Resource, library
Metric library
The measures every business runs on. Each one with the formula, the variants in use, and the question to ask before you quote it.
Most arguments about numbers are arguments about definitions. This library gives one clear version of each measure, and names the variants so you can tell which one you are looking at.
RevenueFinance
The value of what you sold in a period, before any costs.
Sum of invoice value in the period
1,240 orders averaging €340 gives €421,600 for the month.
Variants people actually use
- Gross versus net of returns and discounts
- Booked versus recognised, which can differ by months
- Including or excluding VAT
What it hides. Whether it came from many small customers or one large one, and whether it repeats.
Ask this. Is this recognised revenue or booked? They are different numbers and both are called revenue.
Gross marginFinance
What is left of revenue after the direct cost of delivering it.
(Revenue − cost of goods sold) ÷ revenue × 100
€420k revenue and €250k direct costs gives a 40.5% gross margin.
Variants people actually use
- What counts as a direct cost: shipping, payment fees, support
- Per product, per customer, or blended
- Before or after discounts
What it hides. A blended margin hides the product that loses money on every sale.
Ask this. What is in cost of goods sold here? Two teams rarely include the same things.
EBITDAFinance
Profit before interest, tax, depreciation and amortisation. A rough proxy for operating cash generation.
Operating profit + depreciation + amortisation
Operating profit of €1.2m with €300k depreciation gives EBITDA of €1.5m.
Variants people actually use
- Adjusted EBITDA, which excludes whatever management considers one-off
- Run-rate EBITDA, annualising a single good month
What it hides. Capital spending, working capital and debt. A business can have strong EBITDA and no cash.
Ask this. What has been adjusted out, and would you have adjusted the same things?
Customer acquisition costSales & marketing
What it costs, on average, to win one new customer.
Total sales and marketing spend ÷ new customers acquired
€90k of spend winning 300 customers gives a CAC of €300.
Variants people actually use
- Including or excluding salaries
- Paid channels only, or all marketing
- By channel, which is where it gets useful
What it hides. That a cheap channel may bring customers who churn in three months.
Ask this. Does this include the team, or only the media spend?
Customer lifetime valueSales & marketing
The total gross profit you expect from a customer across the whole relationship.
Average order value × orders per year × gross margin × expected years
€85 × 4 × 40% × 3 years gives a lifetime value of €408.
Variants people actually use
- Historic versus predicted
- Revenue-based, which overstates it, or margin-based
- Capped at a horizon, usually three years
What it hides. The variation. The average is often pulled by a handful of very large customers.
Ask this. Is this margin or revenue? A revenue-based figure is not comparable with acquisition cost.
Churn rateCustomer
The share of customers who stop buying in a period.
Customers lost in period ÷ customers at start of period × 100
46 lost from 1,200 at the start of the month is 3.8% monthly churn.
Variants people actually use
- Customer churn versus revenue churn, which can move in opposite directions
- Voluntary versus involuntary, such as a failed card
- Gross versus net of upgrades
What it hides. Which customers left. Losing your ten largest accounts and losing two hundred small ones look identical here.
Ask this. Customer churn or revenue churn? And what counts as lost: cancelled, or inactive for ninety days?
Net revenue retentionFinance
How revenue from existing customers changes over a year, including upgrades, downgrades and churn.
(Starting revenue + expansion − contraction − churn) ÷ starting revenue × 100
€1m growing to €1.12m from the same customers gives 112% net revenue retention.
Variants people actually use
- Including or excluding new customers, which should always be excluded
- Cohort-based or blended
What it hides. That strong expansion from a few accounts can mask heavy churn among the rest.
Ask this. Does this include new customers? If it does, it is a growth rate wearing a retention label.
Conversion rateSales & marketing
The share of people who take the action you wanted.
Conversions ÷ total opportunities × 100
480 purchases from 12,000 visitors is a 4% conversion rate.
Variants people actually use
- Which step you measure from: visit, session, or qualified lead
- Per channel, per device, per campaign
What it hides. The volume behind it. Ten per cent of forty people is four.
Ask this. Conversion from what to what? Both ends of the funnel need naming.
Inventory turnoverOperations & supply chain
How many times stock is sold and replaced in a period.
Cost of goods sold ÷ average inventory value
€2.4m of cost against €400k average stock is a turnover of 6.
Variants people actually use
- By category, which is the only level where it is actionable
- Using ending rather than average inventory
What it hides. Which lines are turning fast and which have not moved in a year.
Ask this. Average or ending inventory? On a seasonal business the two differ sharply.
On-time in fullOperations & supply chain
The share of orders delivered complete and on the promised date.
Orders on time and complete ÷ total orders × 100
920 of 1,000 orders arriving complete and on time is 92% OTIF.
Variants people actually use
- Whose date: the original promise or the revised one
- Line-level or order-level, which gives very different figures
What it hides. How late the late ones were. A day and three weeks count the same here.
Ask this. Is this measured against the original promised date or the latest one?
First response timeCustomer
How long a customer waits for the first human reply.
Median time from ticket created to first agent response
A median of 42 minutes against a four-hour target.
Variants people actually use
- Mean or median, and the mean is almost always misleading here
- Business hours only, or around the clock
What it hides. Whether the first reply actually helped. A fast holding message still leaves the problem open.
Ask this. Median or average? And does the clock stop overnight?
Attrition rateHR & people
The share of employees who leave in a period.
Leavers in period ÷ average headcount × 100
18 leavers from an average headcount of 240 is 7.5% annual attrition.
Variants people actually use
- Voluntary, involuntary, or regretted, which is the one that matters
- Annualised from a month, which exaggerates small numbers
What it hides. Who left. Losing your three best engineers is not the same as losing three people.
Ask this. Is this voluntary attrition, and does it separate regretted leavers?
Time to hireHR & people
How long it takes to fill a role, from approval to acceptance.
Median days from requisition approved to offer accepted
A median of 38 days across 22 hires last quarter.
Variants people actually use
- From advert or from approval, which can differ by weeks
- Time to hire versus time to fill, including notice periods
What it hides. The roles you failed to fill at all, which drop out of the calculation.
Ask this. Does the clock start at approval or at advertising?
UtilisationOperations & supply chain
The share of available capacity actually used.
Used capacity ÷ available capacity × 100
1,480 billable hours from 1,850 available is 80% utilisation.
Variants people actually use
- Billable versus productive, which are not the same
- Against contracted hours or actual hours worked
What it hides. Whether the work was worth doing. A hundred per cent utilisation on low-value work is a problem.
Ask this. What counts as available? Holidays and training change this figure a great deal.
Forecast accuracyStrategy & performance
How close the forecast was to what happened.
APE = |actual − forecast| ÷ |actual| × 100, only for non-zero actual
Forecast €1.0m against actual €1.1m gives 9.1% absolute percentage error. A score of 100% minus APE would be 90.9% but can be negative for large misses.
Variants people actually use
- Which forecast version: the first or the last revision
- Weighted by value, or a simple average across items
What it hides. Bias. Consistently forecasting low scores the same as consistently forecasting high.
Ask this. Which forecast version and denominator? With zero actual, APE is undefined: report absolute error or a documented aggregate measure. Check bias separately.
Data freshnessData architecture & governance
How recently the data was updated, against how recently it should have been.
Time since last successful refresh
Refreshed 14 minutes ago, against a one-hour target.
Variants people actually use
- Source freshness versus dashboard freshness
- Per table, or for the slowest table in the report
What it hides. Whether the refresh succeeded. A pipeline can complete and load nothing.
Ask this. Is this when the report refreshed, or when the source data was last true?