Health Scores
How AccountWatch uses AI to calculate client health from Slack conversations.
Health scores are the core metric in AccountWatch. Each client receives a score from 0 to 10 that reflects the overall health of the relationship, calculated by analyzing Slack conversation patterns, sentiment, and engagement signals.
How Scores Are Calculated
AccountWatch uses Google Gemini AI to analyze messages in your tracked Slack channels. The analysis considers:
- Sentiment — Is the conversation tone positive, neutral, or negative?
- Responsiveness — How quickly does each side respond?
- Engagement — Are conversations substantive or just operational?
- Frequency — Is communication regular or has it dropped off?
- Red flags — Complaints, escalation language, deadline concerns
These signals are weighted and combined into a single 0–10 score.
Score Ranges
| Range | Label | What It Means |
|---|---|---|
| 8–10 | Excellent | Strong relationship, positive engagement, expansion potential |
| 7–7.9 | Healthy | Good relationship, normal communication patterns |
| 4–6.9 | Needs Attention | Mixed signals — sentiment dips, slower responses, or reduced engagement |
| 1–3.9 | At Risk | Negative sentiment, disengagement, or explicit concerns raised |
| 0–0.9 | Critical | Severe issues or near-total disengagement |
Update Frequency
Health scores are recalculated weekly via a scheduled job. The score reflects the aggregate analysis of all messages in the trailing period, not just the most recent conversation.
What Doesn't Affect Scores
- Message volume alone — A quiet but healthy relationship won't be penalized
- Internal team messages — Only client-facing channels are analyzed
- Automated bot messages — Routine notifications are filtered out
Improving a Score
If a client's score drops, check the client detail page for the AI-generated analysis. It will highlight specific concerns (e.g., "Client expressed frustration about delivery timeline" or "Response time increased from 2h to 8h average").
Common actions:
- Schedule a check-in call
- Address the specific concern identified by the AI
- Increase responsiveness in the channel
Next Steps
- Track health trends over time
- Set up alerts for score drops (coming soon)