Why Page Analytics Alone Isn’t Enough
Classic AEM reporting answers “what happened on this page” pageviews, bounce rate, time on page. It cannot answer “what should this specific visitor see next,” because that answer requires connecting page-level behavior to the visitor’s entire cross-channel journey: email opens, app sessions, support tickets, past purchases.
Adobe Customer Journey Analytics closes that gap by ingesting every one of those signals into a single data model, and Sensei’s AI layer turns the resulting flood of cross-channel data into surfaced insights an author can act on without writing a query.
Layer 1 – Customer Journey Analytics: The Unified Data Model
CJA sits on the AEP Edge Network and joins data from every connected source such as AEM Edge Delivery pages, mobile apps, call center logs, POS systems all into one queryable data lake, without moving or duplicating the data itself.
- Cross-channel stitching – a single visitor ID follows the customer from an AEM landing page to a mobile app purchase to a support call
- Data views – the same underlying data lake can be sliced into marketing, product, or support-specific views without re-ingesting anything
- Derived fields – calculated metrics (e.g. session quality score) computed once and reused across every report
Layer 2 – Sensei AI: Anomaly Detection and Contribution Analysis
Sensei’s AI models run continuously against CJA’s data model, not on a schedule an analyst sets.
Anomaly Detection: Flags statistically significant deviations i.e. a template’s conversion rate, a checkout step’s drop-off, a content block’s engagement and the moment they occur, with a plain-language explanation of the deviation’s size and confidence.
Contribution Analysis: When an anomaly fires, Sensei automatically tests which dimensions (device, geography, campaign, AEM component variant) explain the change, ranking them by statistical contribution instead of a human guessing which filter to try first.
Layer 3 – Conversational AI Assistant: Ask, Don’t Query
CJA’s Conversational AI assistant lets any team member type a plain-language question “which AEM landing page had the biggest engagement drop this week for mobile visitors in Germany?” and receive a chart-backed answer generated from the same underlying data model, with no query language required.
Where this lives: Surface CJA insight cards directly inside the Universal Editor sidebar so authors see performance context while they edit not three tools away in a separate analytics login.
From Insight to Next-Best-Action
Detecting an anomaly is only half the job. The other half is deciding what to do about it and that connects back into the same personalization stack (Adobe Real-Time CDP + Adobe Target).
- Sensei flags a conversion drop on a hero component for mobile Safari visitors in Germany
- An App Builder action pushes that segment definition into Adobe Real-Time CDP as a live audience
- Adobe Target automatically routes that audience to the next-best-performing variant, without waiting for a human to build a new experiment
- CJA measures the recovery and confirms whether the automated fix worked, closing the loop
Governance note: Automated next-best-action should still route through the same brand and compliance checks used for AI content governance, an AI-detected anomaly should never trigger an unreviewed content swap on a regulated page.
Implementation Checklist
- Connect AEM Edge Delivery page data, mobile, and offline sources into a shared CJA data view
- Define derived fields once for metrics reused across marketing, product, and support views
- Enable Sensei Anomaly Detection on the metrics that matter most to your key AEM templates
- Route anomaly webhooks to Slack/App Builder for triage instead of relying on dashboard checks
- Build an App Builder action to surface CJA insight cards inside the Universal Editor rail
- Wire confirmed anomalies into Real-Time CDP audience creation for automated next-best-action
- Route any automated content change through existing governance/compliance checks first
- Review Conversational AI assistant answers against the underlying workspace query periodically to catch model drift
What to Measure
- Time-to-detection – how long between a real anomaly occurring and Sensei flagging it
- Time-to-action – how long between detection and a corrective experience going live
- False-positive rate – how often flagged anomalies turn out to be normal variance, to tune sensitivity
- Insight adoption – how many non-analyst team members are using the Conversational AI assistant weekly
Final Thoughts
The AEM teams getting the most value from analytics in 2026 aren’t the ones with the biggest dashboards. They’re the ones who’ve collapsed the distance between “we noticed a problem” and “we fixed it” down to minutes instead of days by letting Sensei watch the data continuously and routing what it finds straight back into the personalization and governance systems already in place.
Start with one high-traffic template, one anomaly metric, and one automated action. Prove the loop closes before expanding it site-wide.
