The definition, and
the error attached to it.

Ninety-six terms an owner is expected to already know, written the way we would explain them on a first call: what it means, the formula if there is one, a worked example in real numbers, and the mistake that makes the number wrong. The last part is the reason this exists — the definitions are available anywhere.

Have us check yours See the practices

01/Unit economics

Dispute ratioWith formula: Disputed transactions as a share of settled ones, computed monthly by the card schemes — the one number in commerce with an automatic consequence attached.
Unit economicsWith formula: What one unit of the business earns and costs, measured so that the answer to “should we sell more of this” is arithmetic rather than opinion.
CACWith formula: Everything spent to acquire customers in a period, divided by the number of customers that spending actually produced.
Marginal CACWith formula: What the next customer costs rather than what the average one cost — the only version of CAC that answers whether to spend more.
LTVWith formula: The contribution a customer produces across their whole relationship with the business, discounted for the fact that it arrives later.
LTV : CACWith formula: The ratio of what a customer is worth to what they cost — a sanity check on the model, not a target to optimise.
CAC payback periodWith formula: How many months of a customer's contribution it takes to earn back what they cost to acquire.
Contribution marginWith formula: What is left of a sale after every cost that exists only because the sale happened.
AOVWith formula: Revenue divided by the number of orders that produced it — the simplest lever on unit economics and the most volatile.
Churn rateWith formula: The share of customers who stop buying in a period, and therefore the reciprocal of how long a customer lasts.
Retention rateWith formula: The share of a cohort still active after a given interval — the same fact as churn, read from the surviving side.
Cohort analysis: Grouping customers by when they arrived and following each group separately, so that changes over time are visible instead of averaged away.
ROASWith formula: Revenue attributed to advertising divided by the advertising spend — a channel efficiency ratio, not a profitability one.

02/Demand and conversion

Drip pricing: Revealing mandatory charges one screen at a time, so the price a buyer compares is never the price they pay.
Decoy effect: Adding an option nobody is expected to buy, priced so that one of the real options stops looking expensive.
Forced continuity: A trial that converts to a paid subscription by default, where the operative cancellation deadline is earlier or harder to reach than the offer implies.
Lead: A contact who has given permission to be contacted about a specific thing. Not a customer, not a prospect, and not a metric a business should be run on.
MQL and SQL: Two agreed thresholds on the same contact: one that marketing will vouch for, and one that sales has accepted and is working.
Pipeline stage: A named position in the sales process defined by something the buyer did, not by how the seller feels about it.
Conversion rateWith formula: The share of a defined population that completed a defined action — meaningless until both the population and the action are stated.
Funnel: An ordered model of the steps between first contact and purchase, used to locate where people stop rather than to describe how they think.
Attribution: The rule that decides which touchpoint gets credit for a sale — a modelling choice, never a fact.
UTM parametersWith formula: Tags appended to a link so that analytics can tell where a visit came from — useful exactly to the degree that the naming is disciplined.
Session, user, event: The three units analytics counts in, each measuring something different and none of them a person.
Bounce rate and engagement rateWith formula: Two mirror metrics for whether a visit went anywhere — and the definition changed underneath everyone when analytics moved to GA4.
Vanity metrics: Numbers that rise without anything else improving — followers, impressions, likes — and which, when bought, actively damage the account that carries them.

03/Product and interface

Dark pattern: An interface arranged so that the outcome the business wants is cheaper for the user to reach than the outcome the user came for.
User flow: The ordered path a person takes through a product to complete one goal, including what happens when it goes wrong.
Information architecture: How the content and functions of a product are named, grouped and nested — the structure a person navigates before they read anything.
Jobs to be done: The framing that a customer is not buying a product but hiring it to make progress in a situation.
Wireframe, mockup, prototype: Three different artefacts, each answering a different question, routinely used as if they were three levels of polish.
MVP: The smallest thing that can be built to test the assumption most likely to kill the business — not the smallest sellable version of the eventual product.

04/Search and the web

SEO: Making a site findable by search engines for the questions its buyers actually ask — a structural property of the site far more than a list of tasks performed on it.
Structured data: Machine-readable statements embedded in a page that say what its content means rather than how it looks.
Canonical URL: The one address a piece of content declares as its real one, so that every variant of it consolidates into a single entry rather than competing with itself.
Core Web VitalsWith formula: Three measurements of how a page feels to use — how fast the main content appears, how quickly it responds, and how much it moves while loading.
Black-hat SEO: Tactics that manufacture ranking signals rather than earn them — cheap to buy, effective for a while, and the penalty outlives the supplier.
Impression: One appearance of your page in a result set a person was served — counted whether or not they ever scrolled far enough to see it.
Click: A click on your result in a search engine. It is not a session in your analytics, and the two will never reconcile.
Click-through rateWith formula: The share of appearances that produced a click — and a rate that falls when a site starts being shown for more things.
Average positionWith formula: The mean rank of your appearances weighted by how often each occurred — a single number averaging page one with page nine, and among the most misread statistics in search.
Domain authority: A search tool vendor’s model of how likely a domain is to rank. It is not a search engine signal, no search engine publishes one, and no quantity of it has ever caused a ranking.
Internal link graph: Which of your pages link to which others — the structure that decides what a reader and a crawler can reach from the page they are actually on, as opposed to what you have published.
Index coverageWith formula: The share of the pages you published that a search engine has actually stored — and a gap between the two is ordinary rather than a fault.
Crawl budget: How much attention an automated reader spends on your site — influenced by what you publish and never controlled by it.
Referring domain: One site linking to you, however many times it does so — the unit that carries information, and not the one quoted in proposals.
Rank tracking: A simulated search run on your behalf by a tool. It is not what your customer sees, and the distance between the two is the whole of what a ranking report leaves out.
SERP feature: Everything on a results page that is not one of the ten links — and most of it answers the question without sending anyone anywhere.

05/Assistants and answers

GEO: Being citable by an assistant that is assembling an answer, rather than merely rankable by an engine that is returning links.
Assistant fetch: One request for a page by a declared automated client of an AI system — the only unit of attention in answer engines that a site can count on its own server.
Citation: Being named or linked inside an answer a machine assembled — the unit of answer-engine visibility, and the one your own server cannot see.
Citation shareWith formula: How often you appear among the sources for a set of questions — a number whose entire meaning lives in the set, which is the part suppliers omit.
Retrieval: How an assistant chooses which pages to open before it writes — a selection problem, not a ranking one, and it rewards different things.
Entity: A thing a machine holds a record of — a company, a person, a product — as opposed to a set of pages that happen to mention a name.
llms.txt: A file proposing which of your pages a machine should read first. It is a community convention rather than a standard, and nobody is obliged to fetch it.

06/Systems that answer

Conversation: One exchange between a person and the system, from their first message to the point it is over — and where that point is decides every rate computed below it.
Answer rateWith formula: The share of conversations the system finished on its own — and on its own is doing more work in that sentence than it looks.
Refusal rateWith formula: How often the system says it does not know — the one metric on a deployment where zero is the alarming reading.
Escalation rateWith formula: How often a person takes over — and whether they arrive holding the conversation or starting one.
Knowledge base: The material the system answers from. It is the product; the model is a commodity you rent by the token.
RAG: Look the answer up first, then have the model phrase it — an architecture that turns a question about the model into a question about your material.
Grounding: Every sentence the system produces is traceable to a passage it retrieved — which means an answer can be checked without asking the model to explain itself.
Refusal path: The route the system takes when it has nothing — designed, tested, and the single most valuable thing in a deployment.
Hallucination: A fluent, confident claim with nothing behind it — and the reason it is dangerous is that it is indistinguishable from a correct answer at the moment it is read.
Deterministic layer: The code that holds a veto over what the model produces — rules that always run and always give the same answer to the same input.
Adversarial test set: The questions written specifically to make the system invent an answer — and the only acceptance evidence that means anything.

07/State and recovery

Transaction: A change that either happens completely or does not happen at all — and the argument is always about where it ends.
Domain invariant: A statement about your data that must never be false — enforced by the system that stores it, not by the code that happens to be running.
Idempotency: Sending the same request twice has the same effect as sending it once — the single most valuable property in anything that touches money.
Consistency: Whether two people looking at the system at the same moment see the same thing — and how long “eventually” is allowed to be.
Failure isolation: How much stops working when one thing stops working — a property you design in, and discover the value of exactly once.
Recoverable copy: A copy somebody has restored from. Until then it is a hypothesis, and the word backup is what the hypothesis is called before it is tested.
RPOWith formula: How much recent work you have decided you can afford to lose — expressed in time, and settled by the business rather than by the infrastructure.
RTOWith formula: How long you have decided the business can be down — and the only version of the number worth having is one somebody has timed.
Migration: Moving data from one shape to another while the business keeps running — and the risk is never the data, it is the interval where two shapes are both live.
Data reconstruction: Rebuilding what the system knew from whatever survived it — the discipline that exists because the copy you were counting on is the thing that failed.

08/Buying and accepting

Acceptance criteria: The conditions that decide whether the work is finished — written before it starts, or they will be written afterwards to describe what arrived.
Scope of work: What is being bought, written so that it can be finished — which means written backwards from the acceptance criteria rather than forwards from the wish.
SLAWith formula: A promise with a number and a consequence attached. Without the consequence it is a wish with a percentage in it.
Change order: The written record that scope moved and what it now costs — and in the UAE, the thing without which the extra work is not billable.
Vendor lock-in: What it would cost to leave — and it is a design property of the arrangement rather than a personality trait of the supplier.
Exit clause: What the contract says happens on the day the relationship ends — written while everyone is friendly, because it is unwritable afterwards.
Source code escrow: Source and build instructions held by a third party, released on stated events — insurance against a supplier that stops existing rather than one that stops replying.

09/Security

Two-factor and multi-factor authentication: Requiring a second proof of identity beyond the password — and the factors differ enough in strength that “we have 2FA” says almost nothing on its own.
Account recovery path: The chain of accounts that can reset each other's passwords — and therefore the real security level of every account in it, which is the level of the weakest.
SPF, DKIM and DMARCWith formula: Three DNS records that together decide whether anyone can send mail pretending to be your domain — and whether your own mail reaches the inbox.
Business email compromise: Fraud carried out through a real or convincingly forged company mailbox, usually to redirect a payment — and by value, the most damaging attack a normal business faces.
Password managers and passkeys: Two mechanisms that remove the human from the part of authentication humans are worst at — remembering secrets, and judging which site to give them to.

10/Access and continuity

Consent record: The evidence that a specific person agreed to a specific thing at a specific time — which is what an authority asks for, rather than the tick box that produced it.
Offboarding: The sequence that removes a departing person's access to everything, in an order that does not lock the company out of its own systems on the way.
Least privilege: Giving each person the narrowest access that lets them do their job, so that a compromised or departing account costs the least it can.
Named accounts and shared credentials: Access issued to a person rather than to a login everyone uses — the difference between being able to revoke one person and having to change everything.
Key-person risk: The number of people whose sudden absence would stop something important — and in most small companies the answer for the digital estate is one.
Digital asset register: A written list of every account, domain and system the business depends on, recording who owns it, who administers it, what it costs and how it is recovered.
Handover: The point at which a vendor's work becomes the company's property — defined by what is transferred, not by the project being finished.
Vendor selection and total costWith formula: Judging a supplier on what the work costs over its life — including the rebuild — rather than on the quotation.

Not because anyone was careless. The denominators drift, the definitions were never written down, and nobody has had a reason to check. That is what the audit is.

Start with an audit Discuss an engagement

The definition, andthe error attached to it.

Whether one more customer makes the company money, and how long it takes to find out.

What is counted between a stranger and a paying customer, and where the counting usually goes wrong.

The shapes a product is described in before it is built, and what each of them is actually for.

How a site is found, by a search engine and by an assistant assembling an answer.

Being retrievable and citable by a machine assembling an answer — a different mechanism from ranking, with a different and much smaller set of things that can honestly be measured.

Building something that answers on your behalf — where the failure that costs money is not a wrong answer but a confident one, and the fix is a refusal that works rather than a better model.

What the system believes is true, what guarantees that belief, and what happens on the day it is lost — the part of a platform nobody sees until it is the only thing that matters.

The buyer's side of a contract for technical work — what is being bought, what finished means, what leaving costs, and which of those is decided before the money moves.

The accounts a business is actually attacked through — mail, social, and the recovery path behind both.

Who can reach what, what happens the day they leave, and whether the company still owns its own presence when they do.

Most of these are wrong somewhere in your own numbers.

The definition, and
the error attached to it.