thousand twenty National Bureau of Economic Research working paper designed specifically to avoid the selection problems that affect the rest of the marketing effectiveness literature. The study covers two hundred and eighty eight consumer packaged goods brands, uses an explicit and replicable brand selection protocol available through an academic data center, estimates effects brand by brand under two separate identification strategies consisting of fixed effects and a discontinuity design at the boundaries of television broadcast areas, and publishes every result regardless of its sign, size or statistical significance. The mean long run advertising elasticity is zero point zero two five and the median is zero point zero one four. More than two thirds of the brand level estimates are not statistically different from zero. The median weekly return on investment is negative seventy nine percent, and the return is negative for more than two thirds of brands. The authors state that the magnitudes of the estimated advertising elasticities are considerably smaller compared with results in the existing literature, and that this is consistent with both publication bias and over estimated elasticities in the literature due to confounding factors. A separate peer reviewed study published in two thousand ten covering five years of advertising and scanner data across twenty five categories and seventy brands estimated total long run elasticities of one point three seven for product, zero point seven four for distribution, zero point one three for advertising and zero point zero four for promotion, meaning product and distribution weigh roughly ten times more on long run sales than advertising does.What happens when every result gets publishedMedian long-run elasticity0.014mean 0.025, across 288 brandsEstimates indistinguishablefrom zeroover 2/3Median weekly returnon investment-79%Long-run elasticity by lever, from a separate peer-reviewed study of 25 categories and 70 brandsProduct1.37Distribution0.74Advertising0.13Promotion0.04
288 brands, both identification strategies, all results published. The median long-run elasticity is 0.014 and most estimates cannot be distinguished from zero. Source : Shapiro, Hitsch and Tuchman, Generalizable and Robust TV Advertising Effects, NBER Working Paper 27684, 2020 (2020)

What none of this covers

Three gaps matter for anyone applying these numbers to their own company.

Nothing here is about a brand starting from zero. The IPA cases, the split-cable experiments, the 288-brand study and the 70-brand elasticity study all cover established brands adjusting their spending. Extrapolating a timeline from them to a company nobody has heard of yet rests on nothing.

Almost nothing here is business to business. The rigorous causal work is exclusively in packaged goods, and the authors of the largest study say why: demand data is not available elsewhere. Packaged goods represent roughly a tenth of United States household consumption spending. Everything applied to B2B is extrapolation, plus fewer than fifty self-selected cases.

None of it assigned campaigns randomly to a duration. That is the whole reason no source can tell you how long brand building takes. To answer the question you would need to run the same campaign for six months, eighteen months and thirty-six months across comparable randomized markets, and nobody has published that.

What each widely quoted brand building timeline figure actually measuresWhat each widely quoted brand building timeline figure actually measures, as opposed to what it is usually taken to mean. The six month figure comes from a two thousand thirteen report for the Institute of Practitioners in Advertising and is a crossover point between two aggregated curves of self reported award competition cases, marking where long term effects begin to dominate short term effects, and it is not a measured latency, so nothing becomes visible at month six that was invisible at month five. The two to three year figure comes from the same databank and describes the horizon over which the gap in very large profit effects between emotional and rational campaigns widens, from thirteen against ten percent at one year to thirty against twenty percent at two years and forty three against twenty three percent at three years and beyond, and it is conditional on the cases that happened to be evaluated over those windows. The three year figure comes from fifty five randomized split cable experiments summarized in nineteen ninety five and is a genuine experimental horizon, with a cumulative effect of about twice the year one effect over the following two years, but only where a significant effect existed in year one. The twenty four month figure comes from a two thousand twenty four industry study and is the ceiling of that study’s measurement window rather than a result, and the study was commissioned by a television promotion body. The common feature of all four is that none of them assigned campaigns randomly to a duration, which is why no source can state how long brand building takes.Four figures that are not timelines6 monthsA crossover point between two aggregated curves of self-reportedaward entries. Not a latency.2 to 3 yearsThe horizon over which the emotional versus rational profit gapwidens, conditional on which cases were evaluated that long.3 yearsA genuine randomized experimental horizon. Cumulative effect abouttwice year one, but only where year one already showed an effect.24 monthsThe ceiling of one industry study’s measurement window, commissionedby a television promotion body. Not a result at all.
Four numbers circulate as timelines. None of them is a measured delay between spending and visible effect. Source : Binet and Field, IPA, 2013 and 2017; Lodish et al., Marketing Science, 1995; Profit Ability 2, Thinkbox, 2024 (2024)

What you can plan on

Strip out what the evidence does not support and four practical statements survive.

Set a first checkpoint inside the first year, and mean it. The randomized evidence says an effect that is going to appear is already detectable in year one. A plan whose first honest read is at month thirty is not a patient plan, it is an unfalsifiable one. A checkpoint is only readable if the work under it was sequenced, and positioning first, then the platform, then the identity, then the guidelines is the order we run it in.

Expect the total to exceed what you see early, if you see something. Roughly double the first-year figure across the following two years, in the experimental data. That is the real argument for continuity, and it applies only after the first checkpoint clears.

Do not import a budget split from another sector. The published spread runs from 80:20 to 48:52, and the authors say their data cannot reliably be cut at category level. Use the direction, which is that activation alone underperforms over time, not the number.

Fix what product and distribution are doing first. Long-run elasticity of 1.37 for product against 0.13 for advertising is the least fashionable finding in this entire literature and the most consequential for a company deciding where next quarter’s money goes.

And when someone tells you brand building takes eighteen months, or three years, ask what that number measured. The honest answer, every time, is that it measured how long somebody’s campaign happened to run before someone wrote it up.