The three dimensions
Three AI-derived dimensions are live today. Each one has its own article explaining every value and the exact logic behind it:
Engagement Quality
How much genuine attention did a session show? High, Medium, or Low — scored from active time, browsing depth, interactions, and conversions.
Session Intent
Why did the visitor come? Research, Comparison, Buying, or Support — derived from the pages visited and the events triggered.
Page Categories
What role does each page play? Home, Product, Pricing, Checkout, Blog article, and more — classified from URL, metadata, and page content.
How classification works
Every value is computed by datataste's classification engine while your events are ingested. The engine evaluates dozens of weighted signals — which page categories a session visited, how much active time accumulated, which events fired, whether a conversion happened — and assigns a value only when the evidence clears a confidence threshold. When the evidence is too thin, the honest answer is unknown rather than a guess.
Values are computed by versioned scoring rules, not by a generative model. The same session always produces the same classification, every result is reproducible and explainable, and none of your traffic data is sent to an external AI provider.
The engine also adapts to your business model. datataste automatically detects whether a property behaves like an ecommerce store, a SaaS product, a B2B lead-gen site, a publisher, or a support portal — and weights the signals accordingly. A checkout page is a strong buying signal in a store; a long article read counts for more on a publisher site.
Scoring rules are versioned. When a rule set improves, pages are gradually re-classified with the new version — so classifications get better over time without you doing anything.
Enabling AI dimensions
AI dimensions are opt-in and ship disabled. Enable each dimension individually per property under Settings → Dimensions in the dashboard.
Classification starts the moment you enable a dimension. Historical sessions and events are not backfilled, so charts fill with classified data from that point forward.
Reading the values
Three things are worth knowing when you read AI-dimension charts:
- Unknown is a verdict, not an error. It means the session or page was evaluated and the evidence didn't clearly support any value. A two-second visit is honestly unknown.
- Empty means disabled. If a dimension was switched off when an event arrived, the value simply stays empty for that event — that is not the same as unknown.
- Charts focus on classified traffic. Most AI-dimension charts leave out unknown and unclassified entries, so a chart's total can be lower than your total session count.
Manual overrides
Every automatically derived value can be corrected by hand — a page's category, a session's intent, or its engagement quality. Manual values always win: the engine never overwrites a manual correction, even when scoring rules are updated later. Fixing the occasional edge case is safe; the automation won't fight back.