Layer 0 · Corpus Ingestion

UK Retailer
Intelligence Framework

Intelligent Context Framework (ICF) — corpus-grounded analysis of agent–customer chat sessions, the SOP library, training manuals, SME interactions and journeys.
Firstsource · Intelligence that Operates
Chat Transcripts
5,443
Agent–customer chat sessions across retail CS journeys — deliveries, returns, products & more
Training Materials
30+
Induction decks — order lifecycle, returns, tone of voice, DPA/GDPR, fraud, carriers
SOP / Process Manual
63
CS procedures — shopping, products, deliveries, returns, payments, store, legal & compliance
Contact Categories
10
General Enquiries, Returns & Refunds, Deliveries, Products & Sizing + 6 more queues
Journey Types
10+
Disputed delivery, How to return, Late delivery, Stock check, Faulty product & more
Qualitative Signals
6
Live chat conversations, ops team huddle notes, agent coaching logs, QA calibration sessions, customer verbatims, escalation narratives

Processing Pipeline

1
Parse — tokenise the 16-queue SOP library & 5,443 chat sessions
2
Extract — identify intents, journeys, agent patterns & exceptions
3
Map — build node-edge graph from entity relationships
4
Flag — classify gaps (Process / SOP / Training / Product)
5
Synthesize — assemble ICF with advisory, KB & transcript layers
Layer 1 · Operational Analysis

Operational Analysis

Journey Distribution

Contact Category Share

Quality vs Benchmark

How Contacts Resolve (derived)

Gap Profile

Top Operational Risks

Issue Volume by Domain

Transcript Insight — Agent Patterns

Issue
Exception
System
Decision
Resolution
Layer 2 · Gap Intelligence

Gap Register — 41 Findings

3
CRITICAL
DPA inconsistency, Lost-in-transit refunds, DPA-fail recovery
16
HIGH
Disconnect follow-up, Spam triage, Goodwill matrix, Confirmed-late vs LIT
22
MEDIUM
Duplicate refund tasks, cancellation window, wrong-store collection, multibuy diagnostics
5,443
SESSIONS IMPACTED
Across 5,443 chat sessions and the 16-queue SOP library analysed
All Types
Process
SOP
Training
Product
All Severity
Critical
High
Medium
Layer 3A · SOP Decision Tool

SOP Reviews — Owner Decisions

Total 0 Incorporate 0 Ignore 0 Pending 0 Selected 0
Layer 3B · Capability Development

Training Modules & Performance Appraisals

Training Modules

Performance Appraisals

Layer 3C · Strategic Advisories

Process & Product Advisories

Process Advisories

Product Advisories

Priority × Effort Matrix

Layer 4 · Agent Knowledge Base

Knowledge Base — UK Retailer

Articles
ICF Co-Pilot
Select a category to view articles
Layer 4 · Data Deep Dive

Analytical Deep Dive — Charts & Insights

Slice the 5,443-session corpus by domain, severity, volume and remediation load.
Powered by Firstsource Intelligence

AI-Driven Interventions

Four intelligent pillars working across the customer journey — from proactive alerts to autonomous decisioning.

4
AI Pillars
~20%
FCR Lift · illustrative
~30%
AHT Reduction · illustrative
Illustrative
Modelled on corpus volume
Pillar 01
Proactive Engagement
Detect risk signals before the customer contacts us — delivery exceptions, lost-in-transit scans, failed refunds, returns nearing the 28-day window — and intervene with timely updates.
Pillar 02
Smart Routing
AI concierge classifies intent and complexity in real time, routing each customer to the optimal channel — self-service, AI chat, or specialist.
Pillar 03
AI Co-Pilot
Real-time agent assist during live chat — intent classification, knowledge surfacing, compliance guardrails and auto-wrap — all contextual.
Pillar 04
Agentic AI
Autonomous reasoning over ICF context — observe, recall, probe, reason, govern, recommend — with human-in-the-loop governance.

Intervention Overview

Each pillar addresses a distinct layer of the customer interaction — from pre-contact engagement through to post-contact intelligence. Together they form a closed loop that learns from every conversation.

How the Pillars Connect

1. Proactive Engagement detects risk signals from account data, transaction patterns and ICF context — and sends timely alerts before the customer needs to call.

2. Smart Routing classifies inbound intent and complexity in real time using the ICF entity graph, routing each customer to the right channel first time.

3. AI Co-Pilot assists the agent during live conversations with contextual suggestions, knowledge articles, compliance checks and automated after-call work.

4. Agentic AI reasons autonomously over ICF context for complex decisions — complaint risk scoring, escalation routing, proactive outreach — with human governance.

Projected Impact

66%→80%
First Contact Resolution
9:09→6:30
Avg Handle Time
3.9%→2.0%
Escalation Rate
7.8→8.6
CSAT Score

Pillar 01 · Proactive Engagement

ICF detects risk signals from account data and transaction patterns — and triggers proactive alerts before the customer needs to call.

ICF Signal Feed

Pillar 02 · Smart Routing

AI Concierge classifies intent, complexity and vulnerability in real time — routing each customer to the optimal resolution channel.

Low · Self-Service

Medium · AI Chat + Handoff

High · Specialist Agent

--:--:--ReadyAI Concierge initialised · Domain Harness loaded · ICF connected

Pillar 03 · AI Co-Pilot

Live agent assist during a customer chat — intent classification, knowledge surfacing, compliance guardrails and automated after-call work.

Co-Pilot actions: 0

Live Chat Transcript

Co-Pilot Actions

Pillar 04 · Agentic AI

Autonomous reasoning over ICF context — a transparent chain-of-thought trace for a complaint escalation risk decision.

Reasoning Trace

ICF Context Read

Entity Context
Customer: repeat shopper · 3yr history
Prior contacts: 2 in last 14 days
Flag: repeat contact / frustration
Order: standard delivery · £95
Channel preference: Live chat
Last contact: refund not yet received (unresolved)
Domain Harness Match
Rule: DH-COMP-007 — "Repeat complaint + vulnerability flag + unresolved prior → escalate to complaints specialist within 2 minutes"
Confidence: 0.91
Trust tier: Human-in-the-Loop

Recommended Outcome

Escalation riskHIGH (0.91)
Recommended actionEscalate to complaints specialist
SLA targetResponse within 2 min
Estimated cost avoided£— (repeat-contact & complaint prevention)
Trace latency1.6s total

ICF Intelligence Architecture

Five-layer stack — ICF sits at the centre, binding operational sensors below to domain intelligence above.

Layer 5 — AI-Native Operations
Agents (illustrative — varies by client) e.g. delivery-exception outreach agent, returns deflection assistant, complaint triage bot
Illustrative
Layer 4 — Domain Intelligence
Domain Harness — Skills, Playbooks, Guardrails, Eval e.g. disputed-delivery playbook, DPA two-gate verification, consumer-rights compliance rules
Core OS
Layer 3 — Organisational Context
Intelligent Context Framework (ICF) — entity graph, gap register, decision traces, SOP alignment
Core OS
Layer 2 — Operations Intelligence
Sensors — Process, Task, Conversation, Knowledge, Doc e.g. chat transcripts, call recordings, CRM notes, QA calibrations
Core OS
Layer 1 — Raw Data & Transactions
Systems of Record + Channels (illustrative) e.g. order platform, returns portal, tracking layer, case system, live chat
Illustrative

Why ICF is the Centrepiece

Below ICF — Operations Intelligence

Sensors observe what is actually happening on the ground: process flows, desktop activity, conversations, documents, knowledge use. For this operation this means every chat transcript, complaint narrative, and agent desktop action is captured in real time.

Above ICF — Domain Harness

Encodes 25 years of Firstsource domain expertise — skills, playbooks, guardrails — as the baseline of what should happen. For this retailer: disputed-delivery flows, returns & refund rules, DPA verification, and the SOP/PR procedure library.

ICF Binds the Two

ICF stores the live state of every interaction, decision, exception, and outcome — closing the should-vs-did gap and feeding the learning loop. It connects what the agent actually did with what the playbook prescribed, surfacing gaps before they become complaints.

Top & Bottom — Illustrative

Agents and Systems of Record vary by client. The core value — Layers 2, 3 and 4 — is the intelligence stack that Firstsource owns and operates.

Outcome Forecast

Projected impact across key metrics — each improvement attributed to specific intervention pillars.

First Contact Resolution
66%
80%
Proactive + Co-Pilot + Routing
Average Handle Time
9m 09s
6m 30s
Co-Pilot + Routing
Escalation Rate
3.9%
2.0%
Agentic AI + Smart Routing
Repeat Contact Rate
~22%
12%
Proactive + Co-Pilot
CSAT Score
7.8
8.6
All Pillars
Annual Cost Saving
Baseline
Illustrative
Proactive + Routing + Agentic