The dashboard
A clear read on what your AI agent really accomplishes: who replies, how conversations end, and the value produced.
The "Analytics" dashboard answers a single question: what is your AI agent actually earning you? It does not simply count messages. It shows who really replied, how each conversation ended, how fast, and what that represents in time saved and business value. Everything reads top to bottom, from the number that matters most down to the detail that explains it.
A single control drives the page: the time period. In the top right, choose the last 7, 30, or 90 days. Every number recalculates for that window, and each key metric shows how it has moved compared with the equivalent previous period. The figures cover all of your customer conversations, across every channel.
The four key metrics
The top row is your five-second summary. Four numbers, each with its trend arrow (green when the trend is good, neutral otherwise).
| Metric | What it measures |
|---|---|
| Conversations | The total number of conversations received over the period. |
| Handled by AI | The share of conversations fully handled by the AI, with no one on your team stepping in. This is the autonomy metric: the higher it is, the more the AI is carrying the load on its own. |
| Resolution rate | The share of genuine conversations that ended well. The exact calculation is detailed further down. |
| First response | The median time between a customer's first message and the AI's first reply. The median, not the average, so it isn't skewed by a few extreme cases. |
Two useful clarifications to avoid any misunderstanding. There is no per-agent view: the dashboard aggregates all of your conversations together, it does not compare your agents against each other. And there is no channel filter: channels are shown as a breakdown (see below), but the time period remains the only control on the page.
A "Resolution rate" of 78% doesn't tell you much on its own. The arrow beside it, though, tells you whether it's rising or falling compared with the previous period. It's the movement that signals a problem taking root or an improvement worth keeping.
Who replied, and how it ended
Two cards form the honest heart of the dashboard. The first, "Who replied", splits your conversations into three based on who actually sent the replies: AI only (the AI replied and no one else did), Taken over by a human (someone on your team sent at least one reply), and No reply (no one answered). This is the measure that reveals whether the AI is really working, rather than assuming it is.
The second card, "Resolution", classifies the outcome of each genuine conversation. Three outcomes count toward the rate: Resolved (the customer got what they wanted and didn't follow up unhappy), Transferred (handed to your team), and Unresolved (the customer wasn't satisfied and showed it). The displayed rate is resolved as a share of that total.
Two categories are counted separately and deliberately excluded from the rate: Off-topic (cold outreach, spam, wrong recipient, job applications) and No follow-through (a simple "hello", an emoji, someone who leaves before any real exchange). They appear beneath the card, in gray, to stay transparent, without ever weighing on your score.
That's why "Off-topic" doesn't count as an unresolved conversation. A supplier pitching you a partnership was never a customer to satisfy. By leaving it out of the calculation, the resolution rate stays an honest measure of your real business exchanges.
Instant response and time saved
The "Instant response" card highlights the most solid number: the AI's average response time, available 24/7. When it's reliable, it adds a "X times faster than your team" comparison. That comparison only appears when your human agents have replied often enough over the period and within a realistic time frame. Otherwise, only the AI's speed is shown, with no misleading comparison.
Just below, after-hours coverage makes the value of 24/7 tangible: how many conversations started outside your opening hours (8am to 7pm, Monday to Friday), and how many the AI handled without keeping the customer waiting.
The "Time saved" card translates all of this work into hours. It estimates how long your team would have spent handling the replies the AI sent by hand, and sometimes converts that into full-time working days. It's clearly labeled as an estimate, not a stopwatch measurement: the calculation rests on assumptions you set yourself (see "Where the numbers come from").
If your team rarely replies live, or only several hours later, there's no credible human response time to compare against. In that case the dashboard prefers to show nothing rather than an inflated multiplier. The absence of "X times faster" isn't a bug: it's caution.
Volume and requests
This band explains the why behind the headline numbers. It brings together several complementary readings:
- Replies per day: the day-by-day histogram, stacking the AI's replies and your team's. At a glance you can see the activity peaks and the share carried by the AI.
- Channels: the breakdown of conversations by channel (Instagram, Messenger, WhatsApp, website, email).
- Activity hours: a day-of-week by hour grid that reveals your busy slots, so you know when the AI's availability matters most.
- Request types: the nature of the questions (general question, account or booking, action to carry out, problem to diagnose).
- Handoffs to a human: the share of conversations passed to your team, and above all why. Each reason expands to reveal the exact notes the AI wrote when handing over. This is your concrete list of things to fix.
- Most useful documents and Unanswered questions: which knowledge documents helped answer, and which questions turned up nothing. These gaps tell you precisely what to add to the knowledge base.
The handoff reasons deserve a note: they aren't guessed after the fact. They come from the note the AI itself wrote at the moment it handed the conversation to your team, grouped into broad families (missing information, request for a human, outside its scope, complaint). So you're reading the real reason, in the agent's own words.
The sales section
Collapsed by default when it's empty, the "Sales" section tells the revenue story, drawn from your sales-oriented conversations. It brings together the pipeline value, the value already converted, the qualification rate, and the number of leads.
The conversion funnel tracks the progression New, Interested, Qualified, Converted. Each stage shows the share of leads that reached or passed it, so the shape narrows naturally from one stage to the next. Along the way, the dashboard points out the stage where you lose the most people: your main leak, the one that deserves your attention first.
Finally, "Top leads" lists your highest-value contacts, ranked by amount. One click opens the matching conversation directly in the inbox, so you can step in at the right moment.
Where the numbers come from, and how to tune them
Some numbers are measured directly, others result from an analysis by the AI. The distinction matters for reading them correctly.
| Measured directly | Analyzed by the AI |
|---|---|
| Conversations, replies per day, channels | The resolution rate and the outcome of each conversation |
| Who replied, activity hours | The type of each request and the tone |
| First response time, time saved | The families of handoff reasons |
In practice, an analysis pass reads each conversation and assigns it an outcome (resolved, unresolved, transferred, off-topic, no follow-through), a request type, and a tone. It's these judgments, made with the real context of the exchange in mind rather than a simple keyword count, that feed the resolution and the request types. The analysis runs continuously in the background and only revisits a conversation once it has changed.
"Time saved" rests on two assumptions you control, under Settings then Reporting: the average time per human reply (90 seconds by default) and an agent's hourly rate, expressed in your currency. Adjust them to your reality and the estimate follows.
From this same screen, you enable the weekly email brief: a summary of this dashboard, sent automatically every Monday morning to the recipients you list. A simple way to keep leadership informed without anyone having to open the app.
Because the analysis runs in the background, a very recent conversation may show up in the volume before it has received its outcome. The resolution rate then fills in over the following minutes. Nothing to do: let the pass catch up.