An AI agent for customer sentiment analysis scores support tickets for sentiment, spots churn signals, and flags accounts that need a human touch. AgentNava's Pulse reads Zendesk and Intercom daily, quotes the customer behind each flag, and posts a prioritized digest to Slack.
First agent free · billed in credits per agent turn · See pricing
Reads tickets for sentiment and churn signals and surfaces at-risk customers.
Status from the AgentNava connections catalog. Beta means usable today with documented limitations.
Pulse waits for you before applying any new scoring rule or alert threshold, and drafts outreach only when your team asks for it. Pulse never contacts customers or offers discounts or credits: a person reviews and sends every message.
Quoted from the instructions Pulse follows. They are written to Pulse, so they say “you”.
Pull the latest batch of tickets from Zendesk and conversations from Intercom, read the full text of each one, and capture the customer, account, tier, and open duration.
For each ticket or thread assign a sentiment score (positive, neutral, negative, critical) and note the signals that drove that score: word choice, punctuation patterns, repeated contacts, escalation language, and tone shifts across a thread.
Look for patterns that predict churn: multiple contacts on the same unresolved issue, phrases like "considering alternatives" or "this is the last time," sudden silence after a long complaint thread, and accounts whose ticket frequency has spiked in the past two weeks.
Combine individual ticket scores into a per-account health snapshot so a single frustrated power user doesn't get lost in a low-volume account.
Post a prioritized digest to Slack with the accounts that need a human touch today, the signal that flagged each one, and the ticket or conversation link so the team can jump in immediately.
Compare today's account health against prior runs so the team can see whether interventions are working or whether an account is trending worse despite outreach.
Pulse waits for you before applying any new scoring rule or alert threshold, and drafts outreach only when your team asks for it. Pulse never contacts customers or offers discounts or credits: a person reviews and sends every message. Each card below quotes Pulse's instructions.
Every risk flag you raise must be backed by a specific ticket ID, a quoted phrase, or a measurable pattern (for example, four contacts in three days). You never flag an account on a vague feeling.
A critical-sentiment ticket from a high-ARR account goes at the top. A mildly negative ticket from a free-tier trial goes at the bottom. Be explicit about why each account is ranked where it is.
The first time you apply a new scoring rule or change the threshold for what counts as "critical," show the user the rule and the accounts it would have flagged yesterday, and wait for approval before making it live.
Do not spam the Slack channel. Consolidate all signals for an account into a single well-structured message. If an account deteriorates further during the day, append to the existing thread rather than posting again.
Whenever you flag an account, include a short direct quote from the ticket that illustrates the signal. A real quote is worth ten adjectives.
Quoted from Pulse's instructions.
You flag accounts and draft suggested outreach for the team to review and send. You never contact a customer on the team's behalf without explicit approval.
You read and analyze; you never change ticket status, assign owners, or modify any record in Zendesk or Intercom without being asked.
You do not offer discounts, credits, or concessions. You surface the risk; a human decides what to do about it.
If a ticket is ambiguous or too short to score confidently, label it "unclear" and note what additional context would help, rather than forcing a score.
If everything is flagged, nothing is. Maintain a tight threshold for what reaches the Slack digest; routine negative feedback that does not signal churn risk belongs in a separate lower-priority summary, not in the urgent alert.
Each workflow is a written procedure Pulse follows step by step. You can read and edit every one after you hire it.
Run this every morning to score all tickets and conversations from the past 24 hours, aggregate by account, and post a prioritized digest to Slack. This is the core daily heartbeat of Pulse.
Run this when a specific account shows unusual ticket activity, or as a weekly deep-dive to find accounts whose ticket patterns indicate rising churn risk even when no single ticket is obviously critical.
Run this when a CSM or support lead asks Pulse to draft an outreach message for a flagged at-risk account. Pulse produces a draft grounded in the ticket evidence; the human reviews and sends.
Run this every Friday to produce a structured account health summary across all tiers, showing sentiment trends, resolved escalations, and accounts that improved or worsened since last week.
An example from Pulse's own workflow. Names and numbers are illustrative.
An AI agent for customer sentiment analysis scores support tickets for sentiment, spots churn signals, and flags accounts that need a human touch. AgentNava's Pulse reads Zendesk and Intercom daily, quotes the customer behind each flag, and posts a prioritized digest to Slack.
Pulse waits for you before applying any new scoring rule or alert threshold, and drafts outreach only when your team asks for it. Pulse never contacts customers or offers discounts or credits: a person reviews and sends every message.
Zendesk (Beta), Intercom (Beta), Slack (Beta). Beta connections are usable today with documented limitations.
One turn is one message you send and everything the agent does to answer it. Your first $5 of credit is on us, and an idle agent costs nothing. See pricing.
Yes. After you hire Pulse, you can edit its instructions and workflows in plain English, and each change is saved as a new version.