AI analyst: ask your data and always-on monitoring
Ask questions about your support numbers in plain English, and get an alert — with the likely cause and a fix — when something changes.
The AI Analyst page has two parts. “Ask your data” answers questions such as “Why were more tickets held for review this week?” or “What does paid web search cost us per day?” with numbers and, where it helps, a chart. The monitor checks your account every hour and adds a card when a number changes significantly.
How answers are worked out
The analyst never writes database queries. It can only run a fixed set of reports for your account — summary and comparison with the previous period, a metric per day, a breakdown by category, channel, intent or language, why tickets were held, AI cost by day or by kind of call, quality scores, integration status, settings changes, knowledge gaps, recent tickets and experiments — and it answers from their results. Below each answer you can see which reports it looked at. It can’t change anything.
What the monitor watches
| Card | Raised when |
|---|---|
| Auto-resolve rate fell | The last 24 hours are at least 8 points below the week before, and the change is statistically significant |
| More replies wait for review / go to your team | Review or triage share rose by 8+ points (significant) |
| Average confidence fell | Down by 5 points or more |
| More answers didn’t solve the problem | Failed-answer rate (customer said it didn’t work, reopened, low rating) up 10+ points |
| Paid web searches up | At least twice the usual daily number |
| AI cost per ticket up | 1.5× the usual cost per ticket |
| Replies could not be delivered | The help desk rejected 3+ replies, three times the usual |
| Integration disconnected | A help desk or other integration is in an error state |
| No tickets received | Nothing arrived in 24 hours although you usually get 20+ a day |
Each card names the likely cause — the categories whose numbers moved most, a recent settings change, a broken integration, an auto-send threshold that looks too strict, or topics your knowledge base doesn’t cover — and offers a fix: open the draft articles for knowledge gaps, open settings, reconnect the integration, or run a regression test with one click. A problem that continues is one card that is updated, not a new card every hour.
E-mail summary
Turn on a daily or weekly summary for the account owner under “Alerts & e-mail”: key numbers compared with the previous period and every open card. Weekly summaries go out on Mondays.
- Start with “How did we do this week compared with last week?” — then ask “why?” about whatever moved. The analyst remembers the last few questions in the conversation.
- Ask for a chart explicitly (“show it per day for the last 30 days”) when you want to see a trend.
- Mark cards resolved when you have fixed the cause, and dismiss the ones that don’t matter — dismissed cards stay dismissed for that day.
- The monitor needs some volume: below about 20 tickets a day it stays quiet rather than raising false alarms.
- Asking costs a few AI calls per question; there is a daily limit per account (shown next to the chat). The hourly monitor uses no AI calls.
- Costs shown are estimates from the real token and search counts of each AI call.
Can the analyst see other accounts’ data?
No. Every report runs for the account you are signed in to; the AI can choose which report to run, never whose data.
Who can use it?
Everyone on the team can read the cards. Asking questions, acting on cards and changing the alert settings need the admin role.