Dashboard overview

The Overview tab is the first screen of every workspace and the one you should read every morning. It answers three questions in one glance: how much revenue the AI influenced, whether the machine is healthy (catalog synced, conversations flowing, rescues firing), and what the single highest-value thing to do next is. This page explains every number on it - where it comes from, what it includes, and how to read it without fooling yourself.

Where it lives

Dashboard → Overview - the workspace home. Everything described below renders here; there is nothing to configure on this page itself.
Deeper slices of the same data live in Analytics (funnels, event-level drill-downs, date ranges) - Overview is deliberately a fixed last-30-days snapshot.

AI-influenced revenue and plan-price comparison

The headline shows AI-influenced revenue for the displayed date range. The comparison divides that revenue by the listed monthly plan price when both amounts use the same currency. It is a revenue-to-plan-price ratio, not profit-based ROI, net gain or proven incremental revenue. Under the number sit four micro-stats: conversations started, rescues triggered, rescue conversion rate, and catalog sync status with the last successful sync date.

What “AI-attributed revenue” actually means

Every order that reaches Zubby is checked against recent AI activity for the same shopper. If the order lands within a 30-minute last-touch window of an AI recommendation and at least one ordered product matches that recommendation for the same store and shopper identity, the full order’s value is written to an analytics event, and the hero sums those events over the period. Two properties follow from this design:

It is last-touch and time-boxed, so a chat from last week does not claim credit for today’s order. The window is deliberately short to keep the number honest.
It is directional, not causal. Some of those shoppers would have bought anyway. The number is a management signal, never a billing input - the full method is published at /attribution-methodology, linked directly under the hero.

Influenced ≠ incremental

Full-order credit can overstate the AI’s contribution; missing identity or events can also miss influenced orders. This is neither a causal estimate nor a guaranteed upper or lower bound. Review the holdout limitations before using a comparison to inform decisions.

The KPI row and its deltas

Four cards sit under the hero:

Card	Source
Conversations	Count of conversations started in the last 30 days - every channel, not just the widget.
Cart Rescues	Count of rescue events - proactive hesitation interventions the widget fired. The subline shows how many shoppers engaged.
Rescue Conversions	Rescues whose cart later became an order, with the conversion rate (converted ÷ triggered).
Sync Status	The most recent catalog sync job - Synced, Syncing, Failed, or Not synced - with its completion date.

The arrows on the first two cards compare against the previous period of equal length - the 30 days immediately before the current window. When the previous period had no activity there is no baseline, so the card falls back to a plain subline instead of inventing an infinite percentage.

Revenue chart and quick actions

The Cumulative Revenue Recovered chart plots the same AI-attributed revenue as the hero, bucketed by day over the last 30 days. Despite the legacy “Recovered” label, this is influenced order value, not a separate measure of recovery-email lift. Check date ranges, currency and event ingestion if the chart and hero differ. Next to the chart, Quick Actions is a static set of jump links: Analytics, Cart Rescue, Recovery Emails, Product Catalog, and Settings.

The Next Move card

Below the chart, Zubby shows exactly one suggestion at a time, ranked by impact for your current workspace state, instead of a wall of tips. The ladder it walks:

Catalog not synced → run your first sync (nothing else works until the AI knows your products).
Synced but zero conversations → test the widget on your storefront.
Rescues firing but conversion under 20% → tune the rescue playbook. The current UI compares this with a hard-coded 25% comparison heuristic: 20% triggers the suggestion, while 25% is the reference shown, not a second trigger. Neither is a published, statistically reviewed Zubby benchmark.
Conversations but no revenue → boost recommendations via rules.
Conversion between 0 and 30% with 5+ chats → review replies in the AI Improver.
Everything healthy → compound wins with recovery sequences.

The card is deterministic - it reads the same numbers you see above it, so you can always reconstruct why it said what it said.

Readiness, benchmarks & the learning loop

Launch readiness - a pass/fail checklist (catalog synced, knowledge added, rules configured, and so on) so you can see setup gaps without hunting through settings.
Merchant benchmark panel - internal comparison data from the application query layer. Treat it as directional product data, not a public industry benchmark or causal performance claim.
AI Improver impact - feedback volume this week, active prompt variants, and approximate lift from the learning loop. This card renders nothing until there is data, so don’t look for it in week one.

Prediction panels (paid tiers)

On plans that include predictions, two extra panels appear: Inventory stockout (variants projected to run out within 14 days, worst first) and Churn risk (shoppers with a churn score of 60+, highest first). Both are computed asynchronously, and both are hidden entirely on tiers below the predictions entitlement - an empty panel means no at-risk rows, a missing panel means your plan doesn’t include it. See Plans & usage.

Reading it day-to-day

First: sync status. A failed or stale sync silently degrades everything downstream - recommendations, rescue, recovery.
Second: the deltas. Absolute numbers vary with traffic; the vs-previous-period arrows tell you whether the machine is trending the right way.
Third: review the Next Move. It is a deterministic heuristic based on the current workspace state, not proof that it is the highest-value action for your business.

Gotchas

A brand-new store (no conversations, no revenue, no completed sync) sees a welcome empty state instead of the dashboard - that is expected, not a loading failure.
Revenue attribution depends on orders reaching Zubby through the store connection. If orders exist but the hero reads zero, check Store details first.
The revenue-to-plan-price badge only appears when the multiple is at least 1× - below that threshold, influenced revenue is below the listed monthly plan price and the hero shows your plan cost and a nudge instead. This is not a profit or breakeven calculation.

Analytics & weekly review - deeper funnels and custom date ranges.
Cart Rescue - the live intervention behind two of the four KPI cards.
Cart & browse recovery - the post-abandonment email side.
Orders - per-order AI attribution flags.