AEM Sites Playbook, Chapter 17: Measurement, personalization and experimentation
Can the team trust measurement and personalized/default experiences?
November 10, 2026
Start with a decision and an event contract
Use this chapter when instrumenting a site, validating a personalization integration or deciding whether a change helped customers. State the decision the measurement should inform. Then define the event, triggering action, required context, destination, responsible owner and quality checks. A collection of tags without that agreement can produce data nobody can interpret.
Separate operating measures from customer outcomes. Approval wait, author task failures and publication-to-observation time help assess the content process. Customer task completion or an experiment's outcome answers a different question. Publishing faster does not establish incremental revenue or improved experience by itself.
Define page/component identifiers and event meaning with analysts and frontend engineers. Cover initial navigation, relevant client-side navigation, repeated interactions, missing context and duplicate sends. Identify consent and identity responsibilities with the appropriate teams; do not assume a working event means the required policy has been applied.
Verify data contracts at each interface
Adobe's Client Data Layer integration guidance supports structured page/component instrumentation and Analytics/Target handoffs. The component exporter discussion shows why that contract must be explicit: custom data appeared in the browser data layer while the requested JSON exporter behavior remained disputed. Some code was exchanged privately and is absent from the discussion. Do not invent its implementation or equate the two interfaces.
Follow one controlled action through the browser event, tag/rule decision, outgoing request and intended reporting destination using authorized tools. Compare field values with the content fixture. A network request establishes transmission, not necessarily correct interpretation in the report. Retain evidence at the checkpoint the analyst will use.
Use the deployment-specific Analytics integration. Current Cloud Service configuration and tag integration differ from legacy framework approaches. Check the applicable API/credential lifecycle before reusing an older recipe.
Treat personalization as a content dependency
Record the targeted content, activity/offer ownership, identity/audience dependency, default experience, measurement and release timing. Validate the default when decision delivery fails or the visitor does not qualify. Check performance and content freshness through the real customer route.
The Target goal-metrics discussion describes an activity that saves but fails during publication. A later author reply recommends a service-pack level without showing an applied upgrade and successful retest. Use it as a target-specific investigation case, not a verified universal fix or a reason to add dummy reporting configuration. Apply the current Target integration guidance to the actual architecture.
For an experiment, agree the hypothesis, audience, primary outcome, duration/stopping approach and analysis owner before launch. Validate allocation and instrumentation with the relevant analytics/experimentation specialists. Do not declare a winner from a single favorable observation or from a reporting setup that changed during the trial.
Define the evidence chain for an event or variation
Community discussions distinguish data contracts at each interface from publication and integration behavior on a particular release. The exporter example does not confirm the requested method behavior, and the Target example does not verify the proposed service-pack remedy. Test each boundary instead of treating either suggestion as a confirmed fix.
Scroll horizontally if needed.
| Question | Record before implementation | Evidence to inspect |
|---|---|---|
| What action is measured? | Business meaning, trigger and required context | The intended interaction on a controlled fixture |
| What should happen once? | Event identity and duplicate-handling expectations | Repeated listeners/navigation and outgoing requests |
| Who may supply identity/context? | Consent and privacy responsibilities | Approved states and implementation behavior |
| Where should the data arrive? | Destination and analyst-owned interpretation | The intended reporting observation, not only a network send |
| Which content should a visitor receive? | Qualification, variation and default | Actual delivered/visible experience under defined conditions |
| What if a dependency fails? | Usable fallback and failure owner | Controlled unavailable/denied response scenarios |
Use current Client Data Layer, Analytics and Target guidance for the selected deployment and interface. A client object, component exporter, outgoing request and report are separate contracts even when they share a field name.
Worked scenario of a campaign interaction
The event counts and observations below are hypothetical. A campaign owner wants to know whether visitors open a product-detail action. The implementation also offers an eligible variation, with an approved default when no decision is available. The analyst defines one event per completed action and the required product/locale context.
The team tests initial navigation, relevant client-side navigation, repeat clicks, missing product context, the organization's required consent states and unavailable decision delivery. It agrees what each observation should mean before inspecting a dashboard.
Scroll horizontally if needed.
| Test | Required behavior | Illustrative initial result |
|---|---|---|
| One ordinary action | One event with correct product/locale context | Correct |
| Action after client-side navigation | Same agreed event meaning without duplicate registration | Multiple outgoing events |
| Context missing | Defined handling without misleading attribution | Old product context is reused |
| Decision unavailable | Approved default remains usable | Default renders correctly |
| Relevant consent state | Behavior follows the agreed policy/implementation contract | One path needs correction |
| Reporting destination | Analyst can identify the intended action | Requests arrive in an unintended destination |
A working first click is therefore insufficient acceptance. Engineering corrects listener/context behavior, the policy owner reviews the relevant consent implementation and the analyst confirms the destination. The same fixtures are repeated, including the default experience. No winner or customer-value claim is made from the initial defective measurement.
Separate experiment readiness from experiment conclusions
Scroll horizontally if needed.
| Stage | Decision | Evidence |
|---|---|---|
| Instrumentation ready | Can the required events be interpreted? | Correct triggers, context, destinations and duplicate/missing behavior |
| Experience ready | Are variations/default usable and approved? | Content, qualification, performance and failure tests |
| Experiment authorized | Is the hypothesis and analysis plan agreed? | Audience, allocation, outcome, stopping approach and owner |
| Results interpretable | Did conditions remain valid? | Data-quality review and changes during the observation window |
| Business action | What decision follows from the analysis? | Qualified result, limitations and accountable approver |
Define how template, audience, instrumentation or content changes during a trial are handled. The analysis owner should decide whether those changes invalidate comparison. A visually attractive variation or favorable early number is not a substitute for that judgment.
Operate the measurement contract after release
For each important event, retain a concise record of meaning, required fields, trigger, consumer/destination, validation fixture and owner. Link it to the component/content release that can change it. Recheck ordinary and failure behavior after navigation, model or tag changes.
Give content and channel owners a useful default when personalization is unavailable. Define who sees a delivery/configuration failure and how the default is verified. Keep private identity data out of public content and shared diagnostics according to the organization's controls.
Business owners accept the question and proposed action, analysts accept interpretability, engineers demonstrate the delivery contract and content owners approve the visible experience. Those are distinct sign-offs. The operating record should reveal which one is missing rather than mark the integration globally complete because an activity saved or a beacon was sent.
Retain decision-quality evidence
Complete W17 with event definitions, destinations, consent/identity dependencies, fallback behavior and validation. Business owners approve the question; analysts accept data quality; engineers demonstrate reliable delivery; content teams confirm the correct variation/default. Review instrumentation when templates, navigation, APIs or content models change. Keep the intended customer experience usable when measurement or personalization is unavailable.
W17 Measurement and personalization
Use with chapter 17. Connect the collected event with the decision it should inform.
Scroll horizontally if needed.
| Record | Entry |
|---|---|
| Business question, metric and process/customer-outcome distinction | [enter] |
| Event trigger, field meaning, identifiers and intended destination | [enter] |
| Consent/identity dependencies and responsible owners | [enter] |
| Browser/rule/request/report observation and duplicate/missing tests | [enter] |
| Content variation/default, qualification and failure behavior | [enter] |
| Experiment hypothesis, analysis ownership and decision rule | [enter] |
