AI & Predictive Widgets

AI & Predictive Widgets

AI widgets use statistical analysis and pattern recognition to surface actionable predictions from your historical data. These are the only widgets in NexusRMS that require the Analytics & AI addon subscription (£79/month) — all other widget types, including scatter plots, heat maps, funnels, and combo charts, are included in the base plan.

Addon requirement

Exactly four widget types require the Analytics & AI addon: predictive_revenue, at_risk_clients, equipment_investment_recommendations, and anomaly_alerts. These are classified as AI widgets in the WidgetPermissionService. When the addon is not subscribed, these widgets still appear in the Widget Library but display an upgrade prompt via the AddonUpsellStrip component. Users cannot add them to a dashboard without an active subscription.

Data maturity requirements

AI widgets require sufficient historical data to generate meaningful predictions. New tenants will see a “Building Your Insights” message with a progress bar until enough data has been collected. The minimum data requirement is approximately 3 months of invoices and project history. Revenue forecasting achieves best results with 12 or more months of historical data. The AiInsightsWidget checks data maturity on load and displays the appropriate state automatically.

Predictive Revenue Widget

The Predictive Revenue widget (backed by RevenueForecastCalculator and rendered via AiInsightsWidget) forecasts future revenue based on historical invoice data. The algorithm uses an ARIMA-like time series approach combining three techniques:

1. Linear trend — least squares regression to identify the baseline growth or decline direction
2. Seasonal decomposition — average revenue by calendar month to detect seasonal patterns (peak months, slow periods)
3. Moving average smoothing — reduces noise from one-off spikes or dips

The forecast projects trend forward with a seasonality adjustment and calculates a 95% confidence interval (1.96 times the standard deviation of residuals). The widget displays the predicted revenue total, a trend direction chip (growing, stable, or declining with percentage change), and the forecast period (configurable to 3, 6, or 12 months via forecast_period). Historical backtesting achieves approximately 85% accuracy. A minimum of 3 months of paid invoice data is required; the calculator returns an insufficient-data error if this threshold is not met.

At-Risk Clients Widget

The At-Risk Clients widget (backed by AtRiskClientsCalculator) identifies clients at risk of churn based on behavioural patterns. The risk score (0–100) is calculated using four weighted factors:

• Inactivity (0–40 points) — days since the client’s last project. 60+ days scores 10 points, 90+ days scores 25, and 180+ days scores the maximum 40
• Payment behaviour (0–30 points) — overdue invoices contribute 10 points each (capped at 30), and a pattern of more than 2 late payments contributes 15 points
• Declining activity (0–20 points) — compares the client’s recent 6-month project rate against their historical average. A drop below 50% of the historical rate scores 20 points; below 75% scores 10
• Declining project values (0–10 points) — if recent project values are less than 70% of previous averages, 10 points are added

Risk levels are classified as: critical (70+), high (50–69), medium (30–49), or low (below 30). The widget displays client name, risk score, risk level badge (colour-coded red, amber, or yellow), the primary risk factor (e.g., “No recent activity”, “Overdue invoices”, “Declining project frequency”), and days since last activity. By default, only clients scoring 50 or above are shown, configurable via min_risk_score.

Equipment Investment Recommendations Widget

The Equipment Investment widget (backed by EquipmentInvestmentCalculator) analyses equipment utilisation, demand patterns, and rental income over a configurable look-back period (default 12 months) to recommend investment decisions. It produces two recommendation lists:

• Buy more — equipment with high demand (60%+ utilisation or shortage instances where booked quantity exceeded owned quantity). Each item shows booking_count, utilization_percent, shortage_instances, estimated ROI, and suggested additional quantity to purchase
• Consider selling — equipment with low utilisation (fewer than 3 bookings in the period, owned for at least 6 months). Each item shows booking count, age in months, purchase_price, estimated current value (depreciated), and capital tied up

The widget also includes category performance analysis (bookings per item by equipment category with growth potential ratings) and an overall ROI analysis showing total investment, annual income, ROI percentage, payback period in years, and health rating (Excellent, Good, Fair, or Needs Attention).

Anomaly Alerts Widget

The Anomaly Alerts widget (backed by AnomalyAlertsCalculator) detects unusual patterns across four business areas:

• Revenue anomalies — weekly revenue that deviates more than 2.0 standard deviations from the mean. Both unusually high and unusually low revenue are flagged
• Booking anomalies — significant drops in weekly booking volume compared to the historical average
• Payment anomalies — high rates of overdue invoices (flagged when exceeding 20% of recent invoices, with severity escalating above 25% and 35%)
• Equipment anomalies — unusually high percentages of equipment with zero bookings in the look-back period (flagged when exceeding 30% and at least 10 items)

Anomalies are classified by severity: critical (3+ standard deviations or extreme rates), high, medium, or low. The widget displays each alert with its title, severity badge, category label, and detection date. Alerts are sorted by severity first, then recency. A summary shows the total alert count, critical count, and number requiring attention. The default look-back period is 12 weeks, configurable via lookback_weeks. A minimum of 4 data points is required for statistical analysis.

AI widget placement

In the Executive Dashboard template, the Predictive Revenue widget is placed as a full-width widget in row 5 (position_y: 11, width: 12, height: 3) with forecast_period set to next_quarter and confidence_interval enabled. The AiInsightsWidget component consolidates all four insight types (revenue forecast, at-risk clients, anomaly alerts, and maintenance predictions) into a single card with expandable sections, a confidence badge, and a “How is this calculated?” methodology dialog.

Methodology transparency

The AiInsightsWidget includes a methodology dialog accessible via the footer link. This dialog explains each prediction algorithm in plain language:

• Revenue forecasting — Holt-Winters Triple Exponential Smoothing
• Churn risk scoring — RFM+ Weighted Scoring
• Anomaly detection — MAD-based Modified Z-Scores
• Maintenance predictions — Weibull Reliability Analysis

A confidence badge in the widget header shows the average confidence level across all active insight sections: high (80%+), good (60–79%), moderate (40–59%), or limited (below 40%). Confidence improves as more historical data becomes available.

Tips for AI widgets

• Review AI insights at least weekly for strategic planning — patterns spotted early are easier to act on
• Combine the At-Risk Clients widget with your CRM workflow — reach out to flagged clients before they churn
• The Equipment Investment widget is most valuable after 6+ months of data, when utilisation patterns become statistically significant
• Pay close attention to anomaly alerts with “critical” severity — these represent deviations of 3 or more standard deviations from normal and warrant immediate investigation
• All predictions are statistical estimates, not guarantees — use them alongside your business judgement, not as a replacement for it