Getting Started with AnalyticKit AI Recommendations: A Step-by-Step Guide

AnalyticKit’s AI-Powered Recommendations feature analyzes your combined Web2 and Web3 analytics data every day and delivers prioritized, actionable insights. This guide walks you through everything you need to know to start using it effectively — from understanding how recommendations are generated to leveraging the full API for programmatic access.

Prerequisites

Before you can receive AI-powered recommendations, make sure you have the following in place:

• An active AnalyticKit account — AI Recommendations are included in all plans at no additional cost. If you do not have an account yet, visit our pricing page to get started.
• Web2 tracking installed — The AnalyticKit JavaScript snippet must be installed on your website or application. This enables collection of visitor data, page views, traffic sources, geographic distribution, and device information. See our integrations page for setup instructions.
• At least 7 days of data — The AI agent computes 7-day rolling averages as part of its analysis. With fewer than 7 days of data, the trend calculations will be incomplete and recommendations may be less accurate.
• Web3 wallet tracking configured (optional but recommended) — While the AI can generate useful recommendations from Web2 data alone, the most powerful insights come from cross-linked Web2 + Web3 data. Configure wallet tracking to enable the AI to analyze the full user journey from page visit to on-chain transaction.

How Recommendations Are Generated

Every day at 6:00 AM UTC, AnalyticKit’s AI agent automatically performs the following process for your team:

1. Aggregates Web2 data from ClickHouse — unique visitors, page views, top traffic sources, geographic distribution, and device breakdown
2. Aggregates Web3 data from PostgreSQL — active wallets, transaction volumes, gas metrics, token transfers, and wallet segment distributions
3. Cross-links Web2 sessions with Web3 wallets using normalized wallet addresses
4. Computes rolling averages — both 7-day and 30-day — across all metrics to identify meaningful trends
5. Sends an aggregated summary (not raw data) to the AI model within a controlled 12,000-token budget
6. Generates 3 to 7 recommendations categorized by type and rated by severity

No setup is required. If your team has an active AnalyticKit account with data flowing in, recommendations start generating automatically. There is no feature flag to enable, no configuration to adjust, and no additional cost.

Viewing Your Recommendations

Via the Dashboard

The easiest way to view your recommendations is through the AnalyticKit dashboard. Log in to your account and navigate to the AI Recommendations section. Your latest recommendations will be displayed with their category, severity level, and full details.

Via the REST API

For programmatic access, use the recommendations endpoint:

GET /api/recommendations/latest/

Example response:

{
  "count": 5,
  "generated_at": "2026-02-17T06:00:12Z",
  "recommendations": [
    {
      "id": "rec_a7f3b901",
      "category": "churn_risk",
      "severity": "warning",
      "title": "Returning Visitor Rate Declining Across Mobile Devices",
      "detail": "Mobile returning visitors have dropped 19% over the 7-day rolling average (from 834/day to 675/day), while desktop returning visitors remain stable. Mobile bounce rate has simultaneously increased from 42% to 51%. This suggests a possible mobile UX regression or performance issue.",
      "suggested_action": "Audit recent mobile UX changes, check mobile page load times, and review mobile-specific conversion funnels for drop-off points.",
      "metric_references": {
        "mobile_returning_visitors_7d_avg": 675,
        "mobile_returning_visitors_7d_avg_prior": 834,
        "mobile_bounce_rate_current": "51%",
        "mobile_bounce_rate_prior": "42%",
        "desktop_returning_visitors_7d_avg": 1203
      },
      "created_at": "2026-02-17T06:00:12Z"
    }
  ]
}

Each recommendation object contains everything you need to understand and act on the insight: a unique ID for tracking, the category and severity for prioritization, a human-readable title and detail, a specific suggested action, and the exact metric references the AI used to generate the recommendation.

Understanding Recommendation Categories

The AI agent classifies every recommendation into one of six categories. Here is what each one means and what kind of insights you can expect:

Understanding Severity Levels

Every recommendation is assigned one of three severity levels to help you prioritize your attention:

Prioritization tip: Start each morning by addressing any action_required items first, then review warning items, and batch info items for weekly strategy reviews.

Acting on Recommendations

When you have reviewed a recommendation and taken action (or decided no action is needed), you can mark it as acted upon through the API:

POST /api/recommendations/{id}/acted/

This serves two purposes:

1. Tracking: You maintain a clear record of which recommendations your team has reviewed and responded to versus which were ignored. Over time, this becomes a valuable audit trail of data-driven decisions.
2. Team accountability: In multi-person teams, marking recommendations as actioned prevents duplicate effort and ensures nothing falls through the cracks.

Build a habit of reviewing recommendations daily and marking them as actioned. Over time, you will develop a clear picture of which recommendation categories drive the most value for your specific use case.

Triggering On-Demand Analysis

While the daily 6:00 AM UTC analysis covers most use cases, there are times when you need fresh insights immediately. You can trigger an on-demand analysis up to 3 times per day:

POST /api/recommendations/generate/

This runs the full analysis pipeline — data aggregation, cross-linking, trend computation, and AI analysis — and returns a fresh set of recommendations. Use this when:

• You have just launched a new campaign and want immediate feedback on early performance signals
• You have released a product update and want to monitor for engagement changes or anomalies
• You have noticed something unusual in your dashboard and want the AI’s perspective on whether it is significant
• A client meeting is coming up and you want the freshest possible insights

Viewing Historical Snapshots

Sometimes you want to understand why a specific recommendation was made. The aggregation snapshots endpoint lets you view the raw aggregated data that the AI analyzed on any given day:

GET /api/aggregation-snapshots/?date=2026-02-17

The response includes the complete aggregated dataset for that date: Web2 metrics, Web3 metrics, cross-linked data, and computed rolling averages. This is useful for:

• Verifying recommendations: Cross-check the AI’s claims against the raw aggregated numbers
• Historical analysis: Compare snapshots across dates to understand how your metrics have evolved
• Debugging: If a recommendation seems off, the snapshot shows you exactly what data the AI was working with
• Reporting: Pull specific date ranges of aggregated data for client reports or internal reviews

Tips for Getting the Most Value

After working with teams during the development of this feature, here are the practices that consistently lead to the best outcomes:

1. Install Both Web2 AND Web3 Tracking

The cross-linking of Web2 behavior with Web3 wallet activity is where the AI delivers its most unique and valuable insights. Web2-only recommendations are useful. Cross-linked recommendations are transformative. If you have not set up Web3 tracking yet, visit our integrations page to get started.

2. Check Recommendations Daily

The AI generates fresh insights every morning at 6:00 AM UTC. Build a habit of reviewing them as part of your morning routine. Even a five-minute scan of the day’s recommendations keeps you ahead of emerging issues and opportunities.

3. Act on Action Required Items First

The severity system exists for a reason. When you see an action_required recommendation, treat it as your top priority for the day. These items represent significant changes or anomalies that can have real impact on your project if left unaddressed.

4. Use the Summary Endpoint Weekly

In addition to daily recommendations, review your weekly summary to identify patterns. Are you consistently getting churn risk warnings? That signals a systemic retention issue. Are growth opportunities appearing in a specific geography repeatedly? That signals a market worth investing in deliberately.

5. Mark Recommendations as Actioned

This simple habit transforms recommendations from a notification stream into an accountability system. Over time, your action rate becomes a meaningful metric for how data-driven your team actually is.

API Reference — Quick Reference

Here is a complete reference of all AI Recommendations API endpoints:

All endpoints require authentication and are scoped to your team. For complete API documentation, visit the AI Recommendations page.