Skip to main content

​
Overview

When an assistant acts on behalf of a customer, your team needs to know what happened: the user intent, the selected tool, the authorization state, the downstream result, and the customer-visible response. Satsuma builds observability into the connector from the first launch instead of treating it as a follow-up project.

​
What We Instrument

Sessions

Channel, account, workspace, user state, and high-level conversational context needed for support.

Tool Calls

Tool selection, inputs, outputs, latency, retries, and assistant-visible summaries.

Auth State

Connection health, scope mismatches, token refresh events, and revocation-related failures.

Outcomes

Successful completions, abandoned flows, recoverable errors, and product-specific conversion events.

​
Error Handling Principles

Satsuma designs connector failures to be understandable and recoverable:
  • Tell the assistant whether the user can retry, reconnect, change input, or contact support
  • Avoid exposing internal system details in customer-visible responses
  • Preserve enough diagnostic context for your team to debug the issue
  • Separate user errors, authorization errors, downstream service errors, and channel errors
  • Track recurring failures so the connector can improve over time

​
Analytics Views

Launch analytics typically include:
  • Most-used assistant channels
  • Top connector actions and intents
  • Conversion and completion rates by workflow
  • Auth connection success and reconnect rates
  • Latency, error rate, and downstream dependency health
  • Customer accounts using the connector most often

​
Support Workflow

1

Find the session

Support locates the session by user, account, time window, channel, or connector action.
2

Inspect the trace

The trace shows the action path, auth state, relevant downstream calls, and customer-visible result.
3

Classify the issue

The issue is grouped as product behavior, auth, channel compatibility, downstream dependency, or connector logic.
4

Resolve or escalate

Satsuma and your team use the trace to resolve the issue, tune the connector, or update support guidance.

​
Production Signals

  • Health checks are reporting for every deployed environment
  • Connector sessions and tool calls are visible in analytics
  • Auth failures are distinguishable from downstream failures
  • Support has a documented lookup path for customer issues
  • Launch stakeholders receive agreed usage and reliability reporting