Skip to main content

Documentation Index

Fetch the complete documentation index at: https://setup.cevro.ai/llms.txt

Use this file to discover all available pages before exploring further.

The Taxonomy page is the source of truth for everything you see on Drivers, Overview, and Feedback. It’s a three-level catalog that classifies every conversation into a structured hierarchy. 
Theme   →  e.g. "Payments"
  Need  →  e.g. "Withdraw funds"
    Problem →  e.g. "Withdrawal blocked by KYC verification"
URL: https://app.cevro.ai/insights/taxonomy

​
The Three Levels

​
Theme (top level)

The broadest grouping — operational area. Examples:
  • Payments
  • Account & Security
  • Bonuses & Promotions
  • Game-Related
  • Platform & Technical
  • Support & Service

​
Need (middle level)

A category of player intent within a Theme. Examples (under Payments):
  • Withdraw funds
  • Make a deposit
  • Update payment method
  • Reverse a transaction
Each Need carries a health indicator (green / amber / red) that summarizes how reliably new tickets are landing under it.

​
Problem (leaf level)

The specific failure mode or scenario. Examples (under “Withdraw funds”):
  • Withdrawal blocked by KYC verification
  • Withdrawal pending more than 72 hours
  • Withdrawal rejected by payment provider
  • Player can’t find withdrawal page
Problems are where most of the actionable signal lives. Each Problem holds the example phrases Cevro uses to recognize matching conversations.

​
The Page Layout

Three columns, left to right:
ColumnShows
ThemesAll Themes in your workspace with ticket counts
NeedsWhen a Theme is selected — its Needs with a health dot
ProblemsWhen a Need is selected — its Problems
Click a row to drill in; click again to deselect. The header above the columns shows summary counts (e.g. “6 themes · 28 needs · 156 problems · 142 active”). Toggle Show archived in the top-right of the card to include archived nodes in each column.

​
Health Indicator

Each Need row has a colored dot:
IndicatorMeans
GreenHealthy
AmberWatch — confidence dropping or volume spiking
RedAttention — low confidence or anomaly detected
A Need turning red usually means one of two things:
  1. The taxonomy needs updating — emerging patterns aren’t covered, so tickets land here with low confidence
  2. A real production incident — a sudden spike in volume on a previously stable Need
Problem rows show the dot only when health is not green, so the column stays scannable.

​
Working with Nodes

​
Creating a node

1

Drill to the right level

Themes are top-level. To add a Need, first click a Theme to select it. To add a Problem, first click a Need.
2

Click + at the top of the column

Opens the node form.
3

Fill in the fields

  • Name — short and descriptive (3–100 characters), unique within the parent
  • Description (optional) — short internal note
  • Example phrases — Problem only. 3–10 phrases of what players might say, one per line, each 10–500 characters. These teach Cevro what conversations belong here.
4

Save

The node becomes available immediately. New tickets can be classified into it, and flagged tickets get a chance to match it on the next classification pass.

​
Editing a node

Hover any row → click the pencil icon. Edit the same fields. If you change Example phrases on a Problem, Cevro re-learns the pattern automatically on save.
Edits to Name / Description on a legacy Problem with fewer than 3 stored examples are still allowed — the 3-phrase minimum only kicks in once you touch the examples textarea.

​
Archiving a node

Hover → click the archive icon. Archived nodes:
  • Stop accepting new ticket classifications
  • Are hidden from default views
  • Reappear when Show archived is toggled on
  • Can be restored at any time with the restore icon
Archiving is reversible — nothing is permanently deleted.
Don’t archive a node just because it has low volume. Archive when it’s been superseded by a clearer node or when it represents a flow that no longer exists in your platform.

​
Discovery Suggestions

Cevro continuously scans tickets that didn’t match any existing Problem confidently. When it spots a recurring pattern, it appears as a TaxonomySuggestion in the Discovery panel at the top of the page.

​
Running discovery on demand

From the panel header:
  • Granularity — broad, default, or granular. Lower granularity finds smaller emerging patterns at the cost of more noise.
  • Days back — 1 / 7 / 30 / 90 — how far back to scan for unmatched tickets
Click Start discovery to kick off a run. The panel streams progress through stages: queued → fetching → embedding → clustering → naming → saving → complete.

​
Acting on suggestions

Each suggestion shows the proposed name, description, parent Need, and example phrases. Available actions:
ActionWhat it does
AcceptCreates the node and reclassifies matching tickets. If no parent was inferred, opens the edit modal so you can set one first.
EditTweak name / description / parent / examples before accepting
MergeFold the cluster into an existing Theme / Need / Problem
DismissClear from queue — may resurface in a future run if the pattern persists
RejectHide permanently — won’t be suggested again
This is how the taxonomy stays current without manual sweeps.

​
Tips

  • Promote, don’t recreate. When you spot a recurring pattern, accept a Discovery suggestion or merge it — don’t hand-write it. The suggestions already have curated example phrases.
  • Three examples is the floor. Below 3 phrases, classification gets noisy. The form enforces this.
  • Watch the amber/red dots. They’re the cheapest leading indicator of taxonomy decay.
  • Drivers — the dashboard that visualizes this taxonomy
  • Feedback — service-quality clusters, separate from the Drivers tree
  • Overview — surfaces taxonomy-driven insights at the top level
  • Topic Monitoring — predecessor concept, focused on AIP coverage