> ## Documentation Index
> Fetch the complete documentation index at: https://docs.raum.am/llms.txt
> Use this file to discover all available pages before exploring further.

# Topic processing & backfill

> Configure live or on-close Chatbot classification, smart tags, eligible categories, classifier budgets, widget-specific policy, UTM allowlists, and historical backfill.

Use **Topics → Processing** to control when classification runs and what the classifier is allowed to produce.

## Enable Topics

Topics can be enabled or disabled at the organization policy level.

When disabled, the Chatbot reply path still works; the asynchronous classification/analytics layer is what stops.

## Processing mode

Choose:

* **Live after each complete Assistant turn, plus a final pass**; or
* **Only when the conversation closes**.

Live mode gives fresher topic/sentiment data during an active conversation. On-close mode reduces intermediate classification work.

## Eligible categories

Select which active Taxonomy categories the classifier may use.

The classifier is constrained to that selected set. Create/activate the taxonomy category first, then include it in Processing.

## Smart tags

Enable Smart Tags to generate up to **five normalized, PII-filtered suggestions** per processed conversation state.

You can provide optional guidance describing what makes a useful durable tag for your business.

Smart tags are exploratory labels. Promote recurring ones into Taxonomy categories when you want stable reporting.

## Insight budget

Topics has organization-level daily limits for asynchronous classification:

* classifier token budget; and
* classifier cost budget.

These limits are separate from customer-reply generation limits.

This prevents background analytics from silently consuming an unbounded model budget just because visitors had an unusually philosophical week.

## UTM allowlist

Visitor campaign context is allowlisted, not open-ended.

Configure exact allowed values for:

* UTM source;
* UTM medium; and
* UTM campaign.

An empty allowlist means **collect none** for that dimension.

## Widget-specific policy

A widget can inherit the complete organization Topics policy or use a **complete independent override**.

The override is not a partial merge.

When enabled, RAUM copies a complete policy for the widget so later organization-level changes cannot silently alter the already-published widget's topic/telemetry behavior.

A widget override can independently define Topics enabled state, processing mode, categories, Smart Tags, visitor fields, disclosures, entity mappings, and UTM allowlist.

## Backfill historical conversations

Use **Topics → Backfill** to queue classification for retained historical conversations.

The current admin flow queues up to **500 targets per backfill request**.

Choose the policy to freeze:

* **Current organization policy**; or
* a specific **published widget version**.

The selected complete policy and its provenance are frozen for that backfill. Later settings edits do not retroactively change what policy the historical job was supposed to use.

## Monitor backfill progress

Backfill records track target counts and outcomes such as completed, failed, and cancelled work.

A backfill operates only on data that is still retained. Deleted conversation content does not reappear because analytics asked nicely.

> **[Taxonomy](https://docs.raum.am/chatbot/topics.md)**
>
> Create the stable categories classification can use.

> **[Analytics](https://docs.raum.am/chatbot/analytics.md)**
>
> Review the resulting topic, sentiment, smart-tag, and visitor dimensions.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.