# Account audit scorecard — research

Version 1.0 research, 2026-09-29. Recorded so that any future change to a
threshold can be traced to the reasoning that produced it. **If you change a
constant in `Account-audit-scorecard.js`, change it here too.**

## The research question

Account audit scorecards are everywhere, and almost all of them do the same
thing: compare the account against a benchmark, then sell the plan to reach it.
The question that decides whether a scorecard is a measurement or a sales
technique is simple — **whose target is it?**

If the target is an industry benchmark the operator never agreed to, the score
measures resemblance to somebody else's account. If the target is the
operator's own plan, the score measures attainment. The second is the only one
that answers a question worth asking.

So the research question for this tool became: *can a useful health score be
built entirely from the operator's own inputs and targets, with no external
benchmark anywhere?* The answer is yes, and the constraint is what makes it
honest.

## Competitors reviewed

| Tool | What it does | The gap |
| --- | --- | --- |
| **Optmyzr Account Health Score** | Industry benchmarks for pacing, wasted spend, conversion rate, budget utilisation, campaign structure; weighted into a single grade | The benchmarks are proprietary and unstated. A good score is largely a statement that the account resembles Optmyzr's user base. Also requires a login. |
| **WordStream/LocaliQ Quality Score** | A grade built on a defined set of quality inputs | Pulls most inputs from a connected account. Free tier limited. The score is a gate to a paid tier, and the weights are not published. |
| **SEMrush PPC Keyword Tool** | Account-level audit against a keyword database | Needs a live account connection, which this project refuses on principle. Nothing is calculated locally. |
| **Adwords/Google Ads own recommendations panel** | Platform-side "optimisation score" out of 100 | Scores the *presence* of recommended settings, not performance. A perfectly configured account that loses money scores 100. A useful lesson, inverted. |
| **Agency in-house scorecards (spreadsheet)** | Usually conversion rate, CPA and ROAS against a target | Genuinely the right shape and the direct ancestor of this tool. Nearly always undocumented weights, and usually a single blended number with no per-component view. |
| **AdClearer / Clickguard** | Free calculators for pacing and waste | Single-metric calculators. Useful, but they answer a question and stop; there is no roll-up and nothing to trend month to month. |
| **Klipfolio / Whatagraph** | Account health dashboards with a score | Dashboards that need the account connected and data leaving the browser. |
| **Agency dashboard templates (Notion, Looker Studio)** | Averages several KPIs | The typical failure: the score is an unweighted mean, so one catastrophic component is diluted by four healthy ones. |

### What the category consistently gets wrong

1. **Undisclosed weights.** A weighted average is a value judgement. Presenting
   one as a measurement without showing the weights is the central problem.
2. **Imported benchmarks with no source.** "Your conversion rate is below
   average" is meaningless without knowing whose average, over what mix of
   industries, and in which year.
3. **Unweighted means.** If return on ad spend is 1.1:1 and the other four
   components are perfect, an unweighted mean says the account is fine. It is
   not; the account is losing money.
4. **Scores that punish missing data.** Scoring an unmeasurable component as
   zero conflates "performing badly" with "we could not measure it", and pushes
   operators towards invented figures to make the number look better.
5. **No trend.** One reading tells you where you are today, which you already
   knew. The value is in the series.

### Demand signals

- The most common practitioner question in PPC communities is not "what is my
  ROAS" but "is this account healthy" — the health-score concept is asked for
  directly, and there is no free tool that answers it without a login.
- Recurring complaints about the logged-in options: cost, data leaving the
  browser, and the sense that a score exists mainly to justify a subscription.
- Agency spreadsheets in the same shape as this tool circulate widely, which is
  the clearest evidence that the model is wanted. They are just not documented.

## Decisions taken

### Weights — judgement call, printed on the page

| Component | Weight | Reasoning |
| --- | --- | --- |
| Cost per conversion | 25 | An account can survive a mediocre click-through rate. It cannot survive paying more per order than the order is worth. |
| Return against break-even | 25 | Return is the only component that decides whether the account is a business or an expense. |
| Waste share | 20 | Waste is recoverable this week; poor return usually is not. |
| Budget utilisation | 15 | Under-spending means the plan was never tested; over-spending means it was never agreed. Both are failures, and they are separable from performance. |
| Conversion rate | 15 | A lever that does not cost money, and the slowest of the five to move. |

Rejected: equal weighting (dilutes the two components that decide
profitability); conversion rate weighted highest (it is the least responsive and
the most argued about). The weights sum to 100 so they read as percentages of
influence, and the reasoning is in the page rather than only here.

### Penalty rate 0.5 — judgement call

Ratio 1.00 or better scores 100. Below that, `score = 100 × (1 − 0.5 × (1 − ratio))`,
floored at 0. So a ratio of 1.50 scores 75, 2.00 scores 50, 3.00 scores 0.

Rejected: a cliff at the target (unstable, and a 1% miss reads identically to a
20% miss); a linear score to zero (too harsh — a 2× miss is a problem worth
flagging, not a catastrophe). 0.5 chosen so a 50% miss lands at 75, which
keeps a healthy account with one weak component out of the "needs work" band.

### Budget utilisation band 95–100% — judgement call

The only component with a good and a bad direction, so it is scored on distance
from 100% rather than on a ratio. Inside 95–100% it scores 100. Outside, the
score falls at the same 0.5 rate per percentage point of distance, relative to
the 95 floor.

Rejected: treating under-spend as free (it usually means the plan was never
tested, so the month produced no evidence); requiring exactly 100% (spend
fluctuates daily, and demanding precision the platform does not promise would
flag healthy accounts).

### Grade edges 85 / 70 / 50 — judgement call

Strong, healthy, needs work, urgent. Each is paired with a stated meaning and a
stated next action, because a grade without a consequence is decoration.

### What is a fact and what is a judgement

**Standard ratio definitions, not judgements:** cost per conversion = spend ÷
conversions. Return on ad spend = revenue ÷ spend. Conversion rate = conversions
÷ sessions. Break-even return = 1 ÷ gross margin fraction — the ratio at which
advertising consumes the entire contribution.

**Judgement calls, all named constants:** the five weights, the 0.5 penalty
rate, the 95–100% utilisation band, and the four grade edges.

## Deliberately rejected

- **Impression share.** The single most-cited number in the trade, and it is
  only visible inside the platform. Guessing it would be inventing the most
  requested figure in the category. The tool instead tells the reader to go and
  look at it.
- **Any competitor benchmark.** There is no reliable, free, current,
  industry-adjusted source of one. A benchmark with no stated source is
  decoration.
- **Revenue or performance forecasting.** A projection from one period of data
  is a confident-looking fiction. This is the category where a tool cannot be
  both free and honest, and the project rule is to withhold the figure.
- **Seasonality adjustment.** Needs at least two full years of history to
  separate a bad month from a bad quarter.
- **Scoring unmeasurable components as zero.** Rejected for the reason above:
  it punishes a gap in data rather than a gap in performance, and it rewards
  inventing a number. Excluded and labelled instead.
- **A named overall verdict phrased as a recommendation.** "Reduce bids by 20%"
  cannot be justified from ten inputs. The tool names the weakest component and
  states the next measurement to take.

## Known limitations, stated on the page

- Ten inputs describe an account in aggregate. Concentration risk — one campaign
  carrying the whole account — is invisible to this score.
- The revenue figure is taken on trust. Attributed revenue is the easiest
  number in an account to overstate, and no ratio can detect that.
- Break-even return assumes margin is uniform across the product mix. A blended
  margin can hide a loss-making line.
- A score is a compression. Every excluded component is invisible in the total,
  which is why the component table is given equal prominence to the score.
- Targets are set by humans, and a target set from last period's outcome measures
  nothing. The page says this explicitly because it is the most likely misuse.