Define Relative Time-Window Metrics

Calculated Insights Time-Window Metrics

For calculated insights, use the NOW function to establish dynamic or relative time windows. The SECOND_ADD and SECOND_SUB functions provide more precise control to define the exact time window for your metrics.

• Availability:
  • The NOW, SECOND_ADD, and SECOND_SUB functions are available in the transform node for Visual Insight Builder.
  • These functions are included in the SQL Functions for Insights.
• Filtering: You can define a relative time filter with the Visual Builder filter node "Relative DateTime operator" option.
• Flexibility: You can define and calculate dynamic metrics over configurable time intervals, from seconds to years (365+ days).

Real-Time Time-Window Metrics

Use dynamic time-window metrics for real-time calculation of metrics over specific time frames. Use these metrics for personalized product recommendations informed by recent activity. For instance, you can:

• Count website interactions for the last 10 minutes to recommend products similar to recent views.
• Remove products that weren’t viewed in the last 10 minutes.

How Real-Time Insights Update Metrics

Real-time insights update metrics when a data event is processed.

• Calculation at query time: The value for this metric is calculated at graph query time.
• Handling no events: When no events are processed, the metric value is calculated based on the defined relative time window. For example, if your metric is "count of product clicks for the last ten minutes," and no clicks occur during this time, the metric is 0.
• Client-side processing: Real-time graph clients must process the real-time insight data from the client side to get the true metric at graph query time.

Streamlining Data Aggregation with Submetrics and Time Buckets

To streamline data aggregation and minimize storage within the real-time graph, the time-window process employs submetrics and time buckets within the user-defined time interval. This process eliminates the need to store every event in the real-time graph.

• How it works: When processing a real-time event, the NOW function compares the event time of each event within a 5-minute sub-time window to determine the appropriate window size. The function applies five different window types in a hierarchical order, with each type covering a specific time range relative to the compute time.
• Automatic dimension creation: When a dynamic time window is defined, the function creates the dimensions window_start__c and window_end__c as real-time insight dimensions.
• Improved storage cost: The time window process can result in an approximate metric rather than an exact count of events. The process is optimized to store the most recent data in smaller, precise buckets, grouping older data into larger buckets, to balance accuracy with storage cost.

Window Types and Logic:

• Five-Minute Windows (Most Recent Data)
  • Covers: Last 1 hour from compute time
  • Maximum Buckets: 12 (ensures at least 1 hour of coverage)
• One-Hour Windows (Recent Data)
  • Covers: From 1 hour ago to 1 day ago
  • Maximum Buckets: 23
• Daily Windows (Medium-Term Historical Data)
  • Covers: From 1 day ago to 1 month ago
  • Bucket Examples: [2025-03-19 00:00 - 23:59], [2025-03-20 00:00 - 23:59]
  • Maximum Buckets: ~30 (varies by length of month)
• Monthly Windows (Long-Term Historical Data)
  • Covers: From 1 month ago to 1 year ago
  • Bucket Examples: [2025-02-01 00:00 - 2025-02-28 23:59], [2025-03-01 00:00 - 2025-03-31 23:59]
  • Maximum Buckets: 11
• Yearly Windows (Oldest Data)
  • Covers: All data older than 1 year from compute time
  • Bucket Examples: [2024-01-01 00:00 - 2024-12-31 23:59], [2023-01-01 00:00 - 2023-12-31 23:59]
  • Maximum Buckets: Unlimited (depends on data age)