Note: The AI features in CommunityWFM use only your center's data. CommunityWFM does not connect to outside systems or large language models (LLMs) to acquire data to use for the AI forecasting or anomaly detection.
AI Forecasting
Automated forecasting allows the option to use closed-environment, predictive intelligence to automate forecasting, building and managing automated forecasts to model your business activity. The AI forecast automatically uses up to five years of your historical data to create forecasts and, if desired, generate/publish staffing requirements. The more data in your system, the more accurate the forecast, and as always, you can edit and revise the forecast as needed to meet the needs of your center.
Enable the AI forecasting service
First, enable the service by navigating to Settings > Application settings > Administrative settings > Services and click Automated Forecast Service.
Service enabled? Select Yes to enable AI forecasting to run in the background.
Impersonated user id: Select from the list of supervisor or higher people. There must be a designated user ID for the service to run.
Start time: This time is in UTC (Universal Time Coordinated). Select a time that is usually slow for your center to run this background service. Each day the service will run at the selected time and check if there are any pending forecasts to be run that day. If there are pending forecasts, they will run at this time.
Set up the AI Forecast
Access the AI Forecasting feature through the new main menu option. Select the submenu link or Set up automated AI Forecasts.
Manage existing automated forecasts or create a new one.
Setup > Basic properties
Name: Give your forecast a name and optional description.
Selected activity: Select an activity. A higher level with many included activities will take longer to run.
Generate/publish staffing requirements? Select whether to generate staffing requirements, and whether to publish staffing requirements.
- Do not generate staffing requirements
- Generate staffing requirements but do not publish: This will create a working forecast with staffing requirements in Forecast > Working forecasts.
- Generate staffing requirements and publish: The staffing requirements is published and available to use in a working schedule.
When generated, the forecast description in the list of working forecasts will include the forecast name, date generated, and forecast dates.
For example
Each time the service runs, it will create a new revision of this working forecast.
Forecast recurrence
Parameters entered here determine how often and on what schedule the forecast will run, when the forecast starts after the release date, and how many days are included in the forecast.
Next forecast release date: The date the forecast data will be available. The forecast engine will use all historical data up to this date to create the forecast. This date must be in the future (tomorrow or later).
Recurrence pattern: options include daily, weekly, monthly by day of month (e.g., on the 1st of the month) or monthly by day & week (e.g., every second Friday). The next options will change based on the selected recurrence pattern.
Forecast recurrence interval: select a frequency.
Forecast offset (# of days after release date): number of days after the release date to start the forecast. Must be between 1 and 15 days and less than the forecast horizon. For example, you may release a forecast on Friday that forecasts for dates starting on Monday (3 day offset).
Forecast horizon (# of days after release date): number of days after the release date to end the forecast. Must be 35 days or fewer and greater than the forecast offset.
Preview Upcoming Dates: Click to preview the next five forecast dates based on the selections.
When the recurrence parameters are complete, the tiles will turn green.
Machine learning
Examine the forecast model data used to train this automated forecast.
- Verify historic dates (Earliest Event Time): With more historical events collected, better seasonality (yearly, monthly, weekly) trends can be predicted.
- Verify recent dates (Latest Event Time and Days Since Last Event): Stale data could affect the prediction accuracy with possible uncertainty similar to starting over with no data.
- Data completeness (Counts and Collections): Missing data not reported/collected will affect prediction accuracy.
Use the tiles at the top of the page to examine individual parts of the data in your model.
Verify Dates
This is Valid / green if there are 365 days of data AND new data within the last 7 days. If any activities do not meet this, the icon for Valid? will be yellow, but you can proceed.
Data Completeness
This is Valid / green if the Collection Percent is ≥ 90. If any activities do not meet this threshold, the icon for Valid? will be yellow, but you can proceed.
Data Consistency
This is Valid / green if the contact volume relative standard deviation is ≤ 30 AND AHT relative standard deviation is ≤ 30. If any activities do not meet this, the icon for Valid? will be yellow, but you can proceed.
Manage existing forecasts
View a list of existing automated forecasts including the name, activity, next release date, next forecast period, and the status (enabled or not). This is where you can enable or disable an automated forecast by clicking the toggle, or delete disabled forecasts (enabled forecasts cannot be deleted).
If the next forecast release date has Pending further work, dates have not been selected and the forecast cannot be enabled. Click the name of the forecast to go to the setup page and finish.
Click a forecast name to view and edit the properties. The banner at the top of the page displays if the forecast is complete.
The Basic Properties tab is Valid / green if the forecast is enabled and settings complete. If the icon is yellow, the forecast is not enabled.
Overview of AI Forecasting Flow
AI Anomaly Detection
The anomaly detection engine breaks down complex, noisy data into simple, interpretable parts: a long-term trend, repeating cycles (oscillations), and random noise. It works by transforming a time series into a matrix, decomposing it, grouping similar patterns, and rebuilding the clean data.
When enabled, CommunityWFM will monitor incoming attendance and contact volume data (contact volume, AHT, and calls abandoned) using data-driven statistical methodology to detect outlier data. Anomaly detection for attendance is only available for daily or weekly intervals and not hourly, since attendance is typically a once per shift occurrence.
If detected, selected roles receive a notification.
Anomaly detection can help predict staffing needs, identify times of spikes or troughs in call volume (system failure; fraudulent activity), and analyze campaign performance.
- Rapid incident response: Catch spikes within 60 minutes rather than waiting for daily reports.
- Short-term seasonality: Effective for patterns that repeat daily (e.g., higher traffic during the day, low at night).
- Balanced context: It ignores minor hourly fluctuations (noise) that aren't sustained, reducing alert fatigue.
- Business KPIs: Best for tracking metrics, which may behave differently on Mondays vs. Sundays.
- Accounting for weekly cycles: Enables models to compare this Sunday against previous Sundays.
- Identifying long-term trends: Useful for identifying significant changes in business metrics.
- Ignoring daily fluctuations: Good for metrics that are not time-sensitive, where hourly or daily changes are not actionable.
- Seasonal stability: Used when weekly patterns are consistent but daily patterns are too volatile.
Enable anomaly detection service
Navigate to Settings > Application settings > Administrative settings > Services and enable the service. If not enabled here, the anomaly detection service will not work.
Set up an AI anomaly detection profile
Navigate to AI > Anomaly detection > Anomaly detection profiles.
Select Click here to create a new anomaly detection profile. The profile will include the frequency and notification parameters.
Profile name: Enter a descriptive name.
Profile type: Select Attendance. After saving the profile, you can’t change the profile type.
Set as default profile? You may have one default profile for each anomaly type.
Daily detection and Weekly detection: Enable one or both frequencies. Hourly detection is only available for contact volume profiles.
Hourly detection runs at the top of the hour; daily detection runs within one hour after after the last interval scheduled for that activity and day in the Enterprise Model Work Habits and Hours, and weekly detection runs after 23:00 (11:00 PM) UTC on the evening before the first day of schedule week (as determined by the First day of schedule week in the Enterprise Model settings (Settings > Enterprise model > Enterprise or Site properties). If the first day of the schedule work week is Monday in the Enterprise Model basic properties, it will run Sunday night.
Confidence: Refers to how sure the system is that an occurrence is unusual or abnormal. Select highest, high, medium, or low. Higher confidence may result in fewer reports and miss some anomalies. Lower confidence may return some false positives..
PValue History Length: Refers to how much past data the detection system uses to decide whether a new incoming data point is anomalous. Select highest, high, medium, or low.
- A lower PValue History Length means the model "forgets" previous large spikes or patterns faster, making it more sensitive to recent changes. This can potentially lead to more detected anomalies (including more false positives) if the data has frequent, short-lived variations.
- A higher PValue History Length means the model considers a longer history, making it less sensitive to short-term fluctuations and more likely to alert on persistent changes, thus potentially leading to fewer anomalies detected overall but with higher confidence.
Notification frequency: Select how often to receive notifications - never, any time there is a detected anomaly, or once per day.
Who should receive a notification when an anomaly is detected? What role should receive notifications and via which channel(s)? Select a role from the drop-down and click each notification channel to include. Note: To receive notifications, the person must be in the reporting tree for the activity, and notifications enabled for anomaly detection (Settings > Application settings > Administrative settings > Notifications > Anomaly Detection).
Return to the Activity Anomaly Detection hub and select an Activity.
The alert banner at the top reports whether there are current anomalies detected and if detected, a list of the anomalies.
Alert thresholds
Configure alert thresholds within the anomaly detection profile
Confidence: Refers to how sure the system is that an occurrence is unusual or abnormal. Select highest, high, medium, or low. Higher confidence may result in fewer reports.
PValue History Length: Refers to how much past data the detection system uses to decide whether a new incoming data point is anomalous. Select highest, high, medium, or low.
- A lower PValue History Length means the model "forgets" previous large spikes or patterns faster, making it more sensitive to recent changes. This can potentially lead to more detected anomalies (including more false positives) if the data has frequent, short-lived variations.
- A higher PValue History Length means the model considers a longer history, making it less sensitive to short-term fluctuations and more likely to alert on persistent changes, thus potentially leading to fewer anomalies detected overall but with higher confidence.
Notification frequency: Default is Notify me any time there is an anomaly detected. Options include no notification or once per day.
Who should receive a notification when an anomaly is detected? Select the role that should receive notifications.
Notification configuration
First, configure Channels to notify on in Settings > Application settings > Administrative settings > Notifications.
Select what role should receive notifications and via which channel(s). Note: To receive notifications, the person must be in the reporting tree for the activity.
Run Now
Use the Run Now function to assess anomalies at a leaf-node level of the Enterprise Model. To view anomalies at the folder or site level, set up an anomaly profile.
Access the Run Now option by navigating to AI > Anomaly detection > Anomaly detection run now, the Anomaly detection menu, or from the tile in the anomaly detection hub. This allows running the report on an ad hoc basis or viewing data from past dates.
Activity: Select the leaf-node activity. To view anomalies at the folder or site level, create an anomaly profile.
Through date: Select the end date of the report period. The start date will depend on the selected interval and will be visible in the report.
- Hourly: 14 days before the through date (not available for attendance).
- Daily: 90 days before the through date.
- Weekly: 104 weeks before the through date.
Make your selections for intervals, anomaly type, confidence, and PValue history length, then click Apply options.
If anomalies are detected, the report shows dates of detection and a value.
- For attendance — a count of the check-in type for the selected time period (day/week) - not the number of anomalies.
- For call volume — an average across the time selected values of the anomaly type collected every 15 minutes then averaged by hour/day/week).
The system displays rows exclusively for dates where anomalies have been found.
Attendance
Call volume