AIR Engine™

The problem with Direct traffic

Every attribution report has a Direct bucket. It grows larger the more channels you run, because dark social shares, cross-device revisits, AI halo traffic, and email forwards all arrive with no referrer header and no UTM parameters. Standard attribution writes them off as unknown.

The AIR Engine does not write them off. It examines the context and circumstances of those sessions, compared against the known-channel traffic already in your database, and produces a structured account of where they probably came from and what to do about it.

How it works

When you open the AIR tab, the engine collects all Direct first-touch sessions in the selected date range and evaluates them against 20 pattern rules across five finding categories. Each rule fires when the session data on a page path meets a specific threshold. No data leaves your server. No external API is called. Every result is deterministic: the same data produces the same findings every time.

This is not generative AI. There is no LLM involved in producing a finding. Every output is rule-based arithmetic against your own session database. The word "deterministic" is accurate and intentional.

The five finding categories

Confidence scoring

Each finding carries a confidence score from 0 to 100, derived from five measurable factors already in your session database: whether the pattern rule fired cleanly, how much trackable traffic the same page receives, how far the Direct share of that page exceeds the site average, how consistently Direct sessions follow known-channel sessions on the same page within a short window, and whether Direct volume on that page is rising.

Attribution Health Score

At the top of the AIR tab, before the findings, a single 0-100 score summarises the overall quality of the Direct traffic picture for this installation. It is derived from the classified session ratio, the severity distribution across findings, and channel diversity. Green is Strong (80+), amber is Fair (60-79), red is Critical (below 60).

This is separate from the Insights Health Score, which measures overall attribution setup quality. The AIR Health Score measures specifically how much of the Direct bucket can be accounted for.

Managing findings

Each finding card has a Mark as reviewed action. Reviewed findings move to a collapsed section at the bottom of the report showing the tier, classification, page path, and the date it was reviewed. Findings can be restored to the active list at any time. The reviewed state persists across sessions.

Plain-English reporting

The Copy prompt button formats the full AIR report, including all finding data, into a self-contained prompt on your clipboard. Paste it into any LLM to get a plain-English summary prioritised by confidence, ending with the top three recommended actions. An Open in Gemini button does the same in one step when the report is short enough for a direct URL.

The LLM reads the engine's output. It does not produce or improve the underlying findings. The analysis is already complete before the LLM sees it.

Requirements

• Referrer Attribution Pro v1.15.0 or later
• At least 30 days of session data for reliable results. The engine runs on any data volume, but the timing and trend signals become more meaningful over a longer window.
• Set the date range selector to All time for the most comprehensive evaluation.