Intelligence

Beyond rule-based enforcement, FORG applies statistical analysis to your team's AI usage patterns to surface anomalies, forecast spend, and identify optimization opportunities automatically.

Anomaly detection

FORG establishes a usage baseline for each user and team over the first 14 days of data, then flags deviations that fall outside two standard deviations of normal behavior:

• Spend spikes — a user's daily spend is 3x+ their trailing 7-day average
• Token burst — a single session consumes more tokens than the user's top 1% historical session
• Model drift — a user suddenly shifts from cheaper models to significantly more expensive ones
• Error spike — error rate for a user or project exceeds 10% of calls

Anomalies are surfaced in Dashboard → Intelligence → Anomalies and can optionally trigger webhook notifications.

Spend forecasting

FORG computes a projected month-end spend for each user, team, and the org using a 7-day rolling average daily spend rate:

projected_month_end = spend_to_date + (avg_daily_spend_7d * days_remaining)

Forecast alerts can be configured to fire when the projected month-end exceeds a threshold — useful for catching overspend before the month ends rather than after.

Model efficiency analysis

FORG tracks tokens-per-task at the session level, enabling comparison across developers and models. The Model Efficiency report shows:

• Average tokens per session by model
• Average cost per session by model
• Model mix over time (are developers over-indexing on expensive models?)

Team comparison

The Team Benchmarksview shows relative AI spend per developer across teams, normalized by team size. This helps identify whether a team's spend reflects high productivity or high waste — context FORG cannot determine, but can surface for human review.

Availability

Basic anomaly detection is available on all plans. Spend forecasting and model efficiency analysis require the Team plan or higher. Team comparison and advanced intelligence features require Enterprise.