---
name: brand-health-and-equity-analysis
description: >
  Analyses brand funnel, image and salience data: awareness through consideration
  to advocacy, the conversion ratios between stages where the diagnostic value
  sits, attribute association and perceptual mapping, and the size effect that
  makes larger brands score higher on almost everything for reasons of
  familiarity rather than merit. Use for "analyse the brand tracker", "why is
  consideration falling", "build the brand funnel", "which attributes do we own",
  "we score lower than the market leader on everything", "run a brand image
  map", "what is our brand equity", or "does brand health predict sales".
category: 06 Specialist and Advanced Analysis
ref: "06.03"
tier: 2
inherits: [K2, K3, K4, K5]
---

# Brand Health and Equity Analysis

## 1. One-line description
Turns brand funnel, image and salience measures into a diagnosis of where a brand actually loses people and why, by reading conversion between stages rather than levels, correcting for the size effect that inflates every metric a large brand touches, and stating plainly what a brand measure predicts and what it does not.

## 2. What this skill is used for

**The research problem it solves.** Brand tracking generates more numbers per pound spent than any other research format and produces less usable diagnosis, for four connected reasons. Levels get read instead of conversion: a report notes that consideration is 34% and says nothing about the fact that 34% of a familiar base of 70% is a very different brand from 34% of an aware base of 90%. Bases get compared carelessly: each funnel stage has a different base, and a chart that puts awareness, consideration and usage side by side across five brands is comparing figures whose denominators differ in ways nobody has stated. The size effect goes uncorrected: bigger brands score higher on almost every image attribute, including attributes they have no distinctive claim to, because more people know them and familiarity generates agreement. And the metric gets confused with the thing it is supposed to predict, so a brand health score falls and the report says the business will suffer, without ever having established that this measure has predicted anything for this brand in this category. This skill supplies the base discipline, the conversion reading, the size-effect correction, and an honest account of the link between brand measurement and commercial outcome.

**Where it sits in the research lifecycle.** After data preparation on a brand study or tracker wave, and before insight development. Where the study is a tracker, this skill handles the within-wave diagnosis and hands the over-time reading to trend analysis, which owns the comparability and significance rules for wave-on-wave movement.

**Typical use cases.**
- Building a brand funnel with correct bases and reading the conversion between stages.
- Diagnosing which stage a brand loses people at, and for which audience.
- Analysing brand image data and identifying what a brand is actually associated with once size is accounted for.
- Producing a perceptual map and stating what it can and cannot be read as.
- Comparing a brand against competitors without the comparison being an artefact of their relative size.
- Assessing salience and category entry point coverage.
- Auditing a brand health report before its conclusions drive a marketing budget.
- Answering, honestly, whether the brand metrics predict anything commercial for this brand.

**Who uses it.** Brand and insight managers on the client side; research managers and directors running brand trackers; strategists and planners using brand data as an input to positioning; marketing effectiveness analysts trying to connect brand measures to commercial data; and researchers auditing a supplier's brand deck.

## 3. When to use it

- A brand study or tracker wave has been fielded and needs analysis beyond a table of levels.
- Consideration or preference has moved and somebody needs to know where in the funnel the movement came from.
- A brand scores below a competitor on most image attributes and someone is about to interpret that as a positioning problem.
- A perceptual map has been produced and needs interpreting, or needs its interpretation corrected.
- Brand metrics are being connected to sales, and the strength of that connection needs establishing rather than assuming.
- A category entry point or salience framework is being applied to existing brand data.
- A brand health report is about to justify a budget decision and the evidence needs auditing.
- Two brands of very different size are being compared and the comparison needs to be made fair.

## 4. When NOT to use it

- **The question is about wave-on-wave movement rather than within-wave structure.** Whether a three-point drop in consideration is real, what the measure's normal variation is, whether a sample source change explains it, and how to read a series: all of that belongs to **05.05 Trend and Tracker Analysis**. This skill diagnoses the structure of the brand at a point in time and hands the time series across. Where a movement is the whole question, start there and come back here for the diagnosis of where the movement sits in the funnel.
- **The evidence is being asked to establish that brand activity caused a commercial outcome.** Correlation between a brand measure and sales, however strong, is not evidence that moving the measure moves the sales, and the two share so many common causes (distribution, price, seasonality, category growth, the competitor's activity) that the association is close to uninformative on its own. Causal claims about marketing effect require a design that supports them: **06.04 Experiment and A/B Test Analysis** for tests, **05.06 Correlation, Regression and Causal Claim Control** for the language discipline and the confounding structures in observational data.
- **The brand list was not exhaustive of the competitive set as consumers see it.** Prompted awareness and image are measured against the list shown, and a brand absent from that list contributes nothing to the picture even if it is taking the business. Where the list was built from the client's competitive set rather than the consumer's, every share-style metric in the study is a share of an incomplete universe, and this must be stated rather than analysed around.
- **The bases at any stage fall below the reportable threshold.** Funnel arithmetic compounds base problems: a conversion ratio from consideration to preference on an aware base of 800 may rest on a considering base of 90 and a preferring base of 31. Apply **K4 §7** at every stage and report counts rather than percentages where the base does not support them. A conversion ratio between two small numbers is the least stable statistic in the format.
- **The study measures one brand with no competitive context.** A funnel for a single brand tells you its shape but not whether that shape is good, because the norm depends entirely on the category, the brand's size and the purchase cycle. Without competitor data or a category benchmark, report the internal structure and state that no external standard is available. Do not import norms from another category.
- **The image battery was not asked about all brands.** Attribute association is only interpretable comparatively. Where a battery was asked about the client brand and one competitor but not the rest, the resulting picture is not a positioning map, and the size correction in Step 6 cannot be computed.
- **Segment cuts are being mined for a story.** A brand tracker with twelve brands, twenty attributes and a ten-column banner generates tens of thousands of comparisons. Reading off the ones that flag is the failure mode **05.02 Statistical Testing** exists to prevent, and it is unusually common in brand work because the format invites it.
- **The purpose is to demonstrate that a campaign worked.** Where the analysis is being run after a campaign to find a metric that moved, any metric will eventually be found. Pre-specify what should move, on what base, by how much, before looking, or report the whole set with the number of comparisons made.

## 5. Required inputs

**Required. Without these the skill cannot run. If absent, stop and ask.**
- **The exact question wording and structure for every brand metric.** Spontaneous awareness (unprompted, and whether first mention was captured separately), prompted awareness, familiarity, consideration, preference, usage and advocacy each have multiple standard forms that produce different numbers. "Would consider" and "would seriously consider next time you buy" are different measures. Without the wording, comparison across brands, waves or studies is not possible.
- **The brand list shown, in full, and the rotation applied.** Prompted metrics are measured against this list. Order effects on a long unrotated list are large and systematic.
- **The base definition and base size for every metric, for every brand.** Not the study base. The base that each specific figure sits on, which differs by stage and by brand.
- **The routing.** Which stages were asked only of those who passed a previous stage. A funnel built from independently asked questions behaves differently from one built by routing, and mixing the two produces conversion ratios that are not what they appear.
- **The sample definition.** Category users, brand users, general population, and any quota structure. A brand funnel among a brand's own customers is a different object from one among category buyers.

**Optional, and what each one adds.**
- **Competitor data on identical measures.** Effectively required for interpretation, and listed as optional only because studies arrive without it. With it, conversion ratios can be benchmarked and the size effect can be corrected. Without it, almost every reading in this skill becomes qualified.
- **Market share, penetration or volume data.** Enables the size correction to be computed against a real denominator rather than a survey proxy, and enables the honest check of whether the brand measures track anything commercial.
- **Category entry point or occasion data.** Turns a flat salience measure into a diagnosis of which buying situations the brand is and is not retrieved for, which is usually more actionable than an aggregate awareness figure.
- **Previous waves.** Establish the measure's own variation, without which no movement can be called meaningful. Hand to **05.05**.
- **Media or activity data.** Allows the brand series to be read against what the business actually did, with the causal caveats intact.
- **Open-ended brand associations.** Provide the language people use unprompted, which is a stronger salience signal than agreement with a supplied attribute list and often contradicts it.
- **Weighting scheme and effective base.** Required if any testing is to be done on weighted data.

## 6. Questions to ask before starting

1. **What decision does this feed?** A positioning decision, a budget allocation, a campaign evaluation and a portfolio question need different cuts of the same data, and the last two frequently require evidence this study does not contain. Default if unanswered: produce the funnel diagnosis and the image structure, and withhold anything that reads as campaign attribution.
2. **Is this brand large or small relative to the set, and by how much?** Determines the size of the correction needed before any image comparison is meaningful. Default: compute penetration from the data itself and apply the correction in Step 6 regardless.
3. **Was the funnel routed or asked independently?** Determines whether conversion ratios are arithmetic properties of the routing or genuine behavioural quantities. Default: inspect the data for whether any respondent considers a brand they were not aware of, which reveals the structure.
4. **What is the purchase cycle in this category?** A funnel measured monthly in a category people buy every seven years is measuring recall and intention, not a pipeline. Default: state the cycle if known and read consideration as an attitudinal disposition rather than a stage in a journey where the cycle is long.
5. **Which competitors are in the list, and who decided?** Determines whether the universe is the consumer's or the client's. Default: report the list in the output and flag any obvious absence.
6. **Has any brand measure here ever been checked against a commercial outcome for this brand?** Determines whether predictive claims can be made at all. Default: state that the predictive relationship has not been established in this study and confine claims to what the measure describes.
7. **Are the segment cuts pre-specified?** Determines whether subgroup findings are confirmatory or exploratory. Default: treat all subgroup differences as exploratory and label them.

## 7. Step-by-step methodology

**Step 1. Write out the metric definitions exactly as fielded, before computing anything.** For each stage, record the question, the response options, the base it was asked on, whether the brand list was rotated, and whether the measure is first-mention, any-mention or prompted. This matters more here than in most analysis, because brand metric names are used loosely and two studies both reporting "consideration" can differ by twenty points purely on wording. A correct result is a metric definition table that a reader could use to replicate the study, and it is the first page of the output, not an appendix item.

**Step 2. Build the funnel on stated bases and never on the study base.** Compute each stage for each brand on its correct denominator, and state the denominator in the same cell as the figure. The standard sequence, adapting to what was fielded, runs: spontaneous awareness (base: all respondents), prompted awareness (all respondents), familiarity (base: aware), consideration (base: aware, or familiar, depending on routing), preference (base: considering), usage, current and ever (base: aware or all, depending on the question), and advocacy (base: users). The critical discipline is that **each stage has a different base and the bases differ by brand**, because a brand with 90% awareness has a much larger consideration base than one with 40%. A correct result is a funnel table where every cell carries its own base size, and where no two figures are compared without their denominators being visible.

**Step 3. Compute conversion ratios between adjacent stages, because this is where the diagnosis lives.** For each brand, compute the proportion of those at one stage who reach the next: aware to familiar, familiar to considering, considering to preferring, preferring to using, using to advocating. These ratios are the analytically useful quantities, and they are what the level figures obscure. Two brands with identical consideration can have completely different problems: one converts a large aware base poorly, the other converts a small aware base extremely well. The first has a meaning problem (people know it and do not want it), the second has a reach problem (people who know it want it, and not enough people know it). Those diagnoses lead to opposite investments. A correct result is a conversion table per brand with the ratios, their numerators and denominators, and an identification of each brand's weakest conversion relative to the others in the set.

**Step 4. Apply the base trap discipline to every cross-brand comparison.** Three specific rules. First, **never compare a conversion ratio across brands without noting the base each rests on**: a considering-to-preferring ratio of 60% on a base of 45 and one of 48% on a base of 600 are not comparable in any useful sense, and the smaller one is not a finding. Second, **never read a funnel chart across brands as a comparison of stages**, because the visual encourages exactly that: each brand's bar at each stage sits on that brand's own base, so a shorter bar can indicate a smaller aware base rather than a weaker conversion. Third, **do not compute a conversion ratio through a stage that was not routed**, because if consideration was asked of everyone rather than only of the aware, a respondent can consider a brand they say they do not know, and the resulting ratio can exceed 100% or behave oddly. Where this occurs, report it: it is usually a sign that the prompted awareness question and the consideration question mean different things to respondents. A correct result is a comparison table that a sceptical reader cannot misread, plus an explicit note of any structural anomaly found.

**Step 5. Measure and correct for the size effect before reading any image data.** Larger brands score higher on almost every attribute, including attributes with no logical connection to their size, because more people know them, familiarity generates agreement, and people are more willing to attribute qualities to brands they can picture. This is a robust, well-documented regularity, and reading raw image data without correcting for it produces the single most common false conclusion in brand research: that the market leader is better on everything and the challenger has a broad perception deficit. Three corrections, in increasing order of usefulness. *Restrict the base:* compute each attribute among those aware of, or familiar with, the brand, which removes the largest component of the effect. *Index within brand:* express each attribute as that brand's score on the attribute divided by that brand's average score across all attributes, which reveals what the brand is relatively known for regardless of its overall level. *Model the expectation:* regress attribute score on brand penetration or awareness across all brands in the set, and read the residual, which is the amount by which a brand over or under-performs the score its size would predict. The residual is the honest measure of distinctive association. A correct result is an image table showing raw score, within-brand index, and where computable the size-adjusted residual, with the raw column explicitly labelled as size-confounded.

**Step 6. Read the image structure as differentiation against typicality.** Attributes divide into those every brand in the category scores on (category entry requirements, where a low score is a problem but a high score is not an advantage) and those that separate brands (where ownership is possible). Compute, per attribute, the spread across brands: attributes with small spread are hygiene, attributes with large spread are where positioning lives. Then, for the focal brand, read the size-adjusted residuals: positive residuals on high-spread attributes are the brand's genuine associations. This two-way read is what converts a wall of image percentages into a positioning statement. A correct result is an attribute classification (hygiene against differentiating) and a short list of the brand's genuinely distinctive associations, each with its residual and base.

**Step 7. Produce a perceptual map only with its interpretation limits attached.** Correspondence-style mapping places brands and attributes on shared axes derived from the association matrix. It is a useful summarisation device and it is routinely over-read. Four limits must accompany any map. *The axes are derived, not named*: they are statistical dimensions, and labelling them requires human judgement that should be shown as judgement, not presented as a result. *Proximity between a brand and an attribute is not a score*: in the standard formulation, the meaningful reading is directional (a brand's position relative to the origin in the direction of an attribute) rather than simple distance, and distance between two brands, or between two attributes, is not interpretable the same way as brand-to-attribute proximity. *The map shows only the variance the first two dimensions capture*: report that proportion, because a map explaining 48% of the association structure is hiding more than it shows. *A brand near the origin is not "average"*: it is a brand whose profile is close to the category average, which frequently means undifferentiated, but can also mean it was rated by too few people to move. A correct result is a map with the dimensions' explained variance stated, the axis labels flagged as interpretive, and the base per brand available.

**Step 8. Analyse salience and category entry points where the data permits.** Salience in this framing is not awareness but retrieval: whether the brand comes to mind in the buying situations that matter. Where the study captured spontaneous awareness, first mention, or associations to buying occasions, build the coverage view: for each occasion or entry point, which brands are retrieved and at what rate, on the base of people for whom that occasion is relevant. The diagnostic output is a coverage gap list, the situations where the category is bought and the brand is not thought of, which is a more actionable finding than an aggregate awareness number because it points at specific communication content. Where only prompted data exists, say so: prompted agreement that a brand suits an occasion is a much weaker signal than unprompted retrieval, and the two should never be presented as the same measure. A correct result is an occasion-by-brand retrieval table with bases, and an explicit statement of whether the measure was prompted or spontaneous.

**Step 9. Separate what the metric measures from what it predicts, in writing.** Each brand measure describes a state: prompted awareness measures recognition of a name in a list, consideration measures a stated disposition at the moment of asking, advocacy measures a stated likelihood of recommending. What each predicts is an empirical question about this category and this brand, and it has usually not been answered. Where commercial data is available, test the relationship directly: correlate the brand series against the sales or share series across waves, report the strength, and state that common causes (distribution, price, seasonality, category growth) are unaccounted for. Where commercial data is not available, the honest statement is that the predictive relationship has not been established here, and every downstream claim about business consequence is an assumption. A correct result is a short section, in the deliverable, distinguishing measured state from predicted consequence, per **K2 §2.1**.

**Step 10. Build the diagnosis, which is a claim about where and for whom the brand leaks.** Combine Steps 3 to 8 into a single statement: the stage where conversion is weakest relative to the competitive set, the audience in which that weakness concentrates (with bases and with the exploratory label if the cut was not pre-specified), the image residuals consistent with it, and the salience gaps that would explain it. The test of a good diagnosis is that it is falsifiable and specific enough to be wrong: "the brand converts familiarity to consideration at 41% against a set average of 58%, concentrated among category buyers under 35, and its only positive size-adjusted image residuals are on heritage attributes" is a diagnosis. "Brand health is soft" is not. A correct result is a diagnosis of two or three sentences with every component traceable to a table.

**Step 11. State the limits on the commercial reading, at the point of the claim.** Brand measures are collected from a sample of people answering questions about brands, and the connection to what those people do is mediated by availability, price, promotion, habit and the competitor's behaviour, none of which is in the study. Where the report will inform a budget, the limit belongs beside the recommendation, not in the appendix (**K4 §4.3**), and the sign-off in **K5 §2.5** applies. A correct result is a limitation stated in the same section as the implication it constrains.

## 8. Analytical framework

    Metric definition → Stage level on its own base → Conversion between stages
        → Size correction → Distinctive association → Salience coverage
            → Diagnosis → Bounded commercial implication

The framework's centre is the third term. Almost everything useful in brand analysis is a ratio between two stages rather than a level at one, because a level tells you the brand's size and a conversion tells you the brand's efficiency. The fourth term is what makes cross-brand comparison legitimate at all: without size correction, every image comparison is partly a comparison of how many people know each brand, and the analysis reports familiarity dressed as perception.

The last term is where discipline is most often lost. Everything to its left is measurement and arithmetic on measurement. The step to commercial implication crosses into territory the study does not cover, and the framework is built so that the crossing is visible: the diagnosis is a claim about the survey's picture of the brand, and the implication is a claim about the business, and they are different kinds of claim.

## 9. Output format

**The metric definition table.** Metric, exact question wording, response options, base definition, rotation, brands asked about. First page.

**The funnel table.** One row per stage, one column group per brand.

| Stage | Base definition | Brand A % (n) | Brand B % (n) | Brand C % (n) |
|---|---|---|---|---|

Every cell carries its base. No cell is compared across brands without the bases being visible in the same view.

**The conversion table.** The analytical core.

| Conversion | Brand A (num/den, %) | Brand B | Brand C | Set average | Focal brand gap |
|---|---|---|---|---|---|

**The image table.**

| Attribute | Raw % (size-confounded) | Base | Within-brand index | Size-adjusted residual | Category spread | Classification |
|---|---|---|---|---|---|---|

The raw column carries its label permanently. Classification is hygiene or differentiating.

**The perceptual map.** With: variance explained by the two dimensions shown, a statement that axis labels are interpretive, base per brand, and a note that brand-to-attribute proximity is read directionally.

**The salience coverage table.** Occasion or entry point, base of people for whom it is relevant, retrieval rate per brand, measure type (spontaneous or prompted).

**The diagnosis.** Two to four sentences, each traceable to a table reference.

**Where the evidence is thin**, use these forms:
- `[not reportable: base below 30 at this stage, counts shown]`
- `[conversion not computable: stage asked independently, not routed]`
- `[size correction not computable: image battery not asked for all brands]`
- `[no external benchmark: single-brand study, internal structure only]`
- `[predictive relationship not established in this study]`
- `[exploratory: subgroup difference not pre-specified, one of N comparisons]`

## 10. Quality checks

**K4 §8** runs anyway. These are specific to brand analysis.

1. Does every funnel figure carry its own base definition and base size, in the same view?
2. Is any figure compared across brands whose bases differ without that being stated?
3. Are conversion ratios computed only where the routing supports them?
4. Has any conversion ratio been reported on a denominator below the **K4 §7** threshold?
5. Is the raw image data labelled as size-confounded wherever it appears?
6. Has a size correction been applied or its impossibility stated, before any cross-brand image claim?
7. Are attributes classified as hygiene or differentiating, rather than all treated as positioning opportunities?
8. Does the perceptual map state its explained variance and flag its axis labels as interpretive?
9. Is any brand-to-brand distance on the map being read as a similarity score?
10. Is prompted data anywhere presented as though it were spontaneous retrieval?
11. Does the output distinguish what each metric measures from what it is assumed to predict?
12. Where a commercial link is claimed, is the evidence for it shown, and are the common causes named?
13. Is the brand list reproduced, with any material absence flagged?
14. Are subgroup differences labelled exploratory where they were not pre-specified, with the comparison count stated?
15. Does the diagnosis name a stage, an audience and a base, rather than describing the brand in adjectives?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **Levels read instead of conversion** | A report describing consideration as high or low with no reference to the aware base beneath it | Conversion table is the analytical core; levels appear only with their denominators |
| **The size effect read as superiority** | The market leader ahead on every attribute including ones it has no claim to | Size correction mandatory before any image comparison; raw column permanently labelled |
| **Cross-brand funnel chart misread** | A stage-by-stage bar chart used to say brand B is weaker at consideration | Bases shown in the chart; conversion presented alongside levels |
| **Conversion on a tiny denominator** | A striking conversion ratio built on 30 considerers | **K4 §7** applied at every stage, not just to the study base |
| **Map over-interpretation** | Named axes presented as findings, brand-to-brand distances read as similarity | Explained variance stated, labels flagged interpretive, directional reading explained |
| **Hygiene attribute treated as opportunity** | A recommendation to build an association every brand in the category already has | Spread computed per attribute; hygiene attributes classified before any recommendation |
| **Prompted read as salient** | "Consumers associate us with weekend occasions" from a prompted agreement battery | Measure type on every salience figure; spontaneous and prompted never merged |
| **The incomplete brand list** | A share-style metric in a study missing a real competitor | List reproduced in the output; absence flagged as a limitation on every share metric |
| **Metric-outcome confusion** | A falling brand score reported as a business consequence | Step 9 section required; predictive relationship evidenced or declared unestablished |
| **Campaign attribution by search** | A post-campaign report naming the one metric that moved | Pre-specify what should move; otherwise report the full set and the comparison count |
| **AI: importing category norms** | A benchmark figure with no source, for a category the study did not cover | Norms cited or absent; **K4 §2.4** applies to benchmark figures as to any other source |
| **AI: naming perceptual map axes confidently** | Axis labels presented as output rather than interpretation | Labels always marked as analyst judgement, with the loadings shown |
| **AI: building a funnel from mixed routing** | Conversion ratios above 100%, or oddly stable across brands | Routing inspected in the data before any ratio is computed |
| **Caveat migration** | The size-confounding note present in the working file, absent on the image chart | Caveats travel with their claim (**K2 §7**) |

## 12. AI guardrails

Universal prohibitions are inherited from **K4**. **K4 §3.1** (no significance without a test) and **K4 §3.2** (no causation from correlation) govern this skill in full. The following are specific.

1. **Never report a brand metric without its base definition and base size**, and never use the study base where the metric sits on a routed base.
2. **Never compare image scores across brands of different size without applying or stating the impossibility of a size correction**, and never present raw image comparison without the size-confounded label.
3. **Never compute a conversion ratio through a stage that was not routed from the previous one**, and where the data structure is unclear, inspect it rather than assuming.
4. **Never present a perceptual map axis label as a result.** It is an interpretation of statistical dimensions and must be marked as one, with the explained variance reported.
5. **Never state or imply that a brand metric predicts a commercial outcome** unless that relationship has been tested in the data at hand, and where it has, never without naming the confounders that were not controlled.
6. **Never import a category norm, benchmark or "healthy" threshold that cannot be sourced.** An unsourced benchmark is a fabricated source under **K4 §2.4**, and brand work is full of them.
7. **Never merge spontaneous and prompted measures**, or describe prompted agreement as salience, mental availability or top-of-mind.
8. **Never describe a movement between waves as meaningful** without the trend discipline in **05.05** and a test, and never at all where the sample source or question changed.
9. **Never report an attribute association as a brand strength without checking its category spread.** An attribute on which every brand scores 80% is not a strength.
10. **Never characterise a brand in evaluative adjectives that no measure supports.** "Warm", "trusted", "premium" are claims about specific attribute scores on specific bases, or they are the analyst's prose.

## 13. Best-practice principles

1. **Ratios diagnose, levels describe.** A level tells you how big a brand is at a stage. A conversion tells you what the brand does with the people it has. Almost every actionable brand finding is a conversion finding, and almost every brand report leads with levels.
2. **Assume the size effect until you have removed it.** Before concluding anything about perception, ask whether the pattern would be produced by the brands' relative familiarity alone. Frequently it would, and the analysis that survives that question is worth far more than the one that does not ask it.
3. **A weak stage tells you what kind of problem you have, and they need opposite investments.** A reach problem (small aware base, strong conversion) is a media and distribution answer. A meaning problem (large aware base, weak conversion) is a proposition answer. Spending on the wrong one is the most expensive mistake brand research enables.
4. **Category norms are usually the most influential and least evidenced numbers in the room.** Where a benchmark is quoted, know its source, its category, its question wording and its date, or do not quote it.
5. **The purchase cycle governs how a funnel should be read.** In a category bought weekly, consideration is close to behaviour. In one bought every seven years, it is a disposition with almost no near-term behavioural content, and reading it as a pipeline is a category error.
6. **Brand image batteries measure what a list allowed people to say.** The attributes were chosen by someone, usually years ago, and the brand's real association may not be on the list. Open-ended associations are a cheap and effective corrective and frequently disagree with the battery.
7. **Distinctiveness beats favourability.** Being liked on attributes that every brand is liked on has no competitive value. The analysis should be built to find where a brand differs, not where it scores well.
8. **Advocacy measures are stated likelihoods, not behaviour.** They are subject to the same stated-preference gap as pricing intent, they are strongly affected by scale format and by cultural response style, and cross-market comparison of them is unreliable for that reason alone. Report them as what they are.
9. **Aggregate awareness hides the useful question.** Which situations, needs or occasions the brand is retrieved for is more actionable than how many people recognise the name, because it points to content rather than to spend.
10. **A brand tracker's greatest analytical risk is the sheer number of comparisons it permits.** Decide what you are looking at before you look, and report how many places you looked.
11. **Where the brand and the commercial data disagree, that is the finding.** A brand improving while sales fall, or the reverse, is more informative than either series alone, and it usually points at something outside the brand (distribution, price, a channel change) that the business needs to know about.
12. **Write the diagnosis so it could be wrong.** A brand conclusion that no evidence could contradict is not a conclusion. Name the stage, the audience, the base and the comparison.

## 14. Worked example

**INPUT**

A fictional regional health insurer, Northgate Mutual, runs an annual brand study among 1,500 adults with private health cover or actively considering it. Five brands are measured on a full funnel and a fifteen-attribute image battery. Northgate has 38% prompted awareness against the market leader's 91%. The board's brief says: "we underperform the leader on fourteen of fifteen image attributes, so we have a broad perception problem and need a major repositioning campaign".

**PROCESS**

*Steps 1 and 2, definitions and funnel.* The metric sheet reveals consideration was asked as "which of these would you consider when next choosing cover", routed to the prompted-aware only. Bases per brand differ sharply: Northgate's aware base is 570, the leader's is 1,365. Funnel levels: Northgate prompted awareness 38%, consideration 14% (of all), usage 6%. Leader: 91%, 41%, 29%.

*Step 3, conversion.* This is where the picture changes. Northgate converts aware to considering at 37% (211/570). The leader converts at 45% (615/1,365). The set average is 39%. Northgate is slightly below average on this conversion but not remarkably so. Considering to using, Northgate converts at 43% (90/211) against the leader's 71% and a set average of 52%. That is the weak point, and it is a different problem from the one the board described.

*Steps 4 and 5, base trap and size correction.* The board's "fourteen of fifteen attributes" comparison is computed on all respondents, so the leader benefits from being known to 91% against Northgate's 38%. Recomputing on the aware base closes roughly two thirds of the gap on most attributes. Regressing each attribute on prompted awareness across the five brands and reading residuals changes the picture further: Northgate's residuals are positive on three attributes ("handles claims without argument", "understands people like me", "local"), close to zero on nine, and negative on three ("range of cover options", "modern", "widely accepted by hospitals").

*Step 6, differentiation.* Category spread per attribute shows that seven of the fifteen have a spread under six points across brands, meaning every brand scores similarly: these are hygiene. "Handles claims without argument" has a 21-point spread and Northgate has the highest size-adjusted residual on it in the set. "Widely accepted by hospitals" has a 26-point spread and Northgate's residual is the most negative.

*The judgement call.* The three negative residual attributes could be read as a perception problem to be fixed by communication. But "widely accepted by hospitals" is a factual claim about network coverage, and the analysis has no data on whether Northgate's network is genuinely narrower. Reading a factual deficit as a perception deficit would recommend advertising a claim that may be false. Resolved by reporting the residual, flagging that the study cannot distinguish perception from reality on this attribute, and naming the internal data that would settle it. This is a **K5 §2.1** materiality and **K4 §3.4** hypothesis boundary.

*Step 9, prediction.* Three prior waves exist. Correlating the consideration series against the insurer's new-policy volumes across four points gives a relationship that cannot be estimated meaningfully on four observations, and this is stated rather than reported as a correlation.

**OUTPUT**

> Northgate's weakest funnel conversion is considering to using: 43% (90 of 211) against a set average of 52% and the market leader's 71%. Its aware-to-considering conversion, at 37% (211 of 570), is close to the set average of 39%. The brand's problem is concentrated at the final stage, not across the funnel.

> The board's observation that Northgate trails the leader on fourteen of fifteen image attributes is computed on all respondents, where the leader is known to 91% and Northgate to 38%. Recomputed among those aware of each brand, and adjusted for the association between awareness and attribute agreement across the five brands, Northgate over-performs the score its size would predict on three attributes: handles claims without argument (residual +9 points, base 570), understands people like me (+6), and local (+11). It under-performs on range of cover options (-7), modern (-5) and widely accepted by hospitals (-14).

> Seven of the fifteen attributes show a category spread under six points and are hygiene rather than positioning opportunities. The attribute with the largest spread on which Northgate leads on a size-adjusted basis is handles claims without argument.

> **Hypothesis, requires internal data.** The negative residual on widely accepted by hospitals may reflect a genuine network limitation rather than a perception gap. This study cannot distinguish the two. Northgate's provider network data against the leader's would settle it, and the answer changes the recommendation completely: a communication response is appropriate only if the network is comparable.

> **What could not be established.** Whether any of these measures predicts new policy volume. Four waves is too few to estimate a relationship, and the series share obvious common causes including price positioning and broker activity.

**Researcher sign-off required.** This output will be used to size a campaign budget. The evidence supports a specific, narrow diagnosis (final-stage conversion, with one owned attribute) and does not support the broad repositioning the brief assumed (**K5 §2.5**).

## 15. Advanced usage

**Modelling the funnel rather than tabulating it.** Where sample allows, model each conversion as a function of respondent characteristics and image associations, which identifies which associations are correlated with progressing to the next stage. This is genuinely useful and must carry the full **05.06** caveat set, because "associated with converting" is not "causes conversion", and the image measures and the conversion are collected in the same interview from the same person, which creates a common-method association independent of any real relationship.

**Category entry point mapping at scale.** Where occasion data has been collected for a wide set of buying situations, the coverage matrix (brands by entry points) is the most actionable brand output available, because it converts an abstract salience problem into a list of specific situations the brand is missing from. Combine with **07.02** or **06.06** on open-ended association data for the language people actually use.

**Combining survey brand data with behavioural or search data.** External signals of retrieval (unprompted search volume, direct traffic) provide a second, non-survey read on salience with entirely different biases. Convergence between them raises confidence substantially. Divergence is informative and must be reported rather than reconciled (**K4 §4.1**).

**Distinguishing brand size from brand strength formally.** Across a set of brands, plotting any loyalty or attitude metric against penetration typically shows a strong positive relationship: bigger brands have more loyal customers, a regularity known in the literature and frequently rediscovered as a finding about a specific brand. Fit the relationship across the set and read every brand as a deviation from it. A brand exactly on the line is normal for its size, which is neither good news nor bad, and reporting it as either is a mistake.

**Multi-market brand comparison.** Attribute agreement, scale use and advocacy measures vary by cultural response style as much as by brand perception, so raw cross-market comparison of levels is unsafe. Compare within-market standardised scores or within-market ranks, and treat the interpretation as a **K5 §2.2** point requiring local knowledge.

**Auditing an inherited tracker.** Where a tracker has run for years, the highest-value work is often the definitional audit: which metric definitions have changed, when the sample source moved, whether the brand list has been amended, and which of the reported series are therefore not continuous. Hand the series reconstruction to **05.05**.

## 16. Skill chain

**Recommended previous skills**
- **04.05 Weighting and Base Management.** Hands over the weighting scheme and effective bases, which every funnel figure and every test depends on.
- **04.02 Data Cleaning.** Hands over the cleaned dataset with routing intact, which this skill inspects rather than assumes.
- **05.01 Descriptive Analysis** and **05.03 Cross-Tabulation.** Hand over the distributions and the banner structure that the funnel and image tables are built from.
- **02.01 Survey Questionnaire Design.** Hands over the exact metric wording and the brand list, without which nothing in this skill is interpretable.

**Recommended next skills**
- **05.05 Trend and Tracker Analysis.** Takes the over-time reading: whether a movement is real, what the measure's own variation is, and how to handle wave comparability. This skill diagnoses within a wave and hands the series across.
- **08.01 Finding to Insight Development.** Takes the diagnosis and asks what it means for the proposition, which is where a conversion gap becomes an argument.
- **09.01 Audience Segmentation.** Takes the funnel and image data by audience where the diagnosis concentrates in a specific group.

**Runs well alongside**
- **05.02 Statistical Testing**, which supplies the comparison discipline a brand tracker's enormous comparison count demands.
- **05.06 Correlation, Regression and Causal Claim Control**, which governs any attempt to connect a brand measure to a commercial outcome.
- **13.03 AI Output Verification**, which audits brand decks for unsourced norms, uncorrected size effects and confidently named map axes.

---
A Yazi Supplied Skill and resource.