Analytics

Checkouts, lost revenue, channel performance, and what agents are asking for — four views, one page.

Once live, Analytics shows how your store is doing with agent checkout across four tabs. Checkouts is available on every plan; the other three unlock on the Concierge and Scale plans — without one, you’ll see a locked preview and upgrade prompt, never a blank page.

Everything here follows the store selected in the header, except the Lost Revenue summary and Agent Behavior Insights, which roll up every store on your account.

Checkouts

A live log of every checkout session an agent has started on your store. It has two modes:

Live — checkouts in progress, refreshing every ~15 seconds. Empty until one is running.
History — every past session, filterable by state, cause, date range, and whether it escalated.

Each session shows its state, source surface, start time, last activity, and cart total. Opening one shows the full detail: the cart, run time, a timeline of everything that happened (with raw payloads available as evidence), and — for anything that stalled — what happened, why, and the recommended action. An escalated session also shows whether it’s open, stalled, or resolved, and whether the buyer got a continue link back.

Lost Revenue

Prices every failed and abandoned checkout in the selected month, so you can see exactly what checkout problems are costing you rather than just that they happened.

A summary gives the entry count, the top cause, and how this month compares to last.
Causes breaks the total down, ranked by revenue or count. Fixable causes — payment failures, profile mismatches, stale catalog data, dropped escalations, platform errors, timeouts — are listed separately from checkouts a buyer simply abandoned, so the fixable list is exactly that: things you can act on.
Entries lists every priced loss individually — cause, timing, stage, status (open, acknowledged, resolved), and a recommended action — each linked back to its checkout session. Entries move from open to acknowledged to resolved as you work through them; a few show as “not priced” when the cart data wasn’t available.

Channel Dashboard

Tracks agent checkout sessions from discovery through completion and compares that channel to your regular web traffic. Pick a period (this month, last month, last 90 days, or a custom range up to 92 days) and optionally narrow to one surface — Google AI Mode, Google Gemini, ChatGPT, another agent, or unknown.

KPIs: checkouts completed, revenue, average order value, and completion rate, each compared to the prior period. Revenue is hidden, not shown wrong, if a period mixes currencies.
Funnel: sessions → reached cart → reached checkout → completed. Some sessions are estimated from checkout activity rather than directly observed; the funnel shows the confirmed/estimated split.
Daily trend: checkouts initiated, checkouts completed, and revenue, charted by day (shown in your local time zone; each point is a UTC calendar day).
Agentic vs. web: revenue, average order value, and what share of your total revenue agent checkout represents, next to the same numbers for your web channel.
A lost revenue figure for the period, linking straight through to the full Ledger.

Agent Behavior Insights

What agents are actually asking your store for — useful for deciding what to fix next, separate from what’s already failed. Same period and surface controls as the Channel Dashboard; low-volume items are hidden to avoid presenting noise as signal.

Capability gaps — capabilities agents requested during negotiation that you don’t declare, ranked by estimated revenue impact. Each gap shows request and agent counts, a confidence level (measured, partial, or not measurable — meaning no failed or abandoned checkout could be tied to it, not that it was free), and whether Sarvex can close it or it needs action on your side.
Negotiation — how often agent negotiation with your store succeeds, and why it fails when it doesn’t: unsupported protocol versions, incompatible capabilities, or a payment-handler mismatch where no handler an agent offers overlaps with what you declare.
Product demand — your most-requested products by agent lookup count, broken out by surface. This counts requests only — the protocol doesn’t yet let a lookup be correlated with a later checkout, so cart or completion rates aren’t available.
Attribute gaps — attributes agents asked for that your catalog didn’t return. Shown as not available for now — the UCP protocol has no way for an agent to name a requested attribute yet, so there’s nothing to report until that changes.