A glossary for AI⁠-⁠native agencies.

An experiment comparing two variants to see which performs better — valid only when each variant gets enough traffic (sample size) to reach statistical significance.

The decline in ad performance as the same audience sees a creative too many times — tracked through frequency, the average impressions per person.

An agency operating system is the shared layer where a creative agency's client context, working processes and recurring work live, so capability accrues to the agency rather than to individual chat histories or to an outside consultant.

A program that uses a large language model to plan, take actions through tools, and pursue a goal across multiple steps — not just reply once.

An AI assistant is a conversational tool that answers one request at a time and forgets the work between sessions. An AI agent takes a goal, plans the steps, and carries work forward using tools and shared context. For creative teams the difference is who holds the context: the person, or the system.

A repeatable test suite for an AI feature — a set of inputs, expected behaviour, and a way to score how often the system gets it right.

The coordination layer that directs multiple AI models, tools, and steps toward one goal, deciding what runs, in what order, with which context, and when a human steps in.

An AI workflow is a repeatable sequence of AI steps, from trigger to finished output, that carries a task through without a person running each step by hand.

The methodology for deciding which marketing touchpoints get credit when a customer converts — first-click, last-click, multi-touch, or model-based.

A contested PR metric that estimates the cost of buying earned coverage as paid advertising. Useful as a rough scale, widely criticised as a measure of value.

The percentage of an agency team member’s available hours that are billed to a client — the operating lever behind agency profitability.

A single hourly rate used to bill mixed-seniority work, calculated as the weighted average of the team individual rates.

The accumulated value of a brand beyond its current revenue — the recognition, trust, and pricing power it carries into every future interaction.

The sales volume at which total revenue equals total cost and profit is zero — fixed costs ÷ contribution margin per unit. The floor every campaign has to clear.

Customer acquisition cost — the total sales and marketing spend divided by the number of new customers won in the same period.

The forecast of how much billable work the team can realistically deliver in a given period, used to decide what to sell and when to hire.

The percentage of customers (or recurring revenue) that leaves the business in a given period — the single biggest determinant of LTV.

The share of impressions that result in a click — clicks ÷ impressions, as a percentage. The headline read on creative and targeting relevance.

The profit margin earned on an individual client account, not just the agency average — the number that tells you which clients are worth fighting for.

The discipline of deciding what information to put in front of an AI system at the right time — and what to leave out — so it can do useful work.

The amount of text, measured in tokens, that an AI model can consider at once. It covers the prompt, any documents or history you include, and the model's own reply.

The percentage of visitors or sessions that complete a desired action — conversions ÷ total. The hinge between traffic and revenue.

The fully-loaded cost of an event divided by a chosen outcome — attendees, qualified leads, or meetings. The efficiency metric for experiential and field marketing.

Cost per action (or acquisition) — ad spend divided by the number of actions, such as purchases or sign-ups. The bottom-line efficiency metric for performance media.

Cost per click — the average amount you pay each time someone clicks your ad. Calculated as ad spend ÷ clicks.

Cost per lead — ad spend divided by the number of leads captured. The headline acquisition-efficiency metric for B2B and considered-purchase marketing.

Cost per mille — the price to serve an ad 1,000 times. The baseline unit of cost for awareness and reach campaigns across display, video, social, and out-of-home.

Cost per view — ad spend divided by the number of video views. The pricing unit for video and connected-TV campaigns bought on a view basis.

The average number of days it takes the agency to get paid after invoicing — a direct measure of cash-flow health.

The cost per 1,000 impressions for digital out-of-home — programmatic billboards and screens — where impressions are modelled from plays, screens, and audience data.

The actual revenue earned per hour of billable work — total fees divided by hours worked — after discounts, overruns, and write-offs.

The return on investment from an email programme — revenue generated against the cost of running it. Consistently one of the highest-ROI channels in marketing.

Numeric vectors that represent the meaning of a piece of text, image, or other content — used so a computer can compare two things by semantic similarity, not just keyword overlap.

The share of an audience that interacts with content — engagements divided by reach, followers, or impressions. The core measure of organic and influencer performance.

Continuing to train a base model on a smaller, curated dataset so it adapts to a specific style, format, or domain.

Revenue minus the direct cost of delivering the work — primarily billable salaries — expressed as a percentage of revenue.

When a language model generates plausible-sounding text that is factually wrong — a made-up citation, a fabricated number, a confidently invented detail.

A description of the type of customer most likely to buy, stay, and refer — usually defined by industry, size, role, and a few qualifying behaviours.

How fees for influencers and key opinion leaders are set — benchmarked against reach (CPM) and engagement (cost per engagement) rather than follower count alone.

A neural network trained on huge volumes of text to predict the next token, used today as the engine behind chat, agents, and most generative-AI features.

Customer lifetime value — the gross profit a single customer is expected to generate over the entire time they stay with a business.

An open standard for connecting AI applications to tools, data sources, and systems — letting one model talk to many tools without bespoke integrations.

Marketing-qualified lead — a lead that has crossed a defined engagement threshold, signalling enough intent for sales to follow up.

The deliberate choice of how a product wants to be understood — by whom, against which alternatives, and on what unique value.

The craft of writing the instructions you give a language model — choosing wording, examples, structure, and constraints to get useful, repeatable output.

A scannable 2D barcode that links the physical world to a URL — bridging out-of-home, print, packaging, and events into trackable digital journeys.

The percentage of billable time that actually gets billed and paid — the leakage between hours worked and hours invoiced.

An ongoing engagement priced on a recurring basis — usually monthly — in exchange for a defined amount of work or capacity.

A pattern that pulls relevant documents into the prompt at inference time so a model can answer questions about information it was never trained on.

Return on ad spend — the revenue generated divided by the advertising spend used to generate it, within a fixed attribution window.

Net profit as a percentage of the cost that produced it — (revenue − cost) ÷ cost. The all-in profitability test for any spend.

The gradual expansion of work beyond what the contract priced — extra rounds, new deliverables, "while you are at it" asks — usually unbilled.

A brand’s share of total category presence — mentions, spend, or impressions — against all competitors. A leading indicator of market share.

Sales-qualified lead — an MQL that a sales rep has actively qualified and accepted as worth pursuing, usually after a discovery call.

Tool use is how an AI model does more than talk. It calls external functions, APIs, and software to fetch data and take actions, then uses the results to produce real work rather than just text.

Tags appended to a URL — source, medium, campaign, term, content — that tell analytics tools where a visit came from. The backbone of campaign attribution.