Glossary

Every term used in optimization groups, in one place.

The words that appear in the groups table, the calendar, the Timeline, the automation panel and the allocation view.

Metrics

ACOS (advertising cost of sale) — ad spend ÷ ad revenue, as a percentage. ROAS inverted: 20% ACOS = 5× ROAS. A budget rule, not a profit rule.

Platform ROAS — ad revenue ÷ ad spend, as the platform reports it. Treats GMV as if you kept all of it.

Contribution ROAS — revenue after platform fees, COGS, and shipping ÷ ad spend. Break-even is 1×.

Profit-adjusted ROAS (PA-ROAS) — the same number as True ROAS: net profit after ad spend ÷ ad spend. Break-even is 0, not 1.

Contribution profit — what is left of a sale after platform fees, COGS, and shipping, before ad spend.

Profit after ads — contribution profit once the ad cost is paid.

Profit margin — contribution profit as a share of revenue.

GMV — gross merchandise value: the total order value before any deduction.

Marginal CPA (mCPA) — what one more order costs at the current spend level, read off the model's spend-response curve. Shown on a campaign's detail page, with the contribution per order it has to beat to be worth buying.

Budget

Group Monthly Budget — the most the group may spend across the whole month. Unset means the model sizes the month.

Group Daily Budget — the amount, or ceiling, for an ordinary day. Unset means the model sizes each day.

Budget mode — whether an amount is a ceiling ("spend at most", the day may end under it) or a target ("spend it all", the full amount is allocated even where the last of it earns less).

Model-set / model-managed — a budget layer you left unset, so the model chooses the amount.

Projected — a calendar amount that is a forecast rather than a commitment, because no Group Monthly Budget is set.

Planner sized — a day whose amount the month planner chose, rather than one you set.

Day lock — a date that spends exactly this amount, replacing Group Daily Budget for that date.

Day ceiling — a date that may spend up to this amount and less if the model does not think the rest earns its place.

Extra demand (uplift) — how much busier you know a date will be than history suggests, 0–500%. It moves budget between days; it never raises the month's amount.

Campaign Daily Budget — an optional per-campaign ceiling, so one member cannot take too large a share. Not applied while the group spends to a target.

Unallocated cap — group budget the solve deliberately left unspent, usually because funding another campaign would break the constraint.

Uncertain reserve — budget held back for campaigns whose creation on the platform is not yet confirmed, so the group cannot overspend if they turn out to exist.

Policy

Objective — what the split maximizes: profit, GMV, or orders.

Constraint — the minimum performance a campaign must reach to receive budget.

Distribution — an optional tilt of the split toward recent results, defined by a basis, an influence, and a time window. Formerly called allocation preference.

Influence — how much modelled improvement you trade for that tilt: Small keeps ≥90%, Medium ≥70%, Large ≥40%.

Time window — how far back the distribution basis measures: rolling 3 hours, rolling 24 hours, or last 7 days on a live group.

Growth Exploration — how actively DataGlass spends a little differently from the best-known allocation so the model keeps learning. Off, Low (±20%, the default), or Moderate.

Target ROAS — the return each campaign bids toward on the platform. Carried by a date, not by the group.

The schedule

Change point — a time from which the group runs different settings, holding until the next one. Six a day, at least 30 minutes apart, and every day starts with one at 00:00.

Schedule profile — either the default schedule, which runs on any date without one of its own, or a dated profile, which runs instead of the default on that date.

Running day-to-date ceiling — what a scheduled Group Daily Budget is: the total the group may have spent by that time, so it can only rise as the day goes on.

Automation

Rule — an IF/THEN that watches a metric on a schedule and changes a setting.

Cooldown — hours to wait after a change before the rule may act again.

Consecutive matches — how many runs in a row must match before anything changes.

Dry run / replay — what the rule would have done over past days, and what it would do right now, shown before you activate it.

Conflict — two active rules writing the same setting. They stack: the second computes from the value the first just wrote.

Runs

Run — one solve for one group, on one target date.

Trigger — why the run happened: Scheduled (nightly), Saved (you saved settings), Model ready (training finished), or Preview.

Solve status — Solved, Solving, Infeasible, Failed, or Superseded.

Push status — Applied, Partial, Uncertain, Pending, Applying, Failed, or Not pushed.

Allocation — the per-campaign daily budget a run produced.

Allocation stage — whether a campaign's share came from mature evidence, a still-learning model, or cold start.

Constraint readiness — whether there is enough data to evaluate the constraint yet. A brand-new campaign is not evaluable until it has history.

Funding — whether a campaign received budget this run, or was left unfunded and paused for the day.

Cold start — the fallback for a campaign with no performance history.

Stranded — allocation the model gave to campaigns that are paused on the platform, so it cannot reach them.

Model

Campaign model ready — the group allocates on its own campaign-level model.

Product learning ready — the group's campaigns are pooled at the product level; the stronger state where members share products.

Waiting for ads data — not enough delivery yet to fit anything.

More varied data needed — there is data, but every campaign has only spent at one level, so the model cannot say what a different level would return.

Confidence — how reliable a campaign's evidence is. Drawn as a band; wider means thinner evidence.