Find the hotspot in the numbers. Find the friction in the conversation.
Which interactions are getting longer? What do customers keep asking? This scenario shows how a Decision Agent could bring service metrics together with chat, transcripts, and customer feedback, helping a team investigate the problem and evaluate practical improvements.
Illustrative contact-center scenarioBased on the illustrative scenario in our company presentation. Conversation retrieval requires configuration for each environment. The Always-on Decision Agent and approval and execution interfaces are planned for open-source release.
01Investigate the KPI change
02Read the conversation context
03Evaluate improvements
An example service-quality brief
Handle time is up 23 seconds. Where should we look first?
Compare average handle time (AHT) across service partners and inquiry types, then inspect the conversations behind the priority segment.
HEARTCOUNT / DECISION BRIEFIllustrative data
Change in overall average handle time
+23sec
Versus the reference period · presentation example
Priority segment to investigate
Partner A × personal lendingReview the metric change alongside the conversation context
01 · Find the hotspot
Which interactions are taking longer?
AHT increase by segment · bar length represents seconds added
1Partner A × personal lending+51 sec
52% of the overall AHT increase · illustrative
2Partner B × term deposits+26 sec
Comparison segment · check sample size and variability
3Partner A × card applications+9 sec
Check whether the change is meaningful and recurring
Start with personal-lending inquiries handled by Partner A.
02 · Read the context
What makes customers wait or ask again?
Retrieve transcripts, chat, and customer feedback for the same period and inquiry type. Look for steps that repeatedly require clarification.
Partner A
Personal lending
Transcripts · chat · feedback
Illustrative transcript excerpt · translated from Korean
“Let me check the documents again and get back to you.”
Document checks
Rate and limit questions
Internal handoffs
Holds and repeat explanations
Investigate whether document, rate, and limit checks are extending interactions. Recurring language provides a lead to investigate; it does not establish the cause of the delay.
What the metrics and conversations suggest
The segment contributing most to the increase warrants a closer look at repeated document and term checks.
Improvements to evaluate
Review clearer document guidance and faster access to rate and limit information against the actual workflow and supporting conversation excerpts.
How should the numbers and source conversations be checked?+
The +23 sec, +51 sec, and 52% figures come from the illustrative company presentation. They have not been recalculated from source records. Verifying the contribution requires segment volumes, mix, and a defined calculation method.
A real investigation must align the AHT definition and comparison periods, then check contact volume, inquiry complexity, channel mix, and sample variability. This scenario does not establish statistical significance or causality.
Match conversations to the right customer, inquiry, channel, and time period. Retrieve only records the user is authorized to access, and keep source excerpts distinct from the agent’s interpretation.
Service metrics meet conversation context
The numbers tell you where to look. The conversation tells you what to investigate.
A transcript summary is only part of the picture. Use defined metrics and reviewed analytical procedures to locate the problem, then inspect the relevant conversations to evaluate what could improve.
01Structured data · warehouse / CRM
Service, customer, and outcome data
Average handle time, repeat contacts, first-contact resolution (FCR), satisfaction, and post-contact conversion. Define how to compare customer groups, service partners, products, inquiry types, and channels.
02Unstructured data · conversation context
Transcripts, chat, and customer feedback
Inspect call transcripts, chat, email, and voice-of-customer (VoC) records for inquiry topics, repeated checks, complaints, and signs of churn intent. Keep the source and interaction history available for review.
↓
Connect by customer, interaction, inquiry type, and time
Align the KPI population with the retrieved records, applying access controls and appropriate masking.
↓
DECISION GRAPH
Define what to compare and when to inspect the conversation.
01
Detect the change
Use defined KPIs and a reference period to locate the segments that changed.
02
Prioritize the investigation
Consider impact, recurrence, sample quality, and what the team can actually change.
03
Retrieve relevant context
Inspect the selected segment’s interactions, customers, products, channels, and source excerpts.
04
Build the decision brief
Separate observations, interpretations, proposed improvements, and open questions.
The agent or business application connects conversation search and context retrieval. An implementation could use GraphRAG to navigate relationships among customers, interactions, and products. The public framework does not include built-in contact-log ingestion or GraphRAG.
Turn an investigation into better service
Evaluate a change. Then check whether service improved.
Shorter interactions should also mean better resolution for customers. These are possible improvements for a team to evaluate, rather than actions already taken.
01
Clarify the document-guidance script
Check whether required documents and exceptions can be explained earlier. Start with the guidance that repeatedly needs clarification.
02
Simplify rate and limit lookups
Identify avoidable screen changes and internal handoffs. Agree on a practical scope for changes to the service workflow and agent desktop.
03
Pilot a focused change
Consider a pilot for Partner A’s personal-lending inquiries, with a comparable period and inquiry mix for evaluating the result.
Track efficiency and customer experience together.
Average handle time
Repeat-contact rate
First-contact resolution
Customer satisfaction
Post-contact conversion
A shorter call is not an improvement if it creates more repeat contacts or unresolved issues. Confirm implementation separately from its impact. No measured improvement is claimed in this example.
Investigate the change before the complaints accumulate.
The Always-on Decision Agent reviews AHT, repeat contacts, FCR, and emerging feedback on an agreed schedule. It investigates the conversation context and delivers a briefing on issues worth attention, the evidence, and proposed next steps.
Approval, execution, and outcomes · Planned open-source interfaces
01
Review and approve
An owner reviews the evidence, target inquiry type, and permitted change.
02
Request the change
A connected system receives the approved script, knowledge, or workflow update request.
03
Confirm implementation
Distinguish a received request from a completed update or a failed change.
04
Evaluate the outcome
Compare service KPIs and customer experience to inform the next decision.
Available today. What needs configuration. What comes next.
This scenario builds on the public framework’s analytical capabilities. Contact-center data, conversation retrieval, and execution integrations need an agreed scope for each environment.
Public framework · available today
Reuse reviewed procedures and inspect the evidence
Consistent interaction, customer, inquiry, and time identifiers
Chat, transcript, and VoC search with source retrieval
Access controls, masking, and analytical criteria
Contact-center connectors and conversation retrieval require a separate implementation.
Planned open-source release
Investigate proactively and follow approved changes through
Always-on Decision Agent investigations and briefings
Human review and approval of proposed changes
Execution requests to CRM and service systems
Implementation status and KPI outcome checks
The Always-on Decision Agent and approval and execution interfaces are planned for open-source release and are not part of the current public framework. System connectors and approval conditions are configured during implementation.
One Decision Layer. Different data and judgment for each industry.
Manufacturing connects quality and production history. GPU operations connect workload data and service commitments. Contact centers connect service KPIs and conversation context. The shared foundation lets each organization define what to investigate and how to make a decision.
