Ask the AI analyst
A language model's reading of the last 30 days against the 30 before: what changed, why it matters and what to do next.
In one click you get a short written report on your network: what is growing, what is slipping and the few things worth doing this week. The AI analyst reads the last 30 days next to the 30 before.
Before you start#
- A linked server that has seen players for a few days. Before that the page says Nothing to analyse yet, see Connect your server.
- An AI model. The dashboard can offer one, or a workspace admin connects their own provider and key in Integrations › AI (see Integrations). Without a model, the page says The AI analyst is switched off and the button is not shown.
- The AI analyst permission (
view_analyst). The report reads revenue too, so the seeded Viewer role does not have it. See Team and settings.
Steps#
Open the analyst
On Overview, click Ask the analyst.
Run it
Click Analyse my network. The page lists what it is reading (players, retention, peak hours, acquisition, the store, server health) while it works. You can leave the page: the report will be there when it is ready.
Read the report
Start with the health badge and the headline, then What stands out and What to do next.
Check the figures
Each finding links to the page where you can verify it. Check it there before you act, because a language model wrote the report.
Check that it works#
A report appears with a health badge, a headline, a summary, findings and actions, and a line saying when it was written, by which model and who asked for it. If the button is greyed out, it says when it is available again.
Going further#
Run it again#
Analyse again replaces the report.
- One run at a time per workspace.
- A waiting time between runs: six hours unless the dashboard sets another. The button says when it is available again.
- The report is written in the language of the member who ran it. Opened in another language, it says so and offers to run it again.
The report#
| Part | Content |
|---|---|
| Health | Healthy, Mixed or Needs attention, with who asked for it, when, and which model wrote it. |
| Headline and summary | One sentence, then two to four. |
| What stands out | Three to six findings, each in an area (Growth, Playtime & activity, Players who return, Where players come from, Revenue, Server health, Moderation, Community), marked Improving, Getting worse or Worth knowing, and linking to the page where it can be checked. |
| What to do next | Two to five actions, highest impact first (High impact, Medium impact, Low impact), each tied to an area. |
The model is told to use only the numbers it is given, to cite them, and to give advice specific to a Minecraft network: when to run events, onboarding, vote and comeback rewards, lag on a named server, store bundles, moderation load.
What the model sees
Aggregates only: KPIs against the previous period, revenue, retention, momentum, growth accounting, comebacks, lifecycle segments, session lengths, the busiest hours, the top hostnames and countries, performance and uptime per server, moderation totals, and your chart notes. It sees the workspace's name, timezone, currency and server names. No player names, UUIDs or IP addresses leave the dashboard.
When a run fails
| Message | What to do |
|---|---|
| An analysis for this workspace is already running. | Wait for it; the page refreshes by itself. |
| This workspace was analysed recently. Try again later. | Wait out the waiting time. |
| The model answered in a shape the dashboard could not read. | Run it again, or use a larger model. |
| The AI provider refused the request. | The message names the reason, such as a bad key or an unknown model; fix it in Integrations › AI. |
| The analysis stopped before it finished. | Run it again. |
| This workspace used its AI tokens for this month. | Buy more tokens or use your own AI provider in Integrations › AI. |
Something missing on this page? Tell us on Discord