Reports

Under Reports at the top of the account (/app/reports) you see what happened in your projects and how long your customers waited. Anyone who may see projects may view them.

The overview

At the top you choose the period (7, 30 or 90 days) and a project or All projects. The comparison is always with a period of the same length immediately before. All times are in UTC.

The key figures:

• Conversations
• Solved without handover – share of conversations that managed without a handover to a person
• First reply – usually the assistant's, as median and p90 (90 % of conversations were faster)
• First human – when someone from your team replied for the first time, also as median and p90
• Thumbs up – share of positive ratings

On top of that there are the charts Conversations per day, Conversations per hour and Status (open, with the bot, waiting, nobody on duty, assigned, closed), and Why it was handed over: bot was unsure, customer asked for a human, outside hours, or a rule.

The Conversations, Visitors, Team, AI & knowledge and Frustration radar tabs show tiles such as agents, top pages, origin and campaigns, countries, devices, browsers, languages, channels and the state of the knowledge base. With Choose tiles you set which ones appear; the selection applies to the whole workspace. With Refine you filter by channel, status, country, device, browser, page and agent.

Download as CSV exports with the same filters (semicolon-separated, UTF-8). Every export is recorded in the workspace's audit log.

For very long periods the page shows the most recent conversations and says so.

Whether agents appear in the reports by name, pseudonymised or not individually at all is decided by the owner under Security (see Team, account and security).

Frustration radar

The Frustration radar tab (“Where your customers get stuck”) shows the pages of your website with the most frustration signals. Counted are rage clicks, JavaScript errors, exit intent, inactivity and purchase intent, at most once in 10 minutes per visitor, type and page. Rage clicks and errors weigh most.

For each page you see the trend compared with the previous week, how often a conversation began afterwards and how often the visitor left. Frequent topics sums up what conversations were about, from tags or the first message, without AI. This is also where you switch on the weekly email (see Availability, callbacks and notifications).

Quality (AutoQA)

Under /app/reports/quality every closed conversation gets a score from 1 to 5, without the visitor having to fill in a survey. What is scored: resolved (counts double), response time, tone and the bot's handover. A thumbs up from the visitor results in at least 4, a thumbs down in at most 2.

• Scoring happens as soon as a conversation is closed or has been idle for 6 hours. The score stays provisional for 72 hours in case the visitor comes back.
• The page shows rated conversations, the average, split by bot and human, the distribution and the criteria on average.
• Only owners and admins see the Per agent table. You can switch off per-agent scoring altogether; then it is not even stored which agent handled a conversation. The page reminds you to inform your team and, where applicable, to involve the works council.
• On paid plans you can also get an AI rating for conversations the bot handled on its own. It counts towards the AI allowance.

You also see the score of an individual conversation in the desk's context column.

Error cases

Under Error cases (/app/errors) are the JavaScript errors your visitors actually ran into on your website, grouped, with browser, page and stack. They are sorted by frequency or by when the error last occurred.

A case shows which conversations it came from and, with the visitor's consent, the point in the session recording. With an integration for GitHub or Jira you create an issue from it; recording addresses are not passed on in the process. A case expires 90 days after it was last seen.

If a visitor ran into such an error before the chat, the desk shows it in the Error case card.

A/B tests (beta)

Under A/B tests (/app/ab) you compare variants of the chat button, the chat window or both: shape, size, label, icon, colours, texts, layout and animation. Anyone who may see projects may view them; owners and admins may create, start and apply them.

The essentials in brief:

• One test runs per project at a time, with 2 to 5 variants. Variant A is the unchanged control.
• You choose how many visitors take part, on which devices, in which languages and on which pages, and what is measured: opening the chat, a message, contact details, a goal of your own or thumbs up.
• The result shows rate, difference, confidence interval, probabilities and the sample size needed, split by desktop and mobile. CSV export and Apply for the winning variant as the project brand are on the results page; if you like, both happen automatically, after 7 days at the earliest.
• Once started, variants and target group can no longer be changed. To change them, Duplicate.
• If your project requires consent, nothing is counted without it.

How to create tests and design variants is described in the Customisation part.

Outcomes

Under /app/projects/…/value (Outcomes and industry packs) you see for 7, 30 or 90 days: cases, how many of them the customer confirmed as resolved, open and reopened ones, the assistant's actions by result and the service target breaches. Download outcome report delivers the figures as JSON. The page itself points out that technical success does not yet prove customer satisfaction.

With the industry packs (online shop, software and SaaS, service provider) you add suitable test questions for the assistant's test runs.

Promo card

How often your promo card was shown and clicked in the chat is shown right next to its settings; see The promo card in the chat.

Read log

On the Agency plan reactive.chat records who opened which conversation when, in the desk, in the live console or in the AI log, and which export it was included in. Owners and admins see this under /app/projects/…/read-log and in the desk in the Read by card, with filters and CSV export. What is stored is person, conversation, time and route, no IP address. Without the Agency plan nothing is recorded.

What was changed in the workspace, on the other hand, is in the Audit log under Security; see Team, account and security.