QA Analytics

Enterprise Bot

AI Analytics Platform · Contact Center Intelligence

Turning conversations into an intelligent layer for contact-center performance.

QA Analytics is an AI-powered intelligence layer that helps contact centers understand what is happening across conversations, AI agents, human agents and customer journeys — processed live or in aggregated batches, depending on how each organisation works. It goes beyond reporting to explain what is happening, why, and what to improve next.

Role: UX Architect · Product Strategy · UX Research · AI Experience Design
Team: Product · Engineering · AI/ML · Customer Success · Sales · Leadership
Domain: Enterprise AI · Contact Center Intelligence

00 — Key highlights

Where I contributed

Industry research

Banking, insurance, travel and retail — how each consumes conversation analytics and how often.

Live vs aggregated

Shaped the product around multiple analytics cadences instead of assuming everyone needs real-time.

Analytics IA

Structured conversation, channel, AI agent, human agent, topic, intent, escalation and resolution data.

AI agent performance

Where AI succeeds, fails, hands over to humans — and what to change in its configuration.

Human agent intelligence

Transfers, escalations, resolution behaviour, feedback and coaching needs.

AI improvement signals

Linked analytics to prompt, tool, knowledge, configuration and training changes.

Data Analyst AI

Ask questions in natural language and get a consolidated report — no dashboard filtering.

Configurable analytics

Teams define their own metrics, KPIs, scorecards, tabs and resolution measures.

01 / 08

01 — Opportunity

Enormous conversation data. The challenge is knowing what to do with it.

Every interaction contains signals. Traditional analytics tells teams what happened.

QA Analytics/ Command center

Conversations

48.2k+12.4%

AI containment

71.8%+3.1%

Avg. QA score

84.6-0.8

Escalations

2,914-5.2%

Avg. handle time

4m 52s-11s

First contact res.

78.3%+1.9%

Conversation volume30 days

Handled
Resolved
Escalated

Volume vs. handle timeBy hour

CSATTarget 4.5

4.2of 5.0

Detractors 9%Promoters 64%

Intent mix

Billing31%
Account access22%
Delivery18%
Refunds16%
Other13%

Channel load

Peak load heatmapHour × weekday

Topics: volume × sentimentBubble = growth

QA dimensionsAI vs. human

AI resolution funnel

Started100%
Intent detected92%
Answered by AI78%
Resolved by AI64%
CSAT ≥ 451%

Sentiment by channel

Emerging topicsSorted by volume

Topic	Owner	Volume	Sentiment	Trend
Card declined at checkout	Payments	3,412	-0.42	+18%
Password reset loop	Identity	2,870	-0.31	+9%
Late delivery window	Logistics	2,105	-0.27	-4%
New plan pricing	Growth	1,644	+0.12	+42%
Refund status check	Finance	1,210	+0.34	-12%

What customers ask for
Products they’re interested in
Where they get stuck
Why conversations escalate
How well AI resolves
Where agents need support
Outdated knowledge
Failing tools & workflows
Prompts that need work

What happenedWhy did it happenWhat should we do next?

02 — Research question

Does every contact center need real-time?

Assumption

“Contact centers need real-time analytics.”

Reality

A railway operation may monitor live. A bank gets more value from aggregated analysis across millions of interactions.

How might we design analytics that adapts to how different organisations actually consume conversation intelligence?

03 — Industry research

Cadence, metrics and priorities vary by industry

Research workspace with conversation transcripts and sticky notes clustered by industry

Banking

Customer intent
Transactions
Resolution
Escalations
AI performance
Agent performance
Compliance

Insurance

Customer queries
Policy conversations
Resolution quality
Escalation patterns
Agent effectiveness
Recurring issues

Travel & Retail

High volume
Live monitoring
Channel performance
Customer intent
Enquiries
Escalations
Demand trends

01 / 03

04 — Analytics cadence

73% aggregated. Not everyone needs live.

Instead of one universal experience, the architecture supports multiple cadences and customer-specific requirements.

Conversation dataHow does the organisation consume insights?

Aggregated73%
Trends · Performance · Customer behaviour
Live15–18%
Contact center health · Active conversations · Escalations
TraditionalRemaining
Existing KPIs · Operational reports

Analytics cadence — research signal across customers. One universal dashboard wouldn’t fit.

05 — Product model

Four connected intelligence layers

This became the foundation for the product’s information architecture.

Conversation dataAnalytics engine

Conversation intelligence

Topics & intentEscalationsResolutionCustomer behaviour

AI agent intelligence

AI resolutionAI takeoverPrompt / tool issuesAI improvement

Human agent intelligence

Agent performanceTransfersTraining needsFeedback

Operational intelligence

Channel performanceVolumeTrendsOperational health

Product model — four connected intelligence layers fed by one analytics engine.

06 — UX architecture

From raw conversations to recommended action

The analytics engine is the source of intelligence; different interfaces consume it.

Ingest

Conversation sourcesIngestion layerProcessing mode

LiveLive processing

AggregatedBatch processing

Analyse

Analytics engine

Intelligence

ConversationAI agentHuman agentOperational

Insights

Act

Dashboards & reports

KPIsTrendsDrill-down

Data Analyst AI

Natural language query

Queries the analytics engine

Recommendations

AI improvementHuman trainingOperational action

End-to-end architecture — the analytics engine is the single source of intelligence; dashboards, the Data Analyst and recommendations all consume it.

07 — Conversation intelligence

What customers are actually talking about

ConversationClassification

TopicIntentChannelSentiment / outcome

Trend analysisInsight

Every conversation is classified, then rolled into trends.

Topics

Recurring themes across conversations.

Intent

What customers are trying to accomplish.

Channel

Behaviour across chat, voice, email and more.

Resolution

Whether conversations were resolved successfully.

Escalation

When and why conversations move from AI to humans.

Customer behaviour

Emerging topics and shifts in demand.

01 / 06

08 — AI agent intelligence

Analysing the AI agent itself

AI agent conversations

Resolved by AIHuman takeoverEscalationLow confidenceTool / knowledge issue

Why did it happen?

ImprovePromptKnowledgeToolAgent configHuman training

AI agent intelligence — from counting conversations to diagnosing what to improve.

From

“How many conversations did the AI handle?”

“How well is the AI performing, and what should we improve?”

09 — Human agent intelligence

Where people need support — not a ranking

And where the system itself needs improvement.

Aisha M.

Billing · Tier 2

Strong on refunds+4 this week

Topic handling

Refunds96%
Billing disputes92%
Escalations84%

Daniel K.

Tech support · Tier 1

Coaching suggested-3 this week

Topic handling

Troubleshooting68%
Account access81%
Escalations59%

Priya R.

Orders · Tier 1

Consistent resolver+2 this week

Topic handling

Order tracking93%
Returns85%
Delivery issues79%

Illustrative QA rating cards: an overall score plus topic handling scores that point to coaching, not a ranking.

Conversations handled
Handling patterns
Transfers
Escalations
Resolution
Customer feedback
Missed opportunities
Training
Coaching

Agent insightCoachingTrainingProcess improvement

10 — Analytics to action

Analytics shouldn’t stop at reporting

AnalyticsIdentify patternUnderstand causeAction required?

AI issueUpdate prompt
Knowledge issueUpdate knowledge
Tool issueUpdate tool
Agent issueHuman training
Process issueImprove workflow
No actionKeep monitoring

Measure again — back to analytics

Measure → Understand → Improve → Measure again.

11 — Data Analyst AI

Just ask a question

A separate AI layer that gives a conversational interface to the analytics engine.

Data AnalystQA Analytics engine

We released a new feature last month. Are customers enquiring about it?

Yes. Enquiries about “Split payments” rose sharply after launch in week 4 and now make up 6.8% of all billing conversations.

41% ask how to enable it
27% report it missing in the app
Sentiment is neutral, trending positive

Ask about your conversations…

Traditional

Open dashboard
Find metric
Apply filter
Select period
Compare
Interpret

Data Analyst

Ask a question
AI understands intent
Analytics engine
Analyse data
Consolidated insight
Follow-up questions

12 — Configurable analytics

Humans define what good looks like

Metrics
KPIs
Scorecards
Dashboard tabs
Resolution definitions
Analytical dimensions
Reporting views

OrganisationDefine analytics modelAnalytics engineCustom insights

AI provides intelligence. Humans define what intelligence matters to their organisation.

13 — Dashboard & KPIs

Representing complex analytical information

KPI cards

Conversations, resolution, AI resolution, human takeover, escalation, feedback.

Trends

How performance changes over time.

Distribution

Volume across channels, topics, intents and outcomes.

Comparison

AI vs human, channel vs channel, period vs period, team vs team.

Drill-down

From an executive signal into the conversations and causes beneath it.

01 / 05

01Executive KPI
02Trend
03Category
04Topic / intent
05Conversation
06Root cause
07Recommended action

Signal → explanation → evidence → action.

14 — Executive intelligence

Different questions for leadership

Executive overview → operational analysis → conversation-level evidence.

Business performance

How many conversations are we handling?
Which channels are growing?
Which departments drive demand?

AI performance

How much work is AI handling?
How often does AI resolve?
Where does AI need humans?

Customer experience

What are customers asking about?
Where do they get stuck?
Which topics are rising?

Operational performance

How are agents performing?
Where are escalations rising?
Which teams need attention?

01 / 04

15 — Information architecture

Many perspectives, one calm navigation

QA Analytics

Overview
ConversationsConversation details
AI AgentsAI performance
Human AgentsAgent performance
Channels
Topics & IntentTopic analysis
OperationsOperational trends
Reports
Data Analyst

Information architecture — navigation follows the questions users ask, not internal data structures.

16 — Research → decisions

Research wasn’t a phase

The loop revealed which analytics were universal and which needed to stay configurable by industry.

Customer researchIndustry patternsAnalytical needsProduct hypothesesUX explorationPrototypeCustomer feedback

Feedback loops back to hypotheses → product direction → analytics architecture

17 — Collaboration

One coherent product and experience architecture

Product

Opportunities, prioritisation and analytical capabilities.

Engineering

Data structures, processing, dashboard behaviour and constraints.

AI / ML

Classification, agent behaviour and the Data Analyst architecture.

Customer Success

Customer requirements, feedback and operational use cases.

Sales & Partnerships

Industry-specific needs and customer demonstrations.

Leadership

Alignment with the AI-powered contact-center strategy.

01 / 06

Reporting as a shared workflow

Download reportsShare with teamsOperational reviewsProduct, AI & training decisions

18 — Outcome

From dashboards to an intelligent contact-center layer

Broader analytics model

Live and aggregated analytics, so each industry consumes intelligence at its own cadence.

AI + human intelligence

AI-agent and human-agent performance in one model — automation and people, together.

Actionable AI improvement

Signals that inform prompts, tools, knowledge, agent configuration and training.

Conversational analytics

The Data Analyst removes the need to navigate dashboards for every question.

Configurable intelligence

Organisations define the KPIs, scorecards and views that matter to them.

Stronger positioning

From a reporting feature to an AI intelligence layer for contact-center operations.

01 / 06

19 — The bigger idea

What happened?

What happened, why did it happen, and what should we improve?

“Just ask the system.”