Methodology · version 2026.08.v1

How the Marketing Efficiency Index is scored

Every weight, every scoring ladder, every hard limit and the exact conditions under which our weights do not apply. You can disagree with the model, which is the point.

The six pillars

Economics: Work on: knowing what a customer costs, at what granularity, against which margin, and how long they take to repay it.
Demand: Work on: where new customers come from, how concentrated that is, and how much of your money creates demand rather than harvesting it.
Conversion: Work on: the path from attention to revenue, on the site and in the follow-up. For most businesses this is where the money leaks.
Measurement: Work on: tracing closed revenue back to cause, and feeding it back so bidding optimises to money rather than to form fills.
Retention: Work on: whether a customer's value compounds after the first sale, deliberately rather than by accident.
Cadence: Work on: how long it takes to turn noticing a problem into changing something. This one costs nothing to fix.

The weights

These are expert-judgement weights informed by the evidence, not empirically derived importance coefficients. No dataset exists that regresses these six pillars against firm performance. Swapping them for equal weights barely changes anyone’s ranking, which is the honest defence and a stronger one than pretending they were derived.

Business model	Economics	Demand	Conversion	Measurement	Retention	Cadence
Ecommerce	26%	16%	18%	20%	12%	8%
Subscription	20%	16%	16%	18%	22%	8%
Sales-led	20%	20%	24%	20%	8%	8%
Booking-led	20%	20%	24%	18%	10%	8%
Several, evenly	21.5%	18%	20.5%	19%	13%	8%

How the pillars combine

Compensability is chosen separately at each level. Within a pillar, substitution is allowed: two ways of knowing what a customer costs are genuinely interchangeable. Across pillars it is not: knowing what a customer costs and being able to trace revenue to cause are not substitutes for one another. A geometric mean is concave, so a decline hurts more than an equal improvement helps, and a broken pillar cannot hide behind a strong one. Every pillar is floored at 10 because a geometric mean is zero if any term is zero, which would annihilate the index rather than penalise it.

Within a pillar: Unweighted arithmetic mean of the questions asked in that pillar.
Across pillars: Weighted geometric mean: exp(sum(weight × ln(pillar))).
Pillar floor: 10

The four hard limits

Each of these caps the index regardless of everything else. They fire often and bind rarely, and that is correct: most respondents a limit fires on are already below its ceiling on merit. We report a limit whether or not it lowered the score, because the firing is the finding even when the binding is not.

Spending without sight ceiling 60
Triggers when: measurement pillar below 30
Single-channel dependence ceiling 75
Triggers when: Q3 answered a
Unknown economics ceiling 65
Triggers when: Q1 answered a or DK, and Q2 answered a or DK
Speed to lead ceiling 70
Triggers when: business model is sales or book, and BL1 or BK1 answered a

When more than one fires, the tightest ceiling applies. The ceilings are deliberately all different so the limit that actually bound is always identifiable.

The four positive signals

+4 You settle disagreements with an experiment
+3 Closed revenue feeds back into the ad platforms
+3 You sustain an evidence-based create-versus-capture split
+3 You are investing in demand creation at a stage-appropriate level
+2 You run pre-registered tests at cadence

Capped at 8 and scaled by headroom: bonus_applied = min(bonus, bonus × (100 − raw) / 50). Without that scaling a strong respondent reaches 99, which would overstate the precision of a fourteen-question instrument.

You told us you run tests against a threshold agreed in advance, and also that closed revenue cannot be traced back to what caused it. Those two things cannot both be fully true: a test needs an outcome measure, and the outcome that matters is revenue. We have not credited the positive signals on this page until that is resolved.

Where our own weights do not apply

Under £1m of revenue the create-versus-capture ladder inverts at the top. Sustained heavy investment in demand creation before product-market fit is a failure mode rather than a virtue, so a quarter to a third scores full marks at this size and a sustained 40 to 60% scores slightly lower.

Q4: b → 75, c → 100, d → 85

The visibility index

100 × (1 − don't-know count / don't-know-capable questions on your path)

The denominator counts the questions that offer a don't-know, not all fourteen. Two do not: the question about deliberate tests, because a leader who does not know whether tests are being run has answered the bottom rung; and, on the ecommerce path, the multi-select, whose honest bottom rung is 'none of these'. A flat denominator of fourteen would make 0% visibility unreachable and the number a quiet lie.

The confidence band is ± (2 + 1.6 × don't-know count).

The bands

0-29 Spending Blind. You are spending money without a reliable way to know what it does.
30-49 Leaking. The demand is there. It is escaping between the stages.
50-69 Functioning. It works. It does not yet compound.
70-84 Compounding. Money in reliably becomes more money out.
85-100 Frontier. Very few companies operate here.

The recoverable-spend model

A modelled estimate, not a measurement, and the weakest-warranted figure on the report. The waste curve is clamp(0.05 + 0.15 × (1 − MEI/100), 0.05, 0.2), and we take 30% off our own estimate at the low end. Spend rates come from The CMO Survey 2026, marketing spend as a share of revenue by firm type.

Marketing spend as a share of revenue

Ecommerce: 12%
Subscription: 10.1%
Sales-led: 7%
Booking-led: 7.2%
Several, evenly: 9.1%

Revenue assumed per band

under-1m: £500,000
1m-5m: £3,000,000
5m-20m: £12,000,000
20m-100m: £55,000,000
over-100m: £100,000,000

The two open-ended bands have no midpoint, so the top one is seeded from its lower bound, which is conservative. Correct the figure on your own report and the number moves.

Every scoring ladder

Fixed goalposts with published meanings, held constant rather than derived from the respondent pool, so a score does not move when nothing about you has changed. A don’t-know scores 15: not zero, because not knowing what a customer costs is not the same as knowing it is terrible; and not the midpoint, because imputing uncertainty to the middle is the practice that most threatens validity.

Q1 What does it cost you to acquire one new customer?

0 — We do not calculate it.
35 — We have a rough idea from what the ad platforms report.
62 — We calculate blended cost: all sales and marketing spend divided by new customers.
85 — We calculate it by channel, including staff and agency cost, and review it monthly.
100 — As above, and we know it against contribution margin rather than revenue.
15 — I do not know.

Q2 How long before a new customer has paid back what it cost to win them?

15 — Longer than two years, or we are not confident they ever do.
58 — Somewhere between one and two years.
85 — Between six and twelve months.
100 — Under six months, or the first sale is profitable after all variable costs.
15 — I do not know.

Q3 Where do new customers actually come from?

15 — Almost all from one channel. If it stopped tomorrow, new business would stop.
45 — One channel does most of it, and a second is starting to work.
78 — Two or three channels each produce a meaningful share.
100 — Three or more do, and we know the cost and the lead quality of each.
15 — I do not know where new customers come from.

Q4 Of your marketing money, how much creates demand from people who are not looking for you yet?

30 — Effectively all of it captures existing demand: search, retargeting, inbound.
60 — Mostly capture, with occasional awareness activity.
85 — Roughly a quarter to a third creates new demand.
100 — Around 40 to 60% creates new demand, sustained, and we watch it over a longer window.
15 — We have never split it that way.

Q5 What proportion of your website visitors do the thing you want them to do?

35 — There is a number in analytics but we do not really trust it.
62 — We know it overall, and it is roughly flat year on year.
88 — We know it by device and by traffic source, and how it compares to our category.
100 — As above, and we have deliberately moved it in the past twelve months.
15 — We do not measure it.

Q6 When a sale closes, can you trace it back to what caused it?

10 — Not really. We look at the platform dashboards and take a view.
38 — We can see the last click in analytics, and that is where it ends.
65 — Sales and revenue sit in a CRM and we can attribute them to a channel.
90 — As above, and closed revenue is fed back into the ad platforms.
100 — As above, and when methods disagree we settle it with a holdout or geo test.
15 — I do not know.

Q7 What is the first number your leadership team looks at to judge whether marketing is working?

15 — Traffic, impressions, followers or engagement.
50 — Leads or orders.
58 — Cost per lead, or ROAS from the ad platforms.
88 — Qualified pipeline, or revenue we can attribute to marketing.
100 — Contribution margin, or profit after acquisition cost.
15 — We do not have one number we look at first.

Q8 How much of your revenue comes from customers you already had?

30 — Almost none. It is a one-off purchase and we treat it that way.
52 — Some, but it happens on its own. We do not do anything to cause it.
85 — A meaningful share, and we run programmes deliberately to grow it.
100 — As above, and we track it by cohort and it is improving.
15 — I do not know.

Q9 From noticing something is underperforming to actually changing it, how long?

15 — We usually notice at the quarterly review, if then.
45 — Within a month or so.
75 — Within a week.
100 — A day or two, because someone owns the number and can act without asking.
15 — I do not know.

Q10 In the last 90 days, how many deliberate tests did you run where you decided in advance what would count as a win?

10 — None.
38 — One or two, but we judged them by eye afterwards.
65 — One or two, against a threshold we agreed beforehand.
92 — Three to eight, against a pre-agreed threshold, results written down.
100 — More than eight, on a schedule, with a log of what we learned.

BE1 When did someone here last try to buy from you on a phone, and write down what got in the way?

15 — Never that I know of.
40 — Not in the past year.
65 — In the past year, informally.
88 — In the past quarter, and it produced a list of fixes that got done.
100 — On a schedule, and we test the fixes against a control.
15 — I do not know.

BE2 For every £1 of ad spend, how much total revenue does the business make, and where does that stop being profitable?

15 — We do not track blended efficiency.
45 — We track it, but we do not know our breakeven.
78 — We know both, and we check monthly.
100 — We manage to a contribution-margin breakeven, not a revenue multiple.
15 — I do not know.

BE3 What share of customers buy again within twelve months, and does the first order make money on its own?

20 — We do not know either.
50 — We know roughly one of the two.
80 — We know both, and the first order is around breakeven.
100 — We know both by cohort, and the first order is profitable.
15 — I do not know.

BE4 Which of these run automatically, without anyone remembering to set them off?

25 — Abandoned cart or checkout recovery
20 — Welcome or first-purchase sequence
15 — Post-purchase follow-up
15 — Win-back for lapsed customers
15 — Behaviour-triggered cross-sell
10 — Review or referral request
0 — None of these

BS1 Between signing up and getting the result they came for, how many of your new customers actually get there?

15 — We do not measure activation.
45 — We measure signups, not activation.
75 — We measure activation and know the drop-off point.
100 — We measure time-to-first-value by cohort and have shortened it.
15 — I do not know.

BS2 What happens to the revenue from a cohort of customers over the following twelve months?

20 — It shrinks and we do not know by how much.
50 — It shrinks; we know roughly.
80 — It roughly holds.
100 — It grows, because expansion outruns churn.
15 — I do not know.

BS3 Of the customers who leave, how many left because a payment failed rather than because they chose to go?

15 — We have never separated the two.
45 — We could work it out but have not.
78 — We separate them and recover some failed payments.
100 — We separate them and actively run dunning against a target.
15 — I do not know.

BS4 How does an existing customer end up spending more with you?

15 — They do not, really.
50 — If they ask.
85 — We prompt it, but not systematically.
100 — There is an owned motion with a target, triggered by usage.
15 — I do not know.

BL1 When an enquiry arrives during working hours, how long before a human responds?

10 — Next working day or later, or there is no rule.
40 — Within a few hours.
65 — Within an hour.
88 — Within five minutes, and it is monitored.
100 — Within five minutes, monitored, and routed by lead value.
15 — I do not know.

BL2 How many attempts before you stop chasing an enquiry?

15 — One, then it goes cold.
45 — Two or three.
75 — Four or five.
100 — Six or more, on a defined cadence across channels.
15 — I do not know.

BL3 Of the enquiries marketing produced last month, what share did someone actually contact?

15 — I would not be able to find out.
50 — Most of them, I think.
80 — We know the number and it is above 80%.
100 — We know it, it is above 95%, and unworked leads are flagged automatically.
15 — I do not know.

BL4 After a customer buys, what makes the next sale or the referral happen?

20 — Nothing structured.
50 — The account manager, when they have time.
80 — A defined process with owners and timing.
100 — As above, and sales feeds lead quality back in a way that changes the spend.
15 — I do not know.

BK1 When an enquiry arrives during opening hours, how long before a human responds?

10 — Next working day or later, or there is no rule.
40 — Within a few hours.
65 — Within an hour.
88 — Within five minutes, and it is monitored.
100 — Within five minutes, monitored, and routed by lead value.
15 — I do not know.

BK2 When someone calls, what actually happens?

15 — It rings out fairly often and we do not know how often.
45 — We usually answer; missed calls are not tracked.
78 — We track answer rate and call back the missed ones.
100 — As above, and whoever answers is expected to ask for the booking.
15 — I do not know.

BK3 What happens when someone does not turn up, or does not rebook?

20 — Nothing automatic.
50 — Someone chases if they notice.
80 — Automatic reminders, and a rebooking prompt.
100 — As above, plus a reactivation programme with a measured recovery rate.
15 — I do not know.

BK4 How do reviews and referrals happen?

20 — When a customer decides to leave one.
55 — We ask sometimes.
85 — Every customer is asked automatically at the right moment.
100 — As above, and reviews are tracked as events we can market against.
15 — I do not know.

What this does not establish

It is a self-assessment. It measures what you know about your marketing at least as much as the marketing itself. That is partly deliberate — not knowing is a finding, and the visibility index makes it explicit — but it is a real ceiling on accuracy.
The weights are expert judgement informed by the evidence, not empirically derived importance coefficients. No dataset exists that regresses these six pillars against firm performance. Swapping them for equal weights barely changes anyone's ranking, which is the honest defence and a stronger one than pretending they were derived.
No predictive validity has been established. Nothing here demonstrates that the index correlates with actual firm performance.
The distribution is modelled, not observed. The percentile bands come from simulation with research-informed priors and will be replaced with observed data as a version change, with the old curve left published and historical scores not restated.
One respondent per company. Two executives at the same firm may answer differently, and that variance is currently invisible.
The peer set is thin at launch. No peer figure is shown below 100 completions in a business-model and revenue-band cell, which at launch is every cell.
It cannot see anything. It reads no accounts, crawls no pages, and cannot catch what it was not told.
Cadence has the weakest evidence base of the six pillars. No credible research links reporting cadence to performance, so the pillar is built on decision latency and pre-agreed thresholds, which do have supporting evidence, and it carries the lowest weight for that reason.
One figure diverges from the accompanying specification's validation table. Its all-don't-know row requires treating the deliberate-tests question as offering a don't-know; the question set states the opposite explicitly, and a stated design decision outranks a harness artefact. The reachable worst-visibility scores are therefore 14.8 and 14.1 rather than 15.0 and 14.3.

Where this model does not apply

Companies under £250k of revenue, where most of these questions have no meaningful answer.
Companies whose revenue is predominantly offline or channel-partner-led, where the path from marketing to revenue is not the one this instrument traces.
Any company in its first six months of trading, where there is not yet a repeatable system to score.