REFERENCE / USER ACQUISITION
Metrics, methods and working terms for connecting acquisition decisions to business outcomes.
Definitions depend on the source and decision. Entries marked “Working term” describe Ascentframe usage; agree the exact definition in your project.
A comparison in which eligible units are assigned to alternative versions of one intervention.
Example Assign visitors to two checkout versions and compare completed purchases.
Keep separate A before-and-after comparison is not automatically an A/B test.
The sequence from first contact to a defined customer outcome, with eligible populations and events for each transition.
Example Ad click → signup → payment → app activation.
Keep separate A missing event can look like a missing customer.
Events grouped by when they happened, regardless of when the user was acquired.
Example All purchases made in September, including purchases by older customers.
Keep separate Install-cohort reporting groups the same events by the acquisition date instead.
A platform-specific group of ads sharing delivery settings such as audience, placement or budget. Other platforms may use the term ad group.
Keep separate An ad set can contain several creatives; its aggregate result does not isolate each creative's effect.
A notification that a declared condition has been met or a check has failed.
Keep separate Several alerts can belong to one incident. An alert is neither a diagnosis nor proof that an action succeeded.
A preserved view of the data and settings available at a specific decision time.
Example Keep the information known when a creative was stopped, before later purchases arrived.
Keep separate Using later information to judge an earlier decision introduces hindsight.
The entity assigned to a treatment or control condition: for example, a user or a geographic market.
Keep separate The assignment unit can differ from the unit recorded in the dataset.
Rules that assign credit for an observed outcome to acquisition or marketing interactions.
Example A purchase receives credit under a declared click or view attribution rule.
Keep separate Credit does not establish how many purchases advertising caused.
The allowed interval between an interaction and an outcome for that interaction to receive credit.
Keep separate Changing the window can change reported performance without changing purchases.
Adding or correcting records for an earlier reporting period.
Example A late renewal updates the revenue of its original install cohort.
Keep separate Keep an as-of view when later corrections could change a past decision.
Repeatedly resampling observed units to estimate the variability of a statistic.
Keep separate Resampling must preserve relevant dependence; it does not remove confounding.
Customer acquisition cost
The defined cost of acquiring a new customer divided by the number of new customers acquired.
Keep separate State which sales, marketing and production costs are included. CAC is not necessarily media-only CPA.
A platform's organizing level for an advertising objective and its associated delivery configuration.
Keep separate Platform hierarchies differ; confirm where budgets and rules actually apply.
When a new activity takes outcomes or delivery from an existing activity instead of adding equivalent value to the whole portfolio.
Example A new launch gains attributed sales while the existing campaigns lose them.
Keep separate That pattern is a hypothesis, not proof, of cannibalization.
The design and assumptions that make it possible to distinguish an intervention's effect from other explanations.
Keep separate More data or a better model fit cannot repair an unidentified causal question by themselves.
CTI
Installs divided by eligible clicks under a declared attribution rule and observation window.
Example 100 attributed installs from 1,000 eligible clicks gives a 10% rate.
Keep separate Do not mix incompatible click and install populations.
A group sharing a defined starting event or characteristic, compared over an explicit observation horizon.
Example Users who installed in the same week, observed through day 7.
An interval produced by a method with a stated long-run coverage rate under its assumptions.
Keep separate A 95% confidence interval is not a 95% probability that a fixed true value lies in this particular interval.
Revenue less the explicitly included variable costs for a defined unit and period.
Keep separate State whether acquisition spend is included. Contribution is not automatically cash received or accounting profit.
The delay between an acquisition interaction and the outcome being measured.
Keep separate A purchase delay and a reporting delay are different clocks.
CVR
Defined conversions divided by the eligible interactions or users that could convert.
Keep separate Name both the event and denominator: click-to-purchase and signup-to-payment answer different questions.
The outcome that would have occurred under an alternative action or without the intervention.
Keep separate For the same unit at the same time, that alternative outcome cannot also be directly observed.
Cost per action
Advertising spend divided by the number of specified actions.
Example $600 spend and 30 purchases gives a purchase CPA of $20.
Keep separate Always identify the action and its counting rules.
Cost per click
Advertising spend divided by recorded clicks for the same scope.
Keep separate Average CPC differs from a maximum CPC bid.
Cost per install
Advertising spend divided by attributed installs for the same scope and reporting rules.
Keep separate Cheap installs need not produce valuable customers.
Cost per thousand impressions
Advertising spend divided by impressions, multiplied by 1,000.
Keep separate Viewable CPM uses viewable impressions; keep that denominator separate.
A declared grouping of assets that share a concept, production approach or other learning-relevant feature.
Example Group variants of one customer problem and promise before comparing production output.
Keep separate The grouping needs a reproducible taxonomy; there is no universal family definition.
A hypothesized decline in an ad's effectiveness as audiences become less responsive to it.
Keep separate A weaker week alone cannot distinguish fatigue from delivery, audience mix or product changes.
An identifier used to connect an asset's tests, retests and production launches.
Keep separate Specify whether an edit creates a new identity and how platform copies map to the original asset.
Click-through rate
Recorded clicks divided by impressions, usually expressed as a percentage.
Keep separate A strong CTR does not establish purchase quality or profitability.
How recently a source or report was updated relative to the expected arrival of records.
Keep separate Fresh data can still describe immature outcomes.
The extent to which the agreed outcome window and reporting delay have elapsed for an observation.
Example Compare D7 outcomes only after both cohorts have reached D7 and the declared reporting cutoff.
The entity to which an action or business verdict applies.
Example Pause an ad set, retest a creative, or change a channel budget.
Keep separate One dataset can support several decision units, but they need different comparisons.
Recognizing repeated representations of the same event or entity and applying explicit counting rules.
Keep separate Two events with similar values are not necessarily duplicates.
The advertising exposure and spend that actually occurred, rather than the budget or settings requested.
Example A configured budget of $500 does not establish that $500 was spent.
DiD
Comparing the change in treated units with the change in comparison units.
Keep separate A causal interpretation requires assumptions including appropriate parallel trends; matching alone does not establish them.
An analysis of outcomes around a defined event, using explicit relative-time windows and comparison choices.
Example Trace delivery before and after a budget increase.
Keep separate A visible change after an event does not by itself prove the event caused it.
The time when an event occurred, as recorded by its source.
Keep separate It can differ from the time the event arrived in a report and from the acquisition date used for cohort reporting.
Allocating resources to combinations already supported by evidence, while continuing to check that conditions remain valid.
Example Give a confirmed combination additional production budget.
Spending or testing to learn about candidates whose value is not yet sufficiently established.
Example Test a new concept or an unproven creative–setup combination.
The entity or opportunity that actually receives the intervention or advertising exposure.
Keep separate Assignment does not guarantee exposure; verify what was delivered.
Classifying a target condition as absent when it is present under the declared truth definition.
Example Reject a candidate that meets the pre-defined success criterion under an appropriate evaluation.
Keep separate Later recovery alone does not prove an earlier verdict was wrong under the information available then.
Classifying a target condition as present when it is absent under the declared truth definition.
Example Raise an incident alert for an observation that does not meet the accepted incident definition.
The level of detail represented by one record or analytical result.
Example Campaign × day differs from creative × launch attempt × cohort age.
Keep separate Joining datasets at incompatible grains can duplicate spend or outcomes.
The relative weights of segments contributing to an aggregate result.
Example Overall CPA changes when more budget reaches higher-cost markets, even if each market's CPA is stable.
Keep separate Separate mix changes from changes within groups.
A group withheld from an intervention to provide a comparison under an explicit design.
Keep separate Separately, a model-validation holdout is data reserved for evaluation; it is not automatically an experimental control.
Connecting records believed to describe the same user or entity under declared matching rules.
Keep separate Keep uncertain or unavailable links explicit rather than forcing every record into a match.
A material deviation requiring a bounded investigation or operating response in an agreed environment.
Keep separate An incident groups relevant evidence; it does not imply one known cause.
Ascentframe's structured record of an incident: scope, time window, signals, business exposure, owner and next check.
Keep separate It is a working output for review and routing, not an automatic root-cause verdict.
The additional outcome attributable to an intervention relative to the outcome without it.
Keep separate An attributed purchase can exist without an incremental purchase.
Users grouped by their installation date, with subsequent outcomes assigned to that cohort under declared attribution rules.
Keep separate September cohort revenue need not equal purchases that happened in September.
Incremental return on ad spend
Incremental revenue divided by the advertising spend associated with the defined intervention.
Keep separate Declare whether spend means total treatment spend or an added spend tranche, and define the outcome window.
One creative launched in a particular ad set or campaign setup at a recorded time.
Example Creative ID × ad set × setup × launch timestamp.
Keep separate Several attempts may belong to one creative; future creative value must not be assigned in full to every attempt.
The sequence of declared states through which an entity moves over time.
Example For a creative: test → retest → promotion → production → stop. For a customer: signup → activation → renewal.
Keep separate Name the entity; creative and customer lifecycles are different systems.
Lifetime value · CLV
The defined revenue or contribution associated with a customer over a stated lifetime or horizon.
Keep separate Separate observed cohort value from predicted future value; specify revenue versus margin.
A human action that replaces or changes the state expected from an automated rule.
Example Manually resume an ad set stopped by automation.
Keep separate Record the actor, time and reason so the override does not hide a failing rule.
mROI
The additional outcome associated with a small additional spend increment, divided by that increment.
Keep separate It answers a next-spend question, not the same question as average historical return.
Selecting comparison units with similar observed characteristics relevant to the question.
Keep separate Similarity on recorded fields does not remove all unobserved confounding.
Marketing mix modeling
Modeling how business outcomes relate to marketing activity and other factors, usually using aggregated data.
Keep separate Prediction quality alone does not establish causal validity or the safety of a budget move.
Revenue after the deductions included in an explicit reporting contract.
Keep separate Refunds, store fees and taxes may be treated differently across sources; specify the deductions rather than relying on the label.
The result a decision or intervention is intended to protect or change, with an explicit unit and time window.
Example Activated paying customers through day 7, rather than clicks alone.
A project-specific acquisition-economics label used in some analyses. Its formula and cost coverage must be agreed before use.
Keep separate Ascentframe does not assign a universal expansion or formula to PAM. Confirm revenue, refunds, fees, spend and horizon in the project contract.
The time required for the defined customer contribution to recover acquisition cost.
Keep separate State the cost base and contribution basis; forecast payback differs from observed payback.
Moving a candidate to a declared next stage after it meets the applicable evidence and risk criteria.
Keep separate Qualifying for a stage and actually receiving the next launch or budget are separate events.
Assigning eligible units to conditions through a declared chance mechanism.
Keep separate It supports a valid comparison but does not guarantee exposure, complete measurement or perfect balance in one sample.
A metric calculated from total numerator divided by total denominator across the selected scope.
Example Portfolio CPA = total spend / total purchases, not the unweighted average of ad-set CPAs.
Keep separate A zero denominator needs an explicit undefined-value rule.
A modeled relationship between a defined spend level and the expected outcome, holding declared scenario assumptions in place.
Keep separate A smooth curve can extend beyond the spend range supported by observations.
The pre-agreed evidence and authority required to reverse a change or restore an earlier operating state.
Example Specify what would restore budget after a bounded reduction, who acts and when.
The share of a defined cohort that remains active or returns under a specified event and horizon.
Keep separate App activity, continued subscription and successful renewal are different retention definitions.
A new attempt to evaluate a previously tested candidate under recorded conditions.
Keep separate A retest may change the setup or timing, so it need not replicate the original test.
Return on ad spend
The defined conversion value or attributed revenue divided by advertising spend.
Keep separate ROAS is neither contribution profit nor automatically incremental return.
Return on investment
A return measure whose numerator and investment base must be stated. Financial and modeling systems use different conventions.
Example One financial convention is (revenue − included costs) / included costs.
Keep separate Some models label incremental outcome / spend as ROI. Never compare these without reconciling definitions.
An identifiable specification of a rule's conditions, inputs, schedule, allowed action and target population.
Keep separate Record which version evaluated each object; a new announcement does not prove every object uses the new logic.
A range in which additional spend produces diminishing additional outcomes under the defined context.
Keep separate It differs from creative fatigue, and diminishing return does not automatically mean an unprofitable next step.
A shared specification of a metric's event, counting rule, grain, time basis, source, exclusions and correction policy.
Example Define exactly which subscription events count as purchases before joining reports.
Checking whether a conclusion changes under plausible alternative assumptions or analytical choices.
Example Vary maturity cutoffs or matching tolerances and inspect the decision, not only the number.
Keep separate Stability under chosen alternatives does not rule out every source of bias.
The declared combination of campaign structure, targeting, bids, placements and other launch settings.
Keep separate The intended setup and the configuration actually applied need separate records.
The agreed authority for a specific fact or metric in a defined context.
Example The advertising platform for billed spend; a transaction ledger for completed payments.
Keep separate One system need not be authoritative for every part of the customer journey.
A measure of how consistently the ranks of two variables move together.
Keep separate It measures monotonic association, not a causal effect or a ready-made decision rule.
An intervention affecting units outside its intended treatment group, including the comparison group.
Keep separate Audience overlap and competing campaigns can change the interpretation of a test.
In group comparisons, applying a common set of segment weights to separate composition differences from within-segment performance.
Keep separate Here this means composition standardization, not converting values into z-scores.
The probability that a specified test detects a declared alternative effect under its assumptions.
Keep separate Low power can leave a useful effect unresolved; a non-significant result is not proof of no effect.
A declared condition for pausing an activity or ending a test, with an observation window and action owner.
Keep separate Safety stops, economic stops and statistical stopping rules have different purposes.
Revenue, costs and contribution measured per declared useful unit or outcome.
Example Cost per profitable scaled winner includes failed search attempts as well as successful launches.
UA
The work of bringing new users or customers to a product and evaluating the value and cost of doing so.
Keep separate Install volume, activated users and retained paying customers are different outcomes.
W2A
A journey in which a user moves from a web acquisition or purchase flow into an app.
Keep separate Keep the website, payment, installation and app-activation transitions distinct when reconciling the journey.
A candidate that meets a declared success criterion in a recorded context and observation horizon.
Keep separate A winner label does not guarantee transferability, profitable scale or permanent performance.
