The Reports page answers one question: is this working? It pulls together how many conversations your agent handled, how many it finished without you, how fast it replied, how often it had to hand something over, how satisfied customers were, and how much revenue it influenced, over whatever period you choose.
What it does
One page, seven sections, built from your store's real conversations rather than estimates. You choose a date range, optionally narrow to a single channel, optionally compare against the previous period, and every number on the page updates together. You can export the headline figures as a CSV whenever you want them outside the app.
Where to find it
Reports in the main navigation.
The controls at the top
Three controls sit in a bar above everything else, and they govern every number on the page at once.
Date range. Presets for Today, Yesterday, 7 days, 14 days, 30 days, 90 days, This month and To date, plus a Custom option where you set your own start and end dates on a calendar. The page opens on 30 days unless you pick something else. Everything below reflects the range you pick.
Channel. All channels by default, or narrow to chat, email, Instagram or Facebook. This is genuinely useful rather than decorative: chat and email behave differently enough that a blended average can hide a real problem in one of them. A slow first response overall is a very different story if it turns out to be email only.
Compare to previous period. A checkbox, on by default. While it is on, every figure carries a change indicator against the equivalent preceding stretch of time, so a 30-day view compares against the 30 days before it. This is what turns a number into a trend. Untick it if you only want the raw figures for the period, with no comparison at all.
Export. The button in the page header downloads the headline metrics as a CSV: conversations, resolution rate, deflection rate, CSAT, attributed revenue, median first response, median time to resolution, one-touch resolution and escalation rate. It exports what is on screen for the range and filters you currently have set.
What each section tells you
The headline figures
Five numbers, the ones worth checking weekly.
Conversations, the total your agent handled in the period.
Resolution rate, the share that reached a resolved or closed state.
Deflection rate, the share your agent resolved with no human involvement at all. This is the one that translates most directly into hours your team did not spend.
CSAT, the average satisfaction rating from post-conversation surveys, with the number of responses shown alongside so you know how much weight it carries.
Arbyn revenue, revenue from orders your agent influenced through recommendations, upsells, bundles and recovered checkouts.
Speed and efficiency
Median first response, from the customer's first message to the first reply.
Median time to resolution, from the start of a conversation to it being resolved.
One-touch resolution, the share of resolved conversations closed with a single reply.
Reopened rate, the share of resolved conversations a customer came back to. Lower is better, and a rising number here is worth taking seriously: it usually means conversations are being closed before they were genuinely finished.
Escalation rate, the share handed off to your team. Lower is better, but not at any cost; see the note below about reading it honestly.
Automated conversations, those fully resolved with no human involvement.
Commerce impact
AOV on agent-touched orders, the average value of orders your agent influenced.
Bag to checkout, the share of chat bags your agent helped build that went on to complete checkout.
Conversation lifecycle
Follow-ups sent, the proactive nudges sent when a customer went quiet, in that customer's own language.
Auto-closed, conversations resolved and closed automatically after the customer stopped replying.
Awaiting customer, conversations where your agent replied and is waiting on a response.
Messages handled, replies sent across every channel in the period.
The remaining sections
Below the tiles, the page breaks the same period down further: How Arbyn handles conversations (the flow from arrival to outcome), Volume and trends over time, Top market, AI performance, Escalations and actions, and Customer satisfaction in more depth than the single headline CSAT figure.
Top market
A breakdown of which Shopify Markets your conversations are actually coming from, so you can see at a glance whether a slow week is happening everywhere or just in one region. This is a genuinely different question from your revenue totals, which always stay in your store's own currency and are never split or converted by market, since adding together amounts from different currencies would produce a meaningless number.
Every conversation attributed here is tied to a real Shopify order, never guessed from the language a customer happened to write in. Arbyn reads the order's shipping country, billing country, currency, and locale, and only counts a conversation toward a market when those signals genuinely point to one. A conversation with no linked order, or no order carrying enough signal to place it confidently, is not silently left out of the chart, it's counted as unattributed, and the coverage percentage shown above the chart tells you what share of your conversations in this range could be placed at all.
Coverage will rarely be 100%, and that's expected, not a bug. Not every conversation ends in an order, and not every order carries a clean market signal. A low coverage percentage means exactly what it says: fewer conversations in this range had enough to attribute, not that the chart is broken. Trust the chart more as coverage climbs, and treat a very low coverage figure as a reason to read the breakdown cautiously rather than to disregard it entirely.
How it works
Every figure is computed from your own conversations for the range you selected. Change the range or the channel and the whole page recomputes; nothing is cached from a previous view or carried over from a different filter.
Comparison uses the equivalent preceding period, not a fixed window. Ask for 7 days and the comparison is the 7 days before that. Ask for a custom fortnight and it compares against the fortnight before it. So the change indicator always means "versus the same length of time, immediately before", which is the only comparison that is fair.
An empty range says so. A period with no conversations shows "No data for this range yet" rather than a page of zeroes, because zero and no-data are different facts and reading one as the other would be misleading.
Hover any figure for its definition. Every tile carries a tooltip stating exactly what it counts. This matters more than it sounds: "resolution rate" and "deflection rate" are close enough in ordinary speech to be confused, and they mean quite different things here.
Good to know
Deflection and resolution are not the same number. Resolution counts every conversation that reached a resolved state, including ones your team finished. Deflection counts only the ones your agent completed alone. Deflection is the one to watch if you are measuring workload saved.
A low escalation rate is not automatically good. An escalation is your agent correctly declining to guess. A rate of zero on a store handling refunds and returns is more likely to mean the guardrails are too loose than that the agent is brilliant.
CSAT needs volume before it means anything. The response count is shown next to the score for exactly this reason. Three responses averaging 5.0 is not a satisfaction rating, it is three people.
Attributed revenue is influence, not a claim of sole credit. It counts orders your agent touched through a recommendation, upsell, bundle or recovered checkout. There is a separate article on how attribution is decided.
Filter by channel before drawing conclusions about speed. Email replies and chat replies live on entirely different clocks; the blended median is the least informative version of that number.
The export follows your current filters. It gives you the headline figures for the range and channel on screen, not the whole history.
FAQ
Why is my resolution rate high but my deflection rate low? Because your team is finishing conversations your agent started. Both are being resolved; the difference is who finished them. If you are trying to reduce workload, deflection is the number that moves.
Does changing the channel filter change the comparison too? Yes. The comparison recalculates within whatever filters you have set, so a chat-only view compares chat against chat.
Can I see the individual conversations behind a number? Not directly from a tile. Use the Inbox with matching filters to see the underlying conversations, and Revenue attribution for the order-level detail behind the revenue figures.
What counts as a "conversation" here? A whole thread, not a message. A back-and-forth exchange of ten messages is one conversation, which is also how conversations are counted against your plan.
Why does one of my figures show a dash instead of a number? Because there is nothing to compute it from in that period. A CSAT average with no survey responses, for instance, shows a dash rather than a zero, since zero would read as "customers rated us badly" when the truth is that nobody rated us at all.
Is the export a full data dump? No, it is the headline metrics for the current range and channel. It is meant for pasting into a weekly summary rather than for analysis of raw conversation data.