Choose the right long-term personalization approach
Long-term personalization means tailoring a real-time experience based on customer behavior that stretches back six months, a year, or longer. You can support these use cases without storing years of raw event history in the Profile store: keep the detailed history in the analytical layer, derive a compact signal from it, and promote only that signal for activation. This decision guide compares the available approaches and helps you choose the one that fits your architecture. It explains when, why, and which approach to use. For configuration steps and SQL syntax, follow the linked product documentation.
This guide is for solution architects, platform architects, and technical decision-makers evaluating long-term personalization approaches in Experience Platform. It assumes familiarity with profiles, audiences, the data lake, and the Profile store.
Why long-term history doesn’t belong in the Profile store customer-problem
Many organizations accumulate large volumes of historical event data in the Profile store on the assumption that more stored data produces better personalization. In practice, this drives up Profile store size and Total Data Volume entitlement without improving the experiences your customers receive.
Long-term personalization use cases are common in industries with long purchase cycles or slowly evolving customer relationships:
- A financial services customer who opened a savings account 18 months ago and is now researching mortgage products.
- A retail shopper who browsed seasonal inventory last year but never purchased.
- An airline frequent flyer whose loyalty tier reflects 12 months of cumulative activity.
- A telecommunications customer who churned eight months ago and has since returned, identifiable as at-risk from historical behavior.
In each case the relevant data is historical, but the personalization moment must happen in real time. The problem arises when organizations store all of that event-level history directly in the Profile store to support those moments.
The Profile store is optimized for speed, not volume, and every record it holds counts toward your Total Data Volume entitlement. Raw events from 12, 18, or 24 or more months ago are valuable for analysis but rarely need to live in the Profile store to support activation.
How to identify a long-term personalization use case identify
Use the following questions to determine whether your organization is placing analytical demands on the Profile store that go beyond its intended scope. If two or more apply, you likely have a long-term personalization use case that a derived-signal architecture would serve more efficiently.
- How far back does your segmentation logic look? Audiences built with “any time” logic or lookback windows beyond 30–90 days are a strong signal.
- Are you approaching or exceeding your Total Data Volume entitlement? If so, examine how much of that volume is historical event data that is rarely used for real-time activation.
- What time-to-live is configured on your Profile data? No TTL, or a TTL of 12 or more months, means raw events are likely accumulating without a clear activation purpose.
- Is your industry characterized by long purchase cycles? Mortgages, insurance, B2B contracts, seasonal retail, and travel rewards all involve relationships that evolve over months or years.
- Do you have Data Distiller or Customer Journey Analytics licensed but not used for audience creation? These tools are designed for the analytical work you may currently be asking the Profile store to do.
If these questions point to a long-term personalization pattern, the next step is understanding where each type of data belongs in Experience Platform.
Where your data should live: analytical and engagement workflows workflows
Experience Platform stores data in two repositories that serve different workflows. Matching each dataset to the workflow it supports is the foundation of an efficient long-term personalization architecture.
The guiding principle is to personalize on the signal, not the raw history. You do not need to store years of raw behavioral events in the Profile store to know that a customer qualifies as a loyalty tier member, a high-value buyer, or an at-risk churner. You compute those labels from the historical data in the analytical workflow and store only the label (the signal) in the Profile store. The raw history stays in the data lake, where it does not count toward your Profile entitlement.
Recommended architecture architecture
The recommended architecture separates long-term storage from real-time activation:
- Customer event data accumulates in the data lake. Web, app, purchase, loyalty, and offline records are retained at full historical depth without contributing to your Profile entitlement.
- The analytical workflow derives a signal. Queries or analyses transform months or years of history into a compact output: a loyalty tier, a churn risk score, a lifetime value ranking, or a qualified audience.
- The signal is promoted to the Profile store. Only the output, not the underlying event history, enters the engagement workflow.
- Real-Time Customer Data Platform and Adobe Journey Optimizer activate on the signal. The activation layer operates on lean, current data and responds in milliseconds.
The approaches in this guide are different ways to perform step 2 and step 3. Before choosing among them, confirm that your use case actually requires an analytical workflow beyond Real-Time CDP and Adobe Journey Optimizer.
Before you choose: confirm you have a long-term need qualifier
Not every organization needs an analytical approach. If all of your segmentation and personalization logic operates within a 30–90 day behavioral window, and current profile attributes plus recent events in the Profile store contain everything your decisions require, then Real-Time CDP and Adobe Journey Optimizer alone are sufficient. No additional tooling is required, and activation runs directly from the Profile store.
This baseline is the starting point, not a fourth approach. That limit is where the remaining approaches provide additional value: when segmentation logic extends beyond 30–90 days, or personalization depends on insight derived from months or years of history, storing that history in the Profile store raises Total Data Volume without improving outcomes. If your diagnostic in the previous section pointed to a long-term need, choose one of the three approaches that follow.
Quick chooser quick-chooser
If you already know your goal, use this table to find a likely starting point, then read the approach sections and the decision guide to confirm.
Choose an analytical approach approaches
Each approach below describes what it does, when to use it, and its strengths and limitations. Jump to the approach the Quick chooser pointed you to, or read all three to compare them.
Data Distiller data-distiller
To query, transform, and enrich data at scale using SQL, use Data Distiller, an Experience Platform add-on. For long-term personalization, it reads historical event data from the data lake, applies transformation logic, and writes only the resulting signal to the Profile store.
Using SQL, a data engineer defines the transformation logic and produces a derived dataset, a compact output such as a loyalty tier or churn score. The query is scheduled to refresh automatically and publishes only that signal to the Profile store. Data Distiller can also build and publish audiences directly from SQL without first creating a derived attribute.
- Best for: loyalty and churn scoring, lifetime value, decile and percentile rankings, RFM models, and automated recurring refresh.
- Strengths: produces mathematically exact outputs, refreshes automatically on a schedule, and keeps raw data in the data lake.
- Limitations: requires SQL and a data engineer. The transformation must be defined in advance, so it is not suited to exploratory analysis.
To implement this approach, including an end-to-end worked example, see Long-term personalization with Data Distiller.
Customer Journey Analytics customer-journey-analytics
To analyze customer behavior across channels and over time, use Customer Journey Analytics. For long-term personalization, it provides a visual, no-code environment for exploring historical journeys and a direct path (Audience Publishing) to promote the resulting audiences to the Profile store.
Historical data analyzed in Customer Journey Analytics does not need to be enabled for the Profile store, so months or years of event history can reside there without counting toward your Total Data Volume entitlement.
Analysts explore journeys visually across any time horizon, combine behavioral criteria into an audience, and publish it to the Profile store, where it becomes available in Real-Time CDP and Adobe Journey Optimizer. Audiences can be published once for a campaign or refreshed automatically.
- Best for: audiences discovered through exploration, multi-channel journey analysis, re-engagement campaigns, and one-time audiences tied to a specific event.
- Strengths: self-service for analysts and marketers, with no SQL required. Exploration-first, so audience definitions can emerge from the data. Publishes to activation within minutes.
- Limitations: outputs are audience memberships, not computed attributes written back to every profile. Filter-based precision is approximate rather than exact.
Federated Audience Composition federated-audience-composition
To build and qualify audiences directly from data in an external enterprise data warehouse, without moving the underlying data into Experience Platform, use Federated Audience Composition. Audience definitions are applied against warehouse data using a no-code composition canvas, and only the resulting audience membership is sent to Experience Platform for activation.
- Best for: organizations whose historical data is the system of record in a warehouse such as Snowflake, Databricks, Google BigQuery, or Amazon Redshift, and who want to activate it without a full ingestion project.
- Strengths: the underlying data never leaves the warehouse, which satisfies governance and sovereignty requirements. No large-scale ingestion is required.
- Limitations: dependent on warehouse connectivity and composition. Activation is limited to audience membership.
Compare the approaches comparison
The following table compares the three approaches. Real-Time CDP and Adobe Journey Optimizer on their own remain the baseline for use cases that need only recent data, as described in Before you choose.
Decide which approach to use decision-guide
The Quick chooser gives a starting point. Use the following questions to resolve nuanced or combined cases.
Do you need mathematically precise outputs? If your use case requires exact rankings (for example, the top 10% of customers by a calculated metric or a lifetime value in the eighth decile), choose Data Distiller. If you cannot define the calculation precisely in advance, start with Customer Journey Analytics.
Are you starting from exploration? If you do not yet know what the audience looks like and want to discover patterns visually, start with Customer Journey Analytics. Once a pattern is validated, a data engineer can build a scheduled, precision-scored version in Data Distiller if that rigor is needed.
Do you need a reusable profile attribute or a one-time audience? Data Distiller writes a derived attribute, such as a score or tier, to every profile and can recompute it automatically on a schedule. Use it when the signal must be reusable across many audiences and journeys. Customer Journey Analytics instead publishes an audience membership, which suits a campaign audience or a multi-step journey definition built without engineering involvement.
What do your data location and licensing allow? If historical data must remain in an external warehouse, use Federated Audience Composition. Otherwise, let the tools you have licensed and the availability of engineering support guide the choice.
Keep the Profile store lean keep-lean
Choosing the right approach keeps new historical data out of the Profile store. Actively managing what is already there is a complementary, ongoing practice for every deployment. Rather than repeat that guidance here, use the dedicated resource:
- For license entitlement monitoring, ingestion filters, and dataset hygiene, see Data management license entitlement best practices.
Next steps next-steps
Long-term personalization does not require storing years of raw event data in the Profile store. It requires deriving the right signal from that history and making only that signal available for real-time activation. Confirm you have a long-term need, choose the approach that fits your architecture and team, and follow the linked product documentation to implement it.