# Twenty-nine quarters, three phases: what an ad budget looks like when you rebuild it quarter by quarter

원문: https://citeangle.com/en/research/ad-spend-phases-push-and-harvest

Contents

1. [What a phase looks like](#sec-phases)

2. [Whether spending harder works](#sec-peers)

3. [What the lag test said](#sec-lag)

4. [How the quarters were rebuilt](#sec-method)

5. [What this cannot tell you](#sec-limits)

6. [Reading your own phase](#sec-use)

Questions this page answers: [What does a phase look like in the filings?](#sec-phases) [Does spending harder produce the harvest phase?](#sec-peers) [What did the lag test say about cause?](#sec-lag) [What can this data not tell you?](#sec-limits)

Published August 8, 2026 · Written by [Jay Sim](https://citeangle.com/en/authors/jay-sim) ·
CiteAngle · Built from one listed company's public quarterly disclosures, read on 2026-08-08. The
filer is a listed cosmetics company in Asia and is not named here, so every figure is given as a
share, a multiple or a count, which carries the finding without identifying it.

**Is the ad-to-revenue ratio a lever a company pulls? —**
In these filings it behaves as a readout. Rebuilt quarter by quarter, the line shows three runs. In
the first, advertising outgrew revenue and the share climbed to 35.08% of the revenue booked in that
quarter. Then came 13 consecutive
quarters where revenue outgrew advertising and the share fell to 17.01%, and a second push that is
now two quarters old. Advertising money rose 1.80 times during the falling stretch, so the share
came down while the budget went up. Measured on the same basis, eight listed peers show the run was
not a category season, and the lag test found no delayed effect to sell as cause.

## What does a phase look like in the filings?

The test is one subtraction. Take a quarter's year-over-year revenue growth, subtract the same
quarter's year-over-year advertising growth, and read the sign. Positive means revenue moved faster,
so the share fell. Negative means advertising moved faster, so the share rose. Across 29 quarters
those signs group into runs instead of scattering.

The share peaked, then fell for 13 quarters. It has risen for two.

peak 30.26%
 trough 17.01%
 18.12%
 push
 harvest, 13 quarters
 push again

Advertising and promotion as a share of the revenue booked in the
 same trailing four quarters, rebuilt from one listed company's public quarterly disclosures and
 kept anonymous here.
 Thirteen points are plotted and joined. The quarters between plotted points were computed but are
 not drawn. The 35.08% figure quoted elsewhere on this page is a single unsmoothed quarter, so it
 sits above this line's peak. One company's record, not a forecast.

| Phase | Length | Sign of revenue growth minus ad growth | Ad share of revenue |

|---|---|---|---|

| Push | 10 quarters | negative in 6 of 10, alternating | rose to a peak of 35.08% in one quarter |

| Harvest | 13 quarters | positive in all 13 | 29.07% down to 17.01% |

| Push again | 2 quarters so far | negative in both | 17.01% up to 18.12% |

The middle run is the part worth staring at. Thirteen quarters in a row without a single miss.
If the sign were a coin flip each quarter, a run of 13 quarters turns up 0.0122% of the time.

The obvious reading of a falling share is that the company pulled back. It did the opposite.
Advertising in absolute money rose 1.80 times over those same 13 quarters. Revenue rose faster
still, which is the whole reason the share came down. That arithmetic is worked through on its own
in [our note on what an ad ratio can and
cannot tell you](https://citeangle.com/en/research/ad-spend-ratio-and-awareness-baseline).

That is also why the current reading matters more than the trough. The share bottomed at 17.01%
and has risen in each of the two quarters since — the last three readings in the 29 quarters
rebuilt from the filings. Whatever that company was doing during the harvest, it stopped doing it
two quarters ago.

## Does spending harder produce the harvest phase?

This is where the reading a marketer wants runs into the control group, so it is worth being
blunt. The companies in this set that spent more came out worse on this measure.

The same measurement was run on eight listed peers in the same category across the same span. Not
one of them saw its advertising share fall. Two of them raised advertising far more aggressively
than the subject company did, by roughly 3.4 times against its 2.31, and their shares went up.
Spending harder did not produce a harvest. In this control group it produced a higher share.

What separated the subject was not the size of its advertising increase. It was the revenue
multiple underneath, which came to 3.84 times across the same span.

A negative control sits on the other side of that. Contract manufacturers sell to other brands,
so their advertising line is structurally small. Two of them went through the same measurement.
Their shares sat below 0.5% for one and near 2% for the other, and stayed flat across the same 29
quarters. An accounting change or a shift moving through the whole industry would have left a mark
there. None appears.

Then the timing. In the exact five quarters where the subject's share fell 7.79 points, one
peer's share climbed 2.62 points and another stayed flat. Same market, same quarters, same
accounting basis, and the directions still split. A phase belongs to a company. The calendar was
identical for all three of them.

## What did the lag test say about cause?

If advertising builds something that pays back later, the correlation between advertising growth
and revenue growth should peak one or more quarters out. We tested that directly.

It peaks at zero. Same-quarter correlation is +0.819. One quarter out it is +0.552, and it keeps
falling from there. Run the test in the other direction, with revenue growth first and advertising
growth after, and the correlation is larger at every lag we checked, including +0.668 at one
quarter.

A company that sets its advertising budget as a percentage of revenue would generate exactly that
pattern without any advertising effect at all. So the honest summary is narrow: the phases are in
the data, and what moves between them is not. We can show you which phase a brand is in. We cannot
tell you from this what carries it from one to the next, and nobody who has only read these filings
can either.

That is a smaller claim than the one usually sold alongside a chart like this. It is also the one
the data supports.

## How the quarters were rebuilt, and what the self-check caught

Every figure is on a consolidated basis and every quarter is a three-month standalone number
rather than a cumulative one. Fourth quarters are never filed on their own, so each one is the
annual figure minus the nine-month cumulative. Seasonality is removed by summing the trailing four
quarters before taking any share.

The self-check is one question asked 14 times. For each of seven years, and for each of the two
line items, do the four quarters add up to the filed annual figure? All 14 closed with a residual
under one million won.

The check caught an error on its first run. One filing's prior-three-months column carried a
revenue figure that left the year 5.8 billion won short. Nine-month cumulative minus half-year gave
a different number, and only that one closes the annual total. The advertising column in the same
filing was intact. A formatting error can sit in one line item and not its neighbour, which means
each line has to be checked on its own. Judging the report as one document would have let this
through.

We publish the test and the denominators because a phase claim is worthless if you cannot see
what was divided by what. Running the same read on a live brand across search and AI answers is a
different job, and that part lives in our [measurement protocol](https://citeangle.com/en/methodology).

## What can this data not tell you?

Everything above comes from a single company in one national market. Nothing here says the same
shape appears elsewhere, and the peer table above is a reason to expect that it often does not.

The lag test runs on 25 usable observations. Standard error on those correlations is around 0.21
to 0.23, so two of them sitting closer than about 0.2 apart cannot be separated with a sample of this size.
Quarterly filings are the finest resolution that exists for this line. There is no monthly
disclosure to go get.

No media-level or channel-level figure appears anywhere in these filings. The disclosure format
has no column for it, so no amount of digging produces a split between search, social, and offline
spending. No peer in the set discloses more.

The second push also overlaps a run of new offline retail entries, and those structurally create
channel commission. The rise mixes a spending decision with a channel change, and the filings cannot
separate the two. We looked and could not, which is different from saying it cannot be done with
better data.

## Reading your own phase

Your own share moving up or down cannot tell you which phase you are in. A rising share could be
your push or the whole category repricing, and one line on its own does not distinguish those. The
peer table above is the entire reason we say so: eight companies moved one way while one moved the
other, in the same quarters.

The reading has to be taken next to the competitors your buyers actually see, which is what we
build. You get where your brand is surfaced across every answer surface in scope, and which rivals
are named beside you question by question. The distance between two dated readings is the change.
Where those readings are taken across the US market is set out in our [map of the answer surfaces](https://citeangle.com/en/research/us-ai-search-surface-map).

One caution carries over from our work on [single-run
measurement](https://citeangle.com/en/research/single-run-point-estimate): published work puts run-to-run variance at 10% to 34% from sampling alone, so a
single pass gives a point estimate and a phase judgment made on one reading inherits that noise.
That is why we read the same question repeatedly and report [how often](https://citeangle.com/en/methodology#why-7) a name comes back, not whether it came back once. We
count exposure, clicks, and sales as separate lines and report them separately, because rolling
them together hides which one is stuck.

### Does a rising ad-to-revenue ratio mean a company is losing efficiency?

Not on its own. A share rises when advertising outpaces the sales figure, which is what the
opening of a push phase looks like, and it also rises when revenue stalls. Those are two different
situations that produce the same number. Read the two lines separately before reading the ratio
between them.

### Can you tell which phase you are in from your own numbers alone?

Only partly. Your own numbers give you the sign of the subtraction, which tells you whether your
share is rising or falling. They cannot tell you whether your rivals are doing the same thing, and
that is the difference between a company event and a category one. In this dataset eight of nine
companies moved together and one did not.

Which phase is your brand in, and are the competitors your buyers see moving with
you or against you? Start with a [free AI visibility check](https://citeangle.com/en/services), then take a
dated baseline before the next budget decision.

**What a single quarter costs you.** Without the series around it, an investment quarter
and a failing quarter look the same, and the risk is stopping the spend that was about to pay.
Every reading in our series is sealed under conditions fixed in advance, which is the reason the
line means a change in the world rather than a change in the instrument.

## Why a series, and why ours

The lesson underneath these 29 quarters is uncomfortable for the way most reporting works. A
quarter judged on its own can look like failure and be an investment, or look like efficiency and
be a harvest running out. You cannot tell which from inside the quarter. Only the series tells
you, and a series is readable only if every point in it was measured the same way.

That last condition is where measurement quietly fails, and it is the one we build for.
Questions, engines, run count, judging rule and reporting window are fixed before the first
reading and stay fixed, and each reading is sealed with the conditions it ran under — so the
line you steer by is a change in the world rather than a change in the instrument. Readings that
could not be taken are counted rather than dropped, which matters more in a series than anywhere
else: a denominator shrinking a little each month draws a rising line out of nothing at all.

A ratio is what happened. Knowing the phase tells you where you stand now.

A measured baseline shows where your brand is surfaced across all 17 US surfaces and which rivals are named beside you, question by question. Every reading keeps its status and its receipt, so the next reading has something to be compared against.

[Get your free AI visibility check](https://citeangle.com/en/#snapform)[See the measurement protocol](https://citeangle.com/en/methodology)