Glossary

This glossary defines product-specific terms used across the Skeleton Key docs and app. Where a term has a fuller treatment elsewhere, the entry links to the canonical page once.

Adstock

How long a dollar of media spend keeps influencing your KPI after the period it ran in. TV usually has long adstock; paid search fades fast. See Read your results.

Baseline contribution

In a Marketing Mix Model's contribution breakdown, the share of your KPI credited to organic demand: what you'd have earned with no media at all. Usually the largest slice, and a baseline of 60 to 80 percent is normal. Distinct from the baseline-plausibility health check.

Baseline plausibility

A model-health check that asks whether the model's organic-KPI estimate (what you'd sell with zero media) stays sensible, measured by how likely it is to go negative. A high chance of going negative means media is being over-credited.

Bayesian inference

A way of modeling that gives a range of likely answers instead of one number. You start from what you already believe (priors) and the model balances that against what your data actually shows.

Brand

A top-level workspace inside an organization that holds Campaigns and a Brand Profile. The brand selector in the sidebar filters every page to one brand's content. See Core concepts for the containment model.

Brand kit

A saved set of colors, heading and body fonts, an optional logo, and a short visual-personality note used by newly generated Slides decks. An organization can have one shared kit, and each brand can have its own. See Design & brand kits.

Brand Map

A brand-rooted knowledge graph showing how insights, strategies, tactics, and metrics connect across phases. The map is filtered by campaign at view time. See Brand Maps.

Brand Map edge

A directional link between two elements in a Brand Map. Each edge has a confidence score, and the UI hides edges below 0.75 by default; you can adjust the threshold with the slider. Skip-level edges (a research insight feeding a Media Plan tactic with no Strategy element between) carry a confidence penalty.

Brand Map element

A single node in a Brand Map. Elements are extracted from agent outputs by phase and come in four types: insight (circle), strategy (diamond), tactic (square), and metric (triangle).

Brand Profile

The AI-generated profile of a brand's positioning, audience, values, and tone. A completed Brand Profile is required before any Campaign can be created inside a brand. Generation takes about 2 to 3 minutes. See Brand.

Brand role

A user's permission level on one specific Brand: Admin (manages collaborators and runs all workflows), Standard (runs all workflows), or Viewer (read-only). Brand role is distinct from Org role. See Brand permissions.

Campaign

A workspace inside a Brand for one specific marketing initiative. Each campaign has its own brief and its own Research, Strategy, and Media Plan outputs. See Campaigns.

Chat

The AI assistant grounded in your Brand Profile, Research, Strategy, and Media Plan outputs. It lives at /chat and as an embedded panel on the four output pages. Chat does not search the web. See Chat.

Citation

A link from an output back to its source. There are three types: Brand document (a file you uploaded), Skeleton Key output (an upstream run or the Brand Profile), and External web (a public URL). See Citations.

Contribution

The share of your KPI a Marketing Mix Model credits to a channel, beyond what the baseline would have delivered anyway. A function of both how much you spent and how efficient the channel was, so it isn't the same as ROI.

Control variable

A continuous non-marketing factor a Marketing Mix Model adjusts for in the background, like price, seasonality, or weather, so its effect isn't miscredited to a channel. Contrast with a treatment. See Prepare your data.

Convergence

Whether a Marketing Mix Model's independent sampling runs agreed on the same answer, measured by R-hat. If they didn't converge, the estimates aren't stable and the run shouldn't be trusted.

CPIK (cost per incremental KPI)

How many dollars of spend it takes to drive one extra unit of your KPI. Lower is better. The complement to ROI, and Meridian computes it for any KPI type, not just counts.

Credible interval

A range the true value is likely to fall within, given the data. A 90% credible interval is one the model is 90% confident contains the true value. The ROI ranges on the results page are credible intervals: wider means less certainty.

Diminishing returns

The pattern where each extra dollar of spend on a channel returns less than the last as you spend more, shown as an S-curve (Hill saturation) on the response curves. See Read your results.

Documents

The per-brand file repository for PDFs, Word, Markdown, CSV, and Excel files. Files are uploaded once and selected per run as opt-in context, with a 10 MB cap per file. See Documents.

ESS (effective sample size)

Roughly how many independent draws went into each Marketing Mix Model estimate. Higher is better. Read it alongside R-hat as a convergence signal.

Exploratory Data Analysis

The pass a Marketing Mix Model runs over your mapped data before fitting, returning a verdict of good, usable with caveats, or issues found. It flags data adequacy and problems but warns rather than blocks. See Prepare your data.

Incrementality

The lift a channel actually caused, beyond what would have happened anyway. A Marketing Mix Model estimates incrementality from aggregate data, which is a different question from attribution's user-journey credit. An incrementality test measures it directly for one channel or tactic.

Knots

Flex points in a Marketing Mix Model's baseline trend. More knots let the baseline bend more over time, which credits less of your KPI movement to media. Meridian recommends one knot for a single-region model.

Lineage

The workflow chain visible in the Brand Maps sidebar, with one chain per Research, Strategy, or Media Plan grouping inside a campaign. The Lineage list shows an edge-count badge that updates as you move the confidence slider.

MAPE (mean absolute percentage error)

On average, how far off (in percent) a Marketing Mix Model's per-period prediction was from reality. Lower is better. As a rough industry guide, 10 to 20 percent is typical for this kind of model.

Marginal ROI

What an extra dollar of spend would return at your current investment level. Different from average ROI, which is the return across all of a channel's spend. Marginal ROI is the figure that informs where to move budget. See Read your results.

Marketing Mix Model

A model that measures how much each marketing channel contributed to an outcome, using aggregate historical spend and results rather than individual user journeys. Skeleton Key runs Google Meridian. Also called a media mix model. See Marketing Mix Model.

Media Plan

The output of the Planning phase. A single unified output organized into three on-page sections: Overview, Charts, and Plan Details. See Media Plan.

Mediator

A variable that sits between your marketing and your KPI, like site traffic on the path from ads to sales. Including it in a Marketing Mix Model steals credit from the channels that caused it, so leave mediators out of the model. See Prepare your data.

Optimal frequency

For reach and frequency channels, the weekly average frequency that returns the most per impression. Past it, you pay to show the same people an ad more often than helps. See Read your results.

Organic media

An unpaid driver of your KPI, like email sends or organic social. In a Marketing Mix Model it's mapped without spend, so its contribution shows up without being treated as paid investment.

Organization

The top-level container, also called an Org. Every user belongs to at least one. Owners and Admins can create Brands inside their organization and manage who has access. See Orgs and roles.

Org role

A user's permission level on one organization: Owner (full control, one per org), Admin (everything except deletion or ownership transfer), Member (use workflows on accessible brands), or Guest (view-only access to specific brands).

Phase

One of the workflow phases that produces a core output: Research, Strategy, and Planning. Inside a campaign, each phase requires the previous to be complete before it becomes available. Brand Profile and Campaign are setup, not phases. See Core concepts.

Population scaling

In a multi-region Marketing Mix Model, using each region's population so regions of different sizes are compared fairly. This is why a geo column needs a population column alongside it.

Posterior predictive p-value

A model-health check (also shown as fit plausibility) for whether your actual total KPI looks like a plausible result from the model's own simulations. A very low value means your data is an extreme outlier versus what the model predicts.

Prior

What you tell a Marketing Mix Model to expect before it sees your data, drawn from past incrementality results, industry benchmarks, previous models, or your team's expertise. The model balances priors against what the data shows.

Prior-posterior shift

A model-health check for whether the data was strong enough to update a channel's prior. It flags channels where the prior, not the data, drove the answer, which is a sign to loosen the prior or gather more data.

R-hat

A stability score per parameter in a Marketing Mix Model. Below 1.2 means the sampling runs converged; higher means the run didn't fully settle and should be re-run with more draws.

R-squared

The share of your KPI's period-to-period variation a Marketing Mix Model explains (also called goodness of fit). Closer to 1 is a tighter fit. Read it as a relative-comparison signal, not a pass/fail line.

Reach and frequency

Channels where you have true reach (unique people reached) and frequency (times each person saw the ad) data, like YouTube. Modeling them lets Meridian find an optimal frequency instead of treating every impression alike.

Refine

Iterating on an output by giving the agent feedback. Refining produces a new run that builds on the previous one, and refining an upstream phase marks downstream phases as stale. See Refine.

Rerun

Generating a phase from scratch with the same inputs. A rerun produces a new run, and the previous run stays in version history.

Research

A workflow phase. Five analyst agents (Category, Competitor, Consumer, Culture, and Social) run in parallel; an optional pair of synthesis reports can be generated from their outputs. The analyst run takes about 3 to 4 minutes; the reports add another 4 to 5 minutes if you generate them. See Research.

Restore

Promoting an earlier run from version history to current. Restore produces its own new run that captures the restore event; the version you restored from stays in history.

ROI consistency

A model-health check for whether each channel's learned ROI stays within the range your custom priors imply. It only runs when you've set custom ROI priors.

Run

One generation event on a Brand or Campaign, also called a workflow run. A run can be an initial generation, a Refine, a Rerun, a Restore, or an edit. Version history is the chain of runs on a phase.

Skeleton Key

The product itself: an AI-powered marketing intelligence platform that automates a large share of desk research and first-draft writing for marketing teams.

Slides

A presentation-ready deck built from one finished output. You pick a Research, Strategy, or Brand output, choose an available style, template, and length, and Skeleton Key composes the slides. The deck opens in a viewer and downloads as a self-contained HTML file or prints to PDF. Each generation is its own deck. See Slides.

Staleness indicator

A UI flag on a downstream phase when the upstream phase has been refined or rerun, signalling that the downstream output no longer reflects the latest upstream version. See Core concepts.

Strategist

One of the four agents that run during a Strategy phase: Goals, Audience, Communication, and Media. They run in sequence rather than in parallel.

Strategy

A workflow phase. Four strategists run in sequence (Goals, Audience, Communication, Media) followed by a Strategy Summary. The full run takes about 10 to 13 minutes. See Strategy.

Strategy Summary

The synthesis document produced after all four strategists complete. It distills the four strategist outputs into one report.

Time period

The cadence of your Marketing Mix Model data, one row per period. On a weekly file a period is a week, and time-based reads like adstock are expressed in those periods.

Treatment

A discrete non-marketing event whose effect you want a Marketing Mix Model to estimate, often a 0/1 flag like a holiday or promo. Contrast with a control, which is a continuous background condition. See Prepare your data.

VIF (variance inflation factor)

How much two or more of your columns overlap in a Marketing Mix Model. High VIF means the model can't tell their effects apart, so credit gets split between them arbitrarily.

wMAPE (weighted MAPE)

Like MAPE, but bigger periods count more, which is more representative when your KPI has seasonal spikes. The preferred headline fit signal in a Marketing Mix Model.