A Measurable Definition of Brand Health

01Executive summary

Every company with an audience is accumulating or losing brand equity every week, and almost none of them can see it happening. Revenue confirms the damage or the compounding months later, after the budget is spent and the strategy is set. The industry’s standard answer is a mix of vanity metrics, survey panels, and annual rankings published without accuracy statistics.

IMPRS is our answer: a brand health score from 350 to 900, computed deterministically from 117 metrics tracked across five pillars of audience behavior on Instagram, Facebook, TikTok, YouTube, and X. No survey. No judgment call. No generative AI in the calculation. Two people looking at the same account get the same number.

Before launch, the score was tested retrospectively against six years of real revenue outcomes across 122 companies, always forward in time: the score computed from one year’s behavior, graded against the revenue that arrived the next year. In that backtest, stronger scores were associated with stronger subsequent revenue growth, movement in the score carried more information than its level, and the score out-ranked a benchmark index rebuilt from the industry standard’s published components on the identical panel.

We state the status plainly, because the credibility of a measure depends on it: these results establish that the methodology is promising, not proven. Prospective validation on live accounts, where every forecast is logged before its outcome is known, began in July 2026 and will be published without retrospective editing.

02The problem: brand health is managed blind

Brand is the most valuable asset most modern companies own and the least instrumented. Finance has audited statements. Operations has dashboards to the minute. Brand has follower counts, engagement rates that mean different things on every platform, sentiment tools that grade the mood of comment sections, and an annual industry ranking with no published error rate.

The cost of that blindness has a shape. An audience does not leave loudly. It stops saving posts, stops sharing them into DMs, stops clicking through, and keeps scrolling politely for months while the revenue line still looks fine. By the quarter revenue drops, the underlying asset has been eroding for two or three quarters. The same is true in reverse: a brand quietly compounding attention equity looks identical to a stagnant one if all you watch is follower count.

Three habits of the existing toolkit make it unable to catch this:

• It measures opinion instead of behavior. Surveys and comment sentiment record what people say. Purchases, saves, shares, and return visits record what they do, and the two routinely diverge.
• It measures size instead of health. Spend, followers, and reach tell you who is big. They are close to useless for telling you who is about to grow, and a size-dominated index can even point backwards, because smaller companies grow faster.
• It never grades itself. Rankings are published annually with no accuracy statistics, no confidence intervals, and no public record of whether last year’s ranking predicted anything. A number without a standard behind it is an opinion with a font.

Credit markets solved an analogous problem decades ago. Lenders could not observe creditworthiness directly, so bureaus built a single score from observable repayment behavior and validated it against subsequent defaults. The score is not the asset. It is a calibrated, auditable estimate of a latent one, and entire markets run on it. Brand deserves the same treatment.

03A proposal, not a proclamation

We want to be precise about what we are claiming, because the industry we are entering is full of confident numbers that were never tested.

We’re proposing a measurable definition of brand health. Not announcing that we have discovered what brand health “really is,” and not asserting that our definition is the only defensible one. A proposal of this kind earns acceptance the way scientific proposals do: by being stated precisely, tested against outcomes it did not get to choose, published with its uncertainty attached, and exposed to future data that could prove it wrong.

That framing has practical consequences you can hold us to. Every figure we publish carries its sample size and a 95% confidence interval. Retrospective and prospective evidence are labeled and never blurred. The formula is sealed and versioned, so evidence stays attached to the exact mathematics that produced it. And the full methodology, including the parts that did not work and the claims we retracted before publication, lives in the open on our research pages.

04The seven axioms

The definition rests on seven axioms. Everything in the system, from which signals are collected to how the score is validated, follows from them. If you reject one of these, you should reject the score; if you accept them, the rest is engineering and statistics.

The first axiom sets the data source: revealed preference over stated preference, what audiences do over what they tell a survey. The second and third say that brand health is real but not directly observable, like creditworthiness, and that it leaves behavioral fingerprints. The fourth and fifth justify building an index at all. The sixth is the discipline: a score that predicts nothing external is decoration, so IMPRS is graded against subsequent revenue outcomes. The seventh is the audit condition: identical inputs must produce the identical score, which rules out human judgment and generative models anywhere in the calculation.

05What IMPRS measures

Brand health is the strength of the relationship between a brand and its audience, measured through observable behavior. IMPRS is a deterministic score of that relationship: it estimates how much durable attention equity a brand holds, by measuring how audiences actually behave toward it, and it is calibrated against whether that equity showed up later as revenue growth.

Concretely, the system tracks 117 metrics across the five connected platforms and, when a revenue source is connected, revealed-preference purchase behavior. The metrics divide into two families: signals of how audiences respond to the brand, the depth-of-impression measures like save rates, completion rates, and share behavior, and signals of how the brand is growing its audience, the trajectory measures like acceleration, cadence consistency, and discovery rates. Both families roll up through the five pillars into one number.

Just as important is what IMPRS is not:

• It does not predict revenue directly and does not guarantee business outcomes. It is a measure; forecasting is an application built on top of it, and that application abstains when the signal is weak.
• It does not use generative AI to calculate the score. AI explains the score and recommends actions; it never generates the number.
• It is never adjusted manually, and unsuccessful forecasts are never hidden. The prospective record publishes hits and misses alike.

06The five pillars

The score decomposes into five pillars: Impact, Momentum, Presence, Reach, and Sentiment. Each is a family of behavioral signals, each is graded on its own evidence, and their relative weights differ by segment because the backtest showed different segments run on different engines. An agency’s health is not a creator’s health.

Impact

The depth of the impression. Saves, bookmarks, DM shares, carousel completion: signals that something was worth keeping, not just scrolling past. In the validation panel, Impact was the most stable single signal of long-term brand equity. A brand with consistently high Impact but low Reach is a diamond waiting to be found; the reverse is a liability. Agencies carry the heaviest Impact weighting of any segment.

Momentum

Whether the brand is compounding or plateauing: score trajectory, acceleration against the cohort, cadence consistency, format diversification. Momentum was the strongest single leading pillar in the panel, nearly matching the full composite on its own. For creators the system uses Momentum’s year-over-year change rather than its absolute level, because absolute Momentum inverts at high follower counts as accounts mature.

Presence

Authority and consistency across every platform the brand posts on: profile optimization, visual identity consistency, response behavior, cross-platform parity. Presence has the weakest direct association with future revenue of the five pillars, and we keep it in the formula anyway, as a prerequisite floor: brands with weak Presence almost never achieve strong Impact regardless of content quality.

Reach

How fast the audience grows organically: non-follower impressions, algorithmic surfacing, search and hashtag entry points. Discovery signals, not distribution signals. Reach matters most for small businesses, where local discovery and word-of-mouth amplification are primary revenue drivers, and it is weighted accordingly in that segment.

Sentiment

Revealed preference: repeat purchase, refunds, and churn from a connected revenue source such as Stripe, Shopify, Xero, or QuickBooks. A commitment score, not a mood score. The question is whether customers act, not whether they feel good. Earlier versions of the formula used comment-based sentiment proxies; they measured near zero against outcomes and were removed. When no revenue source is connected, a comment-sentiment preview is shown for context but carries no weight in the score. Comments never move the number.

07The 350-900 scale

IMPRS is reported on a credit-style scale from 350 to 900. The familiar shape is deliberate: credit bureaus turned thousands of loan signals into one number banks could trust, and the range signals the same contract, a calibrated estimate of a latent asset, graded against outcomes.

Two properties of the scale matter to a decision-maker. First, it is size-independent: the score cannot be bought with follower count or ad spend, so a challenger brand and a Fortune 500 brand are graded on the same behavioral basis. Second, movement on the scale is itself a signal, and in the backtest it was the stronger one, which is why the product is built around deltas: the morning briefing, the agency delta report, and the forecasts all track how the score is moving.

08How it was validated, in plain language

The validation panel is 122 companies with public revenue histories across nine industries and six years, 2019 through 2024, spanning Fortune 500 brands, consumer and B2B brands, small businesses, agencies, and creators. The formula was developed and iterated on this panel with hold-out testing at each step, then frozen as V32 with a re-seal gate: any future change to the mathematics creates a new version that must earn its own evidence.

Every test is forward in time. The score computed from year N behavior is graded against revenue realized in year N+1. Same-year fit, the trick behind most impressive-sounding analytics, is never presented as forward evidence. Four findings from the backtest carry the story:

• Ordering held. When the score clearly separated two accounts of the same type, the higher score identified the faster-growing one up to 86.6% of the time for B2B brands and 81.7% for Fortune 500 brands. The top-scoring half of a tier outgrew the bottom half in 80% of panel years for four of six tiers.
• Movement carried the most information. Changes in the score were associated with next-year revenue changes at rho +0.395 pooled, against +0.160 for the score’s level, and +0.563 for consumer brands. A falling score preceded a decline 1.44x more often than the base rate.
• It beat the industry-style benchmark on its own turf. We rebuilt an index from the most-cited industry methodology’s published component categories, handed it trailing revenue as an input, and ran it on the identical panel. IMPRS ordered next-year growth more accurately in every segment tested, 56.5% versus 42.3% pooled. The rebuild is a reconstruction, not the proprietary index itself, and we invite anyone to rebuild both sides independently.
• Discipline beat coverage. The forecasting layer built on the score issues a directional call only when the signal is strong, and abstains otherwise. Where it spoke, across brands and agencies pooled, it was right 84.4% of the time, which is +17.2 points of skill over the base rate. Raw accuracy without that base-rate correction flatters everyone, because revenue rises in most company-years, so we never report it alone.

09What the evidence does and does not say

A measurement company that only publishes its wins is a marketing company. The honest boundaries of the current evidence:

• One panel, iterated. The formula was refined across twelve versions on the same 122-company panel, with hold-out testing each time. That is a well-documented backtest, the first rigorous rung of the evidence ladder, and repeated iteration on one dataset can still produce optimistic results before external data arrives.
• Some tiers are small. The agency panel is 15 firms, so its confidence intervals are wide. The creator panel is 25, so we claim a level result there and no movement result at all. An earlier creator movement figure was retracted before publication when it turned out to be an artifact of baseline years.
• Comparisons hold within a tier, not across tiers. A macro account and a nano account should not be ranked against each other; small accounts grow faster off a small base, and cross-tier ordering inverts. This is published as a known limitation.
• Forecasting is bounded. The engine abstains on weak signals, and for creator accounts it suppresses downward calls entirely, because on the panel those calls inverted. An engine that knows where it is unreliable and says so is the design, not a concession.

10What a trusted measure enables

The point of a validated score is not the number. It is what becomes possible once brand health is legible to people who allocate capital.

• Agencies get receipts. A month-over-month IMPRS delta shown to a client is backed by a published movement association and ordering record, on evidence the client can inspect. Retainer conversations shift from anecdote to instrument. This is why agencies are the launch segment.
• Operators get an early-warning system. Because movement leads outcomes in the backtest, a deteriorating score is a reason to intervene quarters before the revenue line confirms the problem, and a compounding one is evidence a strategy is working before finance can see it.
• Creators and brands negotiate on health, not size. A size-independent score gives an undervalued account a way to prove it, and gives buyers a way to avoid paying size prices for decaying attention.
• Diligence gets a brand line. Investors and acquirers routinely price brand equity on narrative. A deterministic, outcome-validated score with a public error record is a step toward pricing it on measurement.

11What happens next

The backtest is the hypothesis. The live data is the experiment. Since July 2026, every directional call the production engine issues is timestamped and stored before the outcome is known, and is never edited afterward. As outcomes resolve, prospective performance will be published next to the original claims, including the misses, with the same prominence. First prospective results are expected in 2027, and an independent rebuild of the benchmark comparison is openly invited.

If the live numbers come in materially below the backtest, our research pages will say so, because the goal is not to defend a statistic. It is to find out whether brand health, defined this way, is as measurable as we propose it is. The evidence so far says the proposal deserves the test.

The companion documents go deeper: the Technical Paper covers the mathematics, normalization, weighting, and validation methodology, and the Developer Specification defines exactly how the production engine behaves. The live validation record, with every figure’s sample size and confidence interval, lives on the research hub.