Skip to main content

Chat Focus Suggestions

A focus suggestion is the chat's recommendation for which data it should work on. That scope affects much more than loading memories. The same focused scope is also used for tasks such as creating clusters, assessing hypotheses, and collecting evidence.

Why focus suggestions exist​

Teamspaces are often set up by one person who knows the taxonomy in detail. Their colleagues usually know what they want to learn from the data, but not which tags, search phrases, or time ranges best represent that request.

Focus suggestions bridge that gap. The chat analyzes your question, compares it with the available data organization, and recommends a suitable scope before doing heavier work.

That scope can include tags, a search phrase, a time range, or a combination of them. Tags are used when the taxonomy has a reliable way to represent part of your request. A search phrase is used for the parts that still need narrowing but are not cleanly represented by the taxonomy.

The goal is simple: give the chat a more relevant and consistent dataset so it can produce better results.

A better focus helps the chat:

  • Create more coherent clusters
  • Assess hypotheses against the right evidence
  • Collect stronger supporting or contradicting examples
  • Produce answers that are better grounded in related data

Example​

If you ask, What are the user pains during onboarding about passwordless login since our March release?, the chat might suggest a focus like:

  • Tags: [User pain], [Onboarding]
  • Search phrase: passwordless login
  • Time range: March until now

In this example, User pain and Onboarding are good tags because the taxonomy already captures them. Passwordless login stays a search phrase because it narrows the topic further, but may not exist as a reliable tag in the teamspace.

How tag AND and OR behavior works​

When a focus includes multiple tags, it matters whether they apply together or as alternatives.

  • AND means the chat looks for data that matches all tag conditions
  • OR means the chat looks for data that matches any of the tags

For example, a topic tag and a segment tag are often combined with AND because both should be true at the same time. When the suggestion includes multiple tags from the same tag group, the chat treats those tags as alternatives and groups them with OR, since a single memory usually matches one of them rather than all of them.

That still applies if your wording sounds conjunctive. If you ask about several brands, countries, or issue types that all come from the same group, the chat will treat them as alternative ways to narrow the dataset rather than requiring all of them on the same memory.

Search phrases are not only a fallback for when there are no tags at all. The chat can also combine tags and a search phrase in the same focus. This happens when part of your request maps well to the taxonomy, while another part needs to be narrowed through the wording of the topic itself.

How confirmation works​

The chat can misinterpret your question and suggest a focus that is too narrow or too broad. If it continued immediately, it might run heavy analysis on the wrong dataset. Instead, the chat asks you to confirm.

You can think of it like you being the manager of the chat:

  • The chat proposes the dataset it plans to use
  • You review that plan
  • You approve or adjust it

The suggestion also shows the number of matching memories so you can quickly judge whether the focus looks reasonable.

The chat does not always ask for confirmation. It may continue without interrupting when:

  • A focus is already in place
  • The suggested focus is effectively the same
  • There is no meaningful focus to add
  • The message is configured to auto-apply focus suggestions

Messages can opt into automatic focus application. In that mode, the chat still generates a focus when it finds a useful one, but it applies that focus directly instead of waiting for confirmation. For automated messages, that same setting also controls whether focus generation runs at all.

Automated messages never require a human focus decision. When they generate a usable focus, the result is applied directly.

Focus suggestions are available when the message selects Memories with a blank Focus, meaning All Memories. An existing focus or a fixed memory or cluster selection does not enable focus suggestions. Automated messages additionally require automatic focus application to be enabled.

If the chat cannot produce a usable focus suggestion, it continues with the selected sources instead of adding a default memory scope.

How Focus Works in Chat​

Focus defines which data the chat uses to answer your questions. Once applied, it becomes the working scope for the conversation until you change or clear it.

1. Selecting data sources​

  • The chat uses only the data sources selected for each message.
  • You can work with all memories, a specific Focus, selected memories, clusters, records, or external sources.

2. How Focus affects answers​

  • The chat retrieves memories relevant to your Focus and bases its answers on the available evidence.
  • Memory-backed claims include references to their sources when available.
  • If the evidence does not support a specific conclusion, the chat explains the limitation rather than making assumptions.
  • Customer quotes always use original wording, never generated or reconstructed quotes.

3. Working with memories and clusters​

  • Focus: Searches memories matching a defined topic, filters, or timeframe.
  • Selected memories: Works with the exact memories selected.
  • Selected clusters: Limits relevant actions to those clusters without expanding to the full dataset.
  • When multiple sources are selected, the chat chooses the appropriate source for each action rather than combining them automatically.
  • To calculate counts, percentages, or representative breakdowns, use a Focus rather than a fixed memory or cluster selection.

4. Creating and analyzing clusters​

  • The chat automatically quantifies newly created clusters before displaying results.
  • You can request cluster counts and original customer quotes together.
  • Displayed quotes are illustrative examples; counts reflect all matching memories.
  • Clustering, evidence collection, and hypothesis assessment finish before the chat responds.
  • Progress updates may appear while longer operations run.
  • Duplicate requests within the same message do not create duplicate ClusterSets.

5. Timeframes and account metadata​

  • Focus can include date filters, which follow the teamspace timezone.
  • Follow-up questions retain the active timeframe unless you explicitly change or remove it.
  • The chat can use relevant account attributes, such as region, industry, or company size, to explain differences in feedback.
  • These observations are based on retrieved memories, not necessarily the complete dataset. Use quantified breakdowns for statistical comparisons.

6. Changing Focus​

To explore a different topic, dataset, or timeframe, update or clear your Focus before sending the next message.

In short: Focus controls the scope of your questions, selected memories and clusters enable deeper analysis of specific evidence, and quantification provides reliable counts and breakdowns.

FAQ​

What if the suggested focus keeps excluding useful data?​

This usually means a tag isn't applied consistently enough in the teamspace.

For example, a country tag might exist for app feedback but not for survey feedback. If that tag is used for focus, the chat may unintentionally exclude survey data.

In such cases, it is often better to opt that tag out of AI focus.

Should I always approve a focus suggestion?​

No. Approve it when it reflects the dataset you want. Edit or reject it if it would lead the chat in the wrong direction.

Does focus only affect memory loading?​

No. Focus defines the dataset used for all downstream tasks, including clustering, evidence collection, and hypothesis assessment.

Focus updates and memory counts​

Manually replacing the memory selection with Focus in the frontend removes the selected memories or clusters. Backend and automatic Focus updates replace only the Focus identity, preserving selected memories, clusters, and other sources on the same message. These identities are recognized independently of their order; each memory action chooses one identity without merging sources.

A Focus memory count describes the message's full memory context only when Focus is its sole memory source. When other memory sources are selected too, manual Focus counting is skipped, and updating Focus clears the saved count. This avoids showing a partial Focus count as the total selected context.

One memory source per action​

Memory actions use one selected source per invocation: Focus, individual memories, or clusters. A message can attach several sources. The chat chooses one for each action based on the request and the scope described in its plan; no source automatically takes priority over another. Choosing a source is part of the action itself, and the backend checks that it is attached to the message.

Separate retrieval actions may use different attached sources or request different breakdowns from the same Focus. The chat aims to avoid unnecessary work while keeping each retrieval scoped to its chosen source. Repeated clustering, evidence collection, and hypothesis assessment requests still keep the first generated result, as described above.

Aggregation still requires Focus. The backend never creates a default memory scope, whether the datasource field is missing, empty, or contains only other sources. An explicitly selected empty Focus remains a valid All Memories selection; records alone do not create a memory retrieval scope.