Adobe Target Architect Master cheat sheet

at.js and Web SDK run in the browser and request offers from the Target edge; server-side decisions use the Delivery API or Adobe Target Node/Python SDKs.

The Target edge returns experiences plus a decision token; the client renders and fires the display event, which is required for impressions and reporting.

mbox names, location names, and offer content are the unit of decisioning; global mbox replaces the legacy target-global-mbox and can host auto-allocate and auto-target activities.

On-device decisioning (ODD) pre-fetches the rules artifact so decisions are made locally; it trades real-time audience freshness for latency and availability.

A/B and XT use mutually exclusive visitors per experience; MVT partitions by location combinations; AP and Auto-Target use a shared model per activity.

Activity QA links use the qa parameter with a token; they bypass audience qualification but still log to reporting unless excluded.

Implementation order: data layer and audiences, then mbox/location instrumentation, then activities, then reporting validation.

Server-side implementations must pass the customer ID and session context so profile and A4T stitching work across channels.

A/B test: 2+ experiences, traffic split by percentage, one success metric; winner selection is manual or automatic after the test ends.

XT (Experience Targeting): 1+ experiences each bound to an audience rule; the first matching audience wins, so order matters.

MVT: evaluates combinations of changes across locations; full factorial tests all combinations, partial factorial tests a subset.

Auto-Target and Automated Personalization use a contextual bandit model; AP can personalize per visitor across offers, Auto-Target picks among experiences.

Recommendations criteria include popularity, recently viewed, user-based, item-based, and custom; each criterion has a design and a backup recommendation.

Offer decisioning priority: activity priority (0-999, higher wins) then the most specific audience; ties resolve by activity creation order.

Goals: conversion (mbox/click), revenue, engagement; a metric can be primary or secondary, and A4T metrics are defined in Analytics.

Form-based composer edits offers as JSON/HTML; Visual Experience Composer modifies the DOM and requires stable selectors.

Audience rules combine geo, device, browser, custom parameters, profile attributes, and Adobe audiences; AND/OR logic determines qualification.

Custom audiences are evaluated server-side on the Target edge; profile scripts compute derived attributes at profile update time.

Audience Manager segments reach Target through the AAM-Target integration; the visitor's segment membership is read at decision time.

AEP audiences are shared to Target via the Experience Platform destination; streaming segments update near real-time, batch segments on schedule.

Profile attributes persist up to the profile lifetime (default 14 days of inactivity) and are keyed by the Target visitor ID or third-party ID.

Exclude internal traffic with IP-based or QA audiences; audience size estimates in the composer are approximate and cached.

Personalization strategy layers: segment-level (audiences), behavior-level (profile scripts), and model-level (Auto-Target/AP).

A4T links Target activities to Analytics report suites; the Analytics for Target panel shows lift, confidence, and segment breakdowns.

Statistical significance uses a two-tailed test at 90% or 95% confidence; the report shows confidence interval and lift range.

Sample size depends on baseline conversion rate, minimum detectable effect, power (typically 80%), and significance level.

Reporting latency: Target reports update near real-time; A4T data flows through Analytics and can lag by hours.

Auto-Allocate shifts traffic to the best experience; Auto-Target personalizes per visitor using the model.

Metrics: conversion rate = conversions / visitors; revenue per visitor = revenue / visitors; RPV is the default revenue metric.

Exclude outliers and bot traffic; the report's visitor count differs from Analytics visits due to counting method.

A4T requires the Analytics tracking server and the Target-Experience Cloud org link; the visitor ID service stitches identities.

Audience Manager integration: the AAM DIL library or Web SDK forwards segment data; Target reads segments as audiences.

AEP integration: the Web SDK sends events to the Edge Network; Target decisions are returned in the same response.

Adobe Campaign can trigger Target offers via the Delivery API or use Target as the offer source in a campaign.

Third-party integrations use the Delivery API with an OAuth server-to-server credential; the client ID and secret are scoped to the org.

Experience Cloud ID (ECID) is the shared identity; Target visitor IDs are derived from it when the ID service is present.

at.js 2.x supports single-page apps via applyOffer and the getOffers API; the legacy mbox.js is deprecated.

Web SDK (alloy) uses sendEvent with a decision scopes array; Target returns propositions rendered by the client.

The Delivery API endpoint is /rest/v1/delivery; requests include context, execute/pageLoad, and prefetch.

Mobile: the Adobe Experience Platform Mobile SDK with the Target extension; offline mode uses cached offers and queues events.

Server-side decisions must send the customer ID and session ID; the display event is required for impressions.

Response tokens expose attributes (e.g., activity.id, experience.id) for use in analytics and personalization.

Consent: the opt-in service gates Target calls; without consent, no profile or audience data is stored.

Data retention: Target profile data expires after the configured profile lifetime; deletion requests use the Privacy API.

PII must not be sent in mbox parameters or custom parameters; use hashed identifiers and the customer ID.

The Privacy API (GDPR/CCPA) supports access and delete requests scoped to the org and visitor ID.

Security: TLS in transit, encryption at rest, role-based access via Adobe Admin Console product profiles.

Data governance labels in AEP control which attributes can be used for personalization.

Product profiles in Admin Console grant roles: Approver, Editor, Observer; approvals gate activity activation.

Activity QA: use the QA link with the qa parameter; the token is activity-specific and expires.

Troubleshooting order: check the mbox request in the browser network tab, the response offers, then the audience qualification.

Common failures: missing display event (no impressions), audience mismatch (no activity), and offer rendering errors (selector drift).

The Adobe Experience Cloud debugger and the Target trace (mboxTrace) show decision details and audience evaluation.

Activity collisions: higher priority wins; use exclusions to prevent overlapping activities on the same location.

Decide client-side vs server-side vs hybrid before instrumentation; the choice determines which SDK and API are used.

The global mbox is required for Auto-Target, Auto-Allocate, and Recommendations on page load.

Plan mbox/location naming conventions and a data layer contract; consistent names enable reusable audiences and reports.

ODD requires the rules artifact to be downloaded and refreshed; stale rules may miss audience changes.

Capacity planning: the edge handles high QPS; server-side calls count against the org's API limits.

QA plan: validate offers in staging, test audience qualification with QA links, and confirm A4T stitching.

Traffic allocation: equal split for A/B; Auto-Allocate starts equal and shifts based on performance.

Success metrics are defined per activity; A4T metrics must exist in the linked Analytics report suite.

Offer content can be HTML, JSON, or a redirect; redirect offers require a URL and are not compatible with all activity types.

Recommendations backup: if the primary criterion returns no items, the backup criterion supplies offers.

Activity duration: run for at least one full business cycle; avoid stopping early on peeking.

Exclusions prevent a visitor from entering multiple activities on the same location.

Audience qualification is evaluated at decision time; profile attributes are read from the visitor profile.

Custom parameters (mbox parameters) are available only for the current request unless promoted to profile attributes.

Geo targeting uses the edge's IP geolocation; accuracy varies by region and VPN usage.

AEP streaming segments can qualify visitors within minutes; batch segments update on the schedule.

First-party IDs improve cross-device stitching when the customer ID is passed.

Audience overlap: the most specific audience wins within an XT activity; order experiences deliberately.

Lift = (conversion rate of experience - control rate) / control rate, expressed as a percentage.

Confidence interval width shrinks as sample size grows; report the interval, not just the point estimate.

A4T uses the Analytics visitor ID; mismatched IDs break stitching between Target and Analytics.

Segment breakdowns in the A4T panel reveal which audiences drove lift.

Auto-Target reports show the model's predicted vs actual performance per experience.

Guardrail metrics (revenue, engagement) should be monitored alongside the primary metric.

The Experience Cloud org ID links Target, Analytics, AAM, and AEP; a mismatch breaks integrations.

A4T requires the Analytics tracking server and the Target report suite mapping.

AAM segments are evaluated at the edge; segment membership is passed via the DIL or Web SDK.

AEP destinations forward audiences to Target; the destination must be enabled per sandbox.

Campaign integration uses the Delivery API or the offer library; test the offer rendering in Campaign.

Third-party systems authenticate with OAuth server-to-server credentials scoped to the org.

at.js 2.x: getOffers returns propositions; applyOffer renders them; the display event fires on render.

Web SDK: sendEvent with decisionScopes and renderDecisions; the Edge Network routes to Target.

Delivery API: POST /rest/v1/delivery with context, execute, and prefetch; the response contains propositions.

Mobile SDK: the Target extension caches offers for offline; events queue and flush on reconnect.

Server-side: pass the customer ID and session ID; the display event is required for impressions.

Response tokens expose activity and experience attributes for analytics and downstream personalization.

Consent gating: the opt-in service blocks Target calls until consent is granted.

Privacy API: access and delete requests are scoped to the org and visitor ID.

Profile lifetime controls data retention; configure it to match the privacy policy.

PII must not be sent in mbox parameters; use hashed identifiers and the customer ID.

Role-based access via Admin Console product profiles limits who can edit and approve activities.

AEP data governance labels restrict which attributes can be used for personalization.

Approvals: the Approver role is required to activate an activity; Editors can create but not approve.

QA links carry the qa parameter and a token; they bypass audience qualification.

Troubleshooting: inspect the mbox request, the response offers, then the audience qualification.

Common failures: missing display event, audience mismatch, selector drift, and offer rendering errors.

The Target trace (mboxTrace) shows decision details and audience evaluation.

Activity collisions: higher priority wins; exclusions prevent overlap on the same location.

Client-side at.js/Web SDK is simplest for marketing-managed tests; server-side is required for regulated or offline channels.

The global mbox supports Auto-Target, Auto-Allocate, and Recommendations on page load.

Naming conventions for mboxes and locations enable reusable audiences and reports.

ODD requires a rules artifact refresh; stale rules may miss audience changes.

Server-side calls count against the org's API limits; plan for peak QPS.

QA plan: staging validation, QA links, and A4T stitching checks.

A/B: 2+ experiences, percentage split, one success metric; manual or automatic winner.

XT: 1+ experiences bound to audiences; first match wins, so order matters.

MVT: full factorial tests all combinations; partial factorial tests a subset.

Auto-Target and AP use a contextual bandit model; AP personalizes per visitor across offers.

Recommendations criteria: popularity, recently viewed, user-based, item-based, custom; each has a backup.

Priority: activity priority (0-999, higher wins) then audience specificity; ties resolve by creation order.

Audience rules combine geo, device, browser, custom parameters, profile attributes, and Adobe audiences.

Custom audiences are evaluated server-side; profile scripts compute derived attributes at profile update.

AAM segments reach Target via the DIL or Web SDK; membership is read at decision time.

AEP audiences are shared via the Experience Platform destination; streaming updates near real-time.

Profile attributes persist up to the profile lifetime (default 14 days of inactivity).

Exclude internal traffic with IP-based or QA audiences; size estimates are approximate and cached.

A4T links Target to Analytics report suites; the panel shows lift, confidence, and segment breakdowns.

Significance uses a two-tailed test at 90% or 95% confidence; report the interval, not just the point estimate.

Sample size depends on baseline rate, minimum detectable effect, power (80%), and significance level.

Target reports update near real-time; A4T data flows through Analytics and can lag by hours.

Auto-Allocate shifts traffic to the best experience; Auto-Target personalizes per visitor.

Conversion rate = conversions / visitors; RPV = revenue / visitors.

The Experience Cloud org ID links Target, Analytics, AAM, and AEP; a mismatch breaks integrations.

A4T requires the Analytics tracking server and the Target report suite mapping.

AAM segments are evaluated at the edge; membership is passed via DIL or Web SDK.

AEP destinations forward audiences to Target; the destination must be enabled per sandbox.

Campaign integration uses the Delivery API or the offer library.

Third-party systems authenticate with OAuth server-to-server credentials scoped to the org.

at.js 2.x: getOffers returns propositions; applyOffer renders them; the display event fires on render.

Web SDK: sendEvent with decisionScopes and renderDecisions; the Edge Network routes to Target.

Delivery API: POST /rest/v1/delivery with context, execute, and prefetch.

Mobile SDK: the Target extension caches offers for offline; events queue and flush on reconnect.

Server-side: pass the customer ID and session ID; the display event is required for impressions.

Response tokens expose activity and experience attributes for analytics.

Consent gating: the opt-in service blocks Target calls until consent is granted.

Privacy API: access and delete requests are scoped to the org and visitor ID.

Profile lifetime controls retention; configure it to match the privacy policy.

PII must not be sent in mbox parameters; use hashed identifiers and the customer ID.

Role-based access via Admin Console product profiles limits who can edit and approve activities.

AEP data governance labels restrict which attributes can be used for personalization.

Approvals: the Approver role is required to activate an activity; Editors can create but not approve.

QA links carry the qa parameter and a token; they bypass audience qualification.

Troubleshooting: inspect the mbox request, the response offers, then the audience qualification.

Common failures: missing display event, audience mismatch, selector drift, and offer rendering errors.

The Target trace (mboxTrace) shows decision details and audience evaluation.

Activity collisions: higher priority wins; exclusions prevent overlap on the same location.

Client-side at.js/Web SDK suits marketing-managed tests; server-side is required for regulated or offline channels.

The global mbox supports Auto-Target, Auto-Allocate, and Recommendations on page load.

Naming conventions for mboxes and locations enable reusable audiences and reports.

ODD requires a rules artifact refresh; stale rules may miss audience changes.

Server-side calls count against the org's API limits; plan for peak QPS.

QA plan: staging validation, QA links, and A4T stitching checks.

A/B: 2+ experiences, percentage split, one success metric; manual or automatic winner.

XT: 1+ experiences bound to audiences; first match wins, so order matters.

MVT: full factorial tests all combinations; partial factorial tests a subset.

Auto-Target and AP use a contextual bandit model; AP personalizes per visitor across offers.

Recommendations criteria: popularity, recently viewed, user-based, item-based, custom; each has a backup.

Priority: activity priority (0-999, higher wins) then audience specificity; ties resolve by creation order.

Audience rules combine geo, device, browser, custom parameters, profile attributes, and Adobe audiences.

Custom audiences are evaluated server-side; profile scripts compute derived attributes at profile update.

AAM segments reach Target via the DIL or Web SDK; membership is read at decision time.

AEP audiences are shared via the Experience Platform destination; streaming updates near real-time.

Profile attributes persist up to the profile lifetime (default 14 days of inactivity).

Exclude internal traffic with IP-based or QA audiences; size estimates are approximate and cached.

A4T links Target to Analytics report suites; the panel shows lift, confidence, and segment breakdowns.

Significance uses a two-tailed test at 90% or 95% confidence; report the interval, not just the point estimate.

Sample size depends on baseline rate, minimum detectable effect, power (80%), and significance level.

Target reports update near real-time; A4T data flows through Analytics and can lag by hours.

Auto-Allocate shifts traffic to the best experience; Auto-Target personalizes per visitor.

Conversion rate = conversions / visitors; RPV = revenue / visitors.

The Experience Cloud org ID links Target, Analytics, AAM, and AEP; a mismatch breaks integrations.

A4T requires the Analytics tracking server and the Target report suite mapping.

AAM segments are evaluated at the edge; membership is passed via DIL or Web SDK.

AEP destinations forward audiences to Target; the destination must be enabled per sandbox.

Campaign integration uses the Delivery API or the offer library.

Third-party systems authenticate with OAuth server-to-server credentials scoped to the org.

at.js 2.x: getOffers returns propositions; applyOffer renders them; the display event fires on render.

Web SDK: sendEvent with decisionScopes and renderDecisions; the Edge Network routes to Target.

Delivery API: POST /rest/v1/delivery with context, execute, and prefetch.

Mobile SDK: the Target extension caches offers for offline; events queue and flush on reconnect.

Server-side: pass the customer ID and session ID; the display event is required for impressions.

Response tokens expose activity and experience attributes for analytics.

Consent gating: the opt-in service blocks Target calls until consent is granted.

Privacy API: access and delete requests are scoped to the org and visitor ID.

Profile lifetime controls retention; configure it to match the privacy policy.

PII must not be sent in mbox parameters; use hashed identifiers and the customer ID.

Role-based access via Admin Console product profiles limits who can edit and approve activities.

AEP data governance labels restrict which attributes can be used for personalization.

Approvals: the Approver role is required to activate an activity; Editors can create but not approve.

QA links carry the qa parameter and a token; they bypass audience qualification.

Troubleshooting: inspect the mbox request, the response offers, then the audience qualification.

Common failures: missing display event, audience mismatch, selector drift, and offer rendering errors.

The Target trace (mboxTrace) shows decision details and audience evaluation.

Activity collisions: higher priority wins; exclusions prevent overlap on the same location.