AI insights & Ask

The console opens on what changed and what it costs

You should not have to know which chart to build. InsightClickIQ computes the numbers that matter on every screen, words them, and keeps the working one click away.

AI insights & Ask Agents Journeys Heatmaps & zones Forms Replay Segments Reports MCP Mobile Experiments Surveys Comments & sharing AI agent analytics Data platform

Digest

This week, explained, before anyone asks

Each project gets a weekly (or daily) narrative built from real KPI deltas against the previous period, the pages and sources that moved most, and the top signals by impact. It ends with one thing to do. The background loop writes every project’s digest once a day, so it is there when the console opens.

The same Overview carries an insight card: a headline, findings that link to the screen that proves them, the 3–8 numbers the narrative rests on, and suggested questions.

InsightClickIQ Overview for Northwind Outfitters with the ‘This week, explained’ digest headed ‘Reddit traffic collapsed 88%, dragging visitors down 32% and orders down 37%’, an AI insight bar, eight KPI tiles (6,696 visitors, 8,241 sessions, 870 errors, 1,042 rage clicks) and a 30-day traffic chart with a ‘v2.4 — new checkout f…’ release marker on Oct 20. — The weekly digest explains the numbers in plain language: visitors fell 2,459 → 1,674 and completed orders 168 → 106, and it ends with ‘One thing to do’. On the right it lists the costliest signals: ‘/ loads slowly: LCP p75 is 4.8s’ (~55 lost) and ‘Apply coupon’ on /cart (~7 lost).

Signals with impact

A cost on every anomaly, so you fix the right thing first

Traffic and conversion anomalies, new traffic sources, frustration hot spots, error spikes and slow pages are detected statistically every 30 minutes. Each is then sized:

Touched: the sessions and people affected, per kind (for example, sessions that hit the rage-clicked element in the last 7 days)
Lost conversions: touched sessions that did not convert × the project’s conversion rate
Revenue at risk: lost conversions × the average order value, when orders carry a value
Score 0–100 and a recommendation; the list is sorted by it
Investigate or Analyze with an agent from the signal itself

Signals page filtered to Frustration, with four cards including ‘’Apply coupon’ on /cart frustrated 78 sessions this week’ showing ~6 lost conversions, 78 sessions, $1,841 at risk, an impact score of 68, a ‘Do:’ recommendation and a rage/dead-click chart. — Each signal comes with its impact (~6 lost conversions, 78 sessions, $1,841 at risk on the coupon button: 53 rage and 110 dead clicks) and a concrete ‘Do:’ step. Explain, Investigate and Analyze with an agent sit on every card.

Ask

Ask in words, get numbers, charts and the sessions behind them

“Why did conversions drop this week?” returns an answer in plain language with the real figures, the charts behind it, a drivers breakdown of which source, device, page or country carried the change, and sessions to watch. Turn on Deep analysis and the question goes to an agent instead, which drills down step by step.

The model plans QuerySpecs against the project’s real catalog of event names, properties and top values. The engine validates and runs them; a second call narrates only the numbers that came back. Any answer can be kept as a widget or a dashboard.

Ask AI conversation answering ‘Why did conversions drop this week?’ with a written answer, an ‘Order Completed — This Week vs Last Week’ tile showing 111 orders down 33%, and a four-step purchase funnel at 11.3% overall conversion (107 of 946). — The answer opens with ‘Orders dropped 32.7% this week — 111 vs 165 last week’ and ties it to checkout errors that peaked on Oct 22 (103 errors, 25 rage clicks). The funnel under it shows where people dropped out: −574 after Product Viewed and −160 after Added to Cart.

Explain

Explain on every chart and widget

Every chart in Analyze and every dashboard widget has an Explain button. It runs the chart’s own spec, compares it with the previous period, finds the drivers for a single series and reads the result back to you, cached for ten minutes.

Seventeen surfaces work this way, from the Overview and Live to a single error, a single person or a single form, through one endpoint and one checked output shape.

Analyze page with a 30-day pageview trend line (peak on Oct 15, v2.4 marker on Oct 20) and the ‘Explain this chart’ panel open, showing an AI narrative, five tagged findings (Anomaly, Risk, Down, Win, Tip) and figure chips. — Explain reads the chart back to you: 27,488 pageviews, peak 1,757 on 15 Oct, and ‘91% of traffic has no UTM attribution’. Each finding links to a next step, such as ‘Explore traffic sources’.

Investigations

Many sessions in, a handful of themes out

Pick a filter (frustrated sessions on /cart, abandoners of a form, the sessions behind a signal) and InsightClickIQ samples up to 40, summarises each one, then groups them into themes. Every theme cites evidence as a session and an offset in seconds; evidence that does not match a real highlight is dropped. Click a chip and the replay opens at that moment.

Investigation ‘Frustrated sessions that did not convert’ (watched 20 of 295 sessions) with a ‘What the AI saw’ summary, the theme ‘Fatal ‘Place order’ JS crash blocks every checkout submission’ (20 sessions, 100% of the sample) with timestamped evidence chips, a five-step ‘What to do’ list and the sessions watched. — The AI watched 20 replays and found one theme covering all of them. Evidence chips such as ‘0:37 Place order JS error — ‘days’ undefined’ jump straight to that moment in the replay.

Dashboards

Dashboards three ways, each widget explainable

Six templates, a drag-and-drop builder, or “generate a dashboard for the checkout funnel”. A widget can be added from a sentence.

Web analytics dashboard with an AI insight bar, KPI widgets (6,695 visitors, 8,238 sessions, 12.5% bounce rate, 1m 12s average session), a 30-day visitors and pageviews chart, a Top pages table and a Referrers bar list. — A ten-widget dashboard built from the ‘Web analytics’ template. Every widget has its own AI button, and the referrers list shows (direct) 3,063, google.com 2,180 and reddit.com 815.

How it works

A grounded answer: the model plans, the engine computes, the model words

The model sees compacted results, never raw events, and never writes SQL. Insight cards are cached for 2 to 10 minutes per scope.

What you can ask

Questions this answers

Type them into Ask, give them to an agent as a goal, or send them through MCP from your own assistant.

Why did conversions drop this week?Numbers, a chart, the drivers and sessions to watch

Which traffic source grew the most in the last 7 days?A ranked table against the previous period

What is costing us the most conversions right now?The top signals by impact, with a recommendation each

Show sessions on /cart that rage-clickedA session list, replays first, ready to investigate

Generate a dashboard for the checkout funnelA saved dashboard built from your own events

Explain this chartPeriod over period, the drivers, and what to look at next

Outcomes

What teams do with it

Monday without the scramble

The digest says what moved and why, with the single costliest signal and one thing to do, before the weekly meeting.

Fix order set by money

Signals sorted by lost conversions and revenue at risk, not by whoever noticed first.

Answers anyone can check

Every figure in an answer came from a query you can open in Analyze.

Where it stops. Explanations are as good as what was captured: a conversion the site never sends as an event cannot be found. Drivers are computed for a single series; questions the catalog cannot express are answered with what it can, and say so.

Agents Session replay Segments Reports