No runs yet. Use “Run the agents” below.
01
The cause
What caused the drop
02
The money
Where the money sits
03
What to fix first
Two rankings, two tracks
04
Can we trust the scores?
A data-quality problem in the scores
✦
The deliverable
The leadership brief
Written by the agent team, then audited line by line by fact_check.py, which is plain Python.
▶
Live
Run the agent team
Runs python -m nps_insights_agent.main --trace on your model. Takes about a minute. Watch the three phase-one agents start at the same moment.
OrchestratorAgentdirector · calls every specialist as a tool · returns the brief word for word
1 three calls at once
ScoreAnalystAgentNPS, trend, drill-down
VoiceOfCustomerAgentthemes and quotes
ChurnRiskAgent₹ at risk and data quality
2 then
ActionPlannerAgentfixes ranked two ways, callbacks
3 finally
ReportWriterAgentthe one-page brief
fact_check.pyplain-Python auditor
waiting
running
done
failed
$ press “Run the agents” to start; the trace streams here line by line
?
Under the hood
Code computes, agents reason
Tools do all the counting and maths in plain Python. Agents read the results, decide what matters and explain it.
ScoreAnalystAgent
Replaces the analyst building the NPS pivot table.
get_nps_overviewget_nps_by_segment ★VoiceOfCustomerAgent
Replaces the analyst reading 536 comments.
get_comment_themesChurnRiskAgent
The finance partner and the data-quality sceptic.
get_revenue_at_riskfind_score_comment_mismatchesget_nps_by_segment ★ActionPlannerAgent
Replaces the CX lead building the action plan.
estimate_fix_impactget_callback_listReportWriterAgent
Replaces the Head of CX writing the CEO note.
no tools: synthesis only
OrchestratorAgent
The director assigning the work in three phases.
the five agents above, each via
.as_tool()★ Shared tool, cached so the second agent gets the answer for free. Reports travel between phases on a hand-off board as forms with exact keys.