Adoption

User segments, personalization depth, quality, sentiment, and AI-generated recommendations to grow each segment.

Adoption answers: who is using our agents, how deeply, and what should we do to grow each segment?

User segments

A donut chart segments your active users into three buckets:

Segment	Description
Power Users	High-frequency, high-engagement users — typically returning daily and using more than one agent.
Regular Users	Users with sustained but moderate engagement.
Basic Users	Light users — a few interactions and rarely returning.

The card shows total users, the count and percentage in each segment, and a trend badge against the previous period.

If the org has no user data yet, the card shows a “No user data” empty state.

Personalization

Four metrics that summarize how much the agents are learning about users:

Metric	What it measures
Instructions	Standing instructions users have given (“always reply in French”, “keep answers under 100 words”).
Preferences	Captured preferences (formats, channels, tone).
Memories	Total memories of any type.
Avg Per User	Memories divided by active users — a proxy for personalization depth.

A type breakdown is plotted below the metrics: Facts / Preferences / Instructions / Relationships, with progress bars and percentages. Hover any label for a tooltip describing what each type stores.

Quality

Four quality metrics rolled up across analyzed conversations:

Metric	What it measures
Quality Score	A 0–100 rollup that combines evaluation scores, resolution rate, and feedback into a single number. Color-coded — green ≥70, amber 40–69, red <40.
Resolution Rate	Percentage of analyzed conversations the LLM judged as resolved.
Feedback Like Rate	Share of feedback items that are likes (vs dislikes).
Sentiment distribution	A horizontal bar showing the positive / neutral / negative split across analyzed conversations.

If no conversations have been analyzed yet, this card shows “No data”.

Common patterns

A short table of recurring usage patterns the analytics pipeline detected — for example “users asking the same workflow question across multiple agents”. Each row shows the pattern, the number of associated instructions, and how many distinct users it covers.

Recommendations

Priority-ranked actions to grow adoption. Each recommendation has:

• A priority badge (high / medium / low).
• A short action string.
• An optional target segment (e.g. “Basic Users”).
• An optional metric value (e.g. “+18% personalization”) showing the projected impact.

Recommendations are advisory — the platform doesn’t apply them.

Aggregation freshness

Adoption metrics are computed from a daily aggregation. After a large analysis run or fresh memory capture, the page may show stale numbers until the next cycle. A “Waiting for aggregation” banner is shown when the most recent aggregation hasn’t completed yet.

You can force a recomputation from the Organization dashboard using the Recalculate metrics menu action.

Where to go next

Memories

The memory store the personalization metrics are built on.

Feedback

Likes, dislikes, and their categorized reasons.

Topics

What users are talking to your agents about.

Agent network

The agent fleet whose adoption you’re measuring.