New in Analytics Platform: Machine Learning-Powered Insights

Bringing Machine Learning to Every Analyst

We are excited to announce the launch of machine learning-powered insights in the Primates Analytics Platform. This release makes advanced machine learning capabilities accessible to every analyst and business user, regardless of their technical background. No Python scripts, no model training, no data science expertise required—just point our ML engine at your data, and it will automatically detect anomalies, forecast future trends, identify correlations, and surface actionable insights that would be nearly impossible to find through manual analysis.

The motivation behind this feature is simple: while machine learning has enormous potential to transform how organizations understand and act on their data, the barrier to entry has been far too high. Building and deploying ML models traditionally requires specialized skills in statistics, programming, data engineering, and machine learning operations. Most organizations do not have enough data scientists to meet the demand for ML-driven analysis, creating a bottleneck where valuable data goes unanalyzed and insights go undiscovered. Our ML-powered insights feature eliminates this bottleneck by embedding pre-trained, automatically tuned machine learning models directly into the analytics workflow.

The ML engine is built on a foundation of carefully selected algorithms that have been optimized for the types of data and analysis patterns most common in business analytics. For anomaly detection, we use a combination of isolation forest and seasonal hybrid ESD algorithms that can identify unusual patterns in both static datasets and time-series data. For forecasting, we use an ensemble of Prophet, ARIMA, and gradient boosting models that automatically selects the best approach based on the characteristics of the data being analyzed. For correlation analysis, we use mutual information and distance correlation metrics that can detect both linear and non-linear relationships between variables.

How It Works

Using ML-powered insights requires no configuration or setup. When you open any dashboard or report in the Analytics Platform, you will see a new "ML Insights" panel that displays automatically generated insights for the data currently being viewed. The ML engine runs continuously in the background, analyzing your data streams and updating its insights in real time as new data arrives. Insights are ranked by significance and actionability, so the most important findings appear at the top of the list.

Each insight includes a natural language description of the finding, a confidence score indicating the statistical significance of the result, a visualization showing the relevant data pattern, and recommended actions that you can take in response. For example, if the anomaly detection model identifies an unusual spike in error rates for a specific API endpoint, the insight might read: "Error rate for endpoint /api/v2/users increased by 340% compared to the baseline, starting at 14:30 UTC. This anomaly has a confidence score of 0.97. Recommended action: investigate recent deployments and infrastructure changes affecting the users service."

You can also interact with the ML engine programmatically through our SDK. Here is an example of how to query for anomalies in a specific metric:

import { PrimatesAnalytics } from '@primates/analytics-sdk';

const analytics = new PrimatesAnalytics({
  apiKey: process.env.PRIMATES_API_KEY
});

// Detect anomalies in the last 24 hours of error rate data
const anomalies = await analytics.ml.detectAnomalies({
  metric: 'api.error_rate',
  timeRange: { last: '24h' },
  sensitivity: 'medium', // low, medium, high
  groupBy: ['endpoint', 'region']
});

console.log(`Found ${anomalies.length} anomalies:`);
for (const anomaly of anomalies) {
  console.log(`  [${anomaly.severity}] ${anomaly.description}`);
  console.log(`  Time: ${anomaly.startTime} - ${anomaly.endTime}`);
  console.log(`  Expected: ${anomaly.expectedValue}, Actual: ${anomaly.actualValue}`);
  console.log(`  Confidence: ${(anomaly.confidence * 100).toFixed(1)}%`);
}

// Generate a 7-day forecast for daily active users
const forecast = await analytics.ml.forecast({
  metric: 'users.daily_active',
  horizon: '7d',
  interval: '1d',
  includeConfidenceIntervals: true
});

Anomaly Detection Deep Dive

Our anomaly detection engine is designed to handle the complexities of real-world business data, including seasonality, trends, missing data, and variable baselines. Unlike simple threshold-based alerting that requires manual configuration and constant tuning, our ML-based approach learns the normal patterns in your data automatically and adapts as those patterns change over time. This means fewer false positives, faster detection of genuine anomalies, and no ongoing maintenance burden for analysts.

The anomaly detection pipeline processes data through several stages. First, the time series is decomposed into trend, seasonal, and residual components using STL decomposition. The seasonal component captures recurring patterns like daily traffic cycles, weekly business cycles, and monthly reporting patterns. The trend component captures long-term directional changes. The residual component represents the random variation that remains after removing trend and seasonality. Anomalies are identified by analyzing the residual component for values that deviate significantly from the expected distribution.

We have tuned our anomaly detection models on millions of real-world time series from our customer base, which gives them strong out-of-the-box performance across a wide range of data types and patterns. However, the models also support customization through several parameters:

• Sensitivity: controls the threshold for flagging anomalies. Higher sensitivity catches more subtle anomalies but may produce more false positives. Lower sensitivity focuses on the most extreme deviations.
• Seasonality hints: while the models detect seasonality automatically, you can provide hints about known seasonal patterns (daily, weekly, monthly) to improve detection accuracy for data with complex seasonal structures.
• Exclusion windows: allows you to mark time periods where known events (maintenance windows, marketing campaigns, holidays) make anomaly detection results unreliable.

"The best analytics tools are the ones that tell you what you need to know before you know to ask. ML-powered anomaly detection transforms analytics from a reactive exercise into a proactive capability." — DJ Patil, former U.S. Chief Data Scientist

Forecasting Capabilities

The forecasting module enables users to predict future values of any metric with confidence intervals that quantify the uncertainty of the prediction. Forecasts are generated using an ensemble of models that are automatically evaluated and weighted based on their performance on historical data. The ensemble approach ensures robust predictions across diverse data types, from smooth trending metrics to highly volatile seasonal data.

Forecasting is integrated directly into the dashboard experience. Any time-series chart can be extended with a forecast overlay that shows predicted values along with shaded confidence intervals. Users can adjust the forecast horizon from one hour to ninety days, and the system will automatically select the appropriate model granularity and update frequency. Forecasts update in real time as new data arrives, continuously refining their predictions based on the latest observations.

The forecasting engine has shown strong accuracy across our beta customer base. The following table summarizes the forecast accuracy metrics across different data types and forecast horizons:

Getting Started and Pricing

ML-powered insights are available today for all Analytics Platform customers on the Professional and Enterprise plans. The feature requires no additional configuration—simply update to the latest version of the Analytics Platform, and you will see the ML Insights panel appear automatically on your dashboards. For customers on the Free and Developer plans, a limited version of the ML insights feature is available that includes anomaly detection for up to five metrics.

We have prepared comprehensive documentation, including a getting started guide, API reference, and a collection of example notebooks that demonstrate common ML insights use cases. Our developer advocate team has also created a series of video tutorials that walk through the setup process and demonstrate the feature's capabilities with real-world data. We are confident that ML-powered insights will transform how you interact with your data, and we are eager to hear your feedback as you explore the new capabilities.

1. Update your Analytics Platform to the latest version to access ML-powered insights.
2. Visit the ML Insights panel on any dashboard to see automatically generated insights.
3. Try the forecast overlay feature on your most important time-series charts.
4. Explore the SDK documentation to integrate ML insights into your custom applications.