Case Study ／ Retail ／ Omnichannel CX

Nationwide Chain Retailer Omnichannel AI Integration — Four touchpoints 1 Customer ID LinkingCustomer-experience redesign NPS +28pt

Nationwide hundred-store Rollout For a major chain Retailer, Generative AI delivered at the core Of customer-experience redesign. Generative AI in motion across four touchpoints: App / E-commerce / store / call Centre , Behavioural-context understanding Customer-service script / Dynamic recommendation / Staff-side assist generation. LLM-based customer understanding and Conventional recommendation model / Demand-forecast model Integration 4-Layer CX architecture , NPS +28pt / LTV +18% / E-commerce CVR 2.3pt Improvement 9-month validation project.

Duration

9 months

IntegrationTouchpoint

4Touchpoint

NPS Improvement

+28pt

LTV Lift

+18%

E-commerce CVR

+2.3pt

Challenge

Customer ID Fragmented — Four touchpointsSiloed organisation Sight Of the single Customer was lost

"Generative AI Customer experience Want to transform" and The stated executive intent was clear. On the frontline, however , Each Of the four touchpoints kept its own customer ledger, Four functions in silos . Generative AI However advanced Also , Input Customer understanding If fragmented Coherent experience Cannot be created. Precondition for leveraging Generative AI and , Organisation Customer-ID integration design Was indispensable.

Four touchpoints Separately Customer ID

App / E-commerce / Store / Call Centre Each independent Customer ID Taxonomy Maintained. Same customer 4 'Different person' and Recorded, Touchpoint Cross.Experience design Technically impossible was.

ConventionalRecommendation Limit / Precondition for leveraging Generative AIAbsent

Each touchpoint ran rule-based recommendation Operations item-by-item; without seeing cross-touchpoint purchase behaviour or customer context, one-sided pushes were frequent. Generative AI Even on deployment , Input and Customer context If fragmented LLM capability Cannot be exercised — a Structural problem Existed.

Demand forecasting and Store Operations Disjunction

Demand forecasting Head Office Owned by an expert Team, Inventory replenishment StoreDecision, E-commerce Inventory Separate system. 3 System Decision Operate independently In order to , Stock-outs / Excess inventory / Lost Opportunity Occur simultaneously Situation became chronic .

NPS Lagging metric, Frontline Cannot notice

Customer Satisfaction had deteriorated, only revealed by the semi-Annual NPS survey. Frontline What What is happeningFrom absentAs is, NPS Numbers Alone Keep dropping, Measure Priority order Cannot decideSituation was.

Channel-Customer Diagnostic

Four touchpoints Role and FragmentationStructure Customer viewpoint decompose — AI Integration guidelines Verbalised on four axes

What we did first , App / E-commerce / store / call Centre — four Touchpoints "Customer How are they Used""organisation HowHow is it operated""Both WhatFriction Whether Exists" 3 Viewpoint evaluation, AI Integration guidelines Four touchpointsPer Separate items Design .

CustomerTouchpoint

TouchpointRole (Customer viewpoint)

Current state Fragmentation

Generative AI LeverageGuideLine

ExpectationEffect

App

Daily Purchase / Product Search / Notification

OtherTouchpoint Purchase history Cannot see

LLM Integration history ContextUnderstanding

RecommendationAccuracy Lift

E-commerce

Targeted purchase / Comparison Review / Large itemPurchase

StoreRegular Treated as first-time

LLM StoreHistory Reflect Customer serviceGeneration

CVR / LTV Improvement

Store

Experience / Trial / Same-day pickup

AppNotification View information Cannot see

LLM Staff-side assist generation

Customer serviceQuality Lift

Call Centre

Inquiry / Complaint response

OtherTouchpoint Situation Inquiry time Unknown

LLM OperatorSideSummary / Proposal generation

Responsetimecompressed / CSAT Lift

CX architecture

Omnichannel CX Four layers — Customer-ID Integration layer from Observability layer Until 1 DesignLanguage Connection

Generative AI Core Placed Four-layer structure. Customer-ID Integration layer LLM Prepares input data, Experience-design layer Generative AI Behavioural boundary Definition, Personalisation-engine layer Generative AI Dynamic Customer service / Recommendation / Assist generation, Observability layer Generative AI Output quality and Customer Satisfaction Continuous monitoring . An independent design where each layer's responsibility is assigned to a primary unit (business / CX Team / AI team / data Team).

Tier 1

Customer-ID Integration layer

Customer Identity Layer

App / E-commerce / Store POS / Call Centre CRM Four touchpoints ID Integration, Generative AI to Input and Becomes"Single Customer context" Build platform.

Where deterministic keys (Member ID / E-mail / Phone number, etc.) cannot integrate, complement with probabilistic Matching using purchase history and behaviour patterns. LLM Input prompt Four touchpointsHistory Data structure prepared to embed, Synchronised with privacy design, integrating only within the customer's consent scope — a transparent design .

Integration scopeApp / E-commerce / Store POS / CRM

MatchingDeterministic ＋ Probabilistic

Tier 2

Experience-design layer

Experience Design Layer

Generative AI "What should be proposed""What Must not be proposedWhether" CX designer Definition, LLM prompt and Translated into Guardrails.

If Generative-AI personalisation becomes excessive, customers feel discomfort. Each touchpoint "LLM may propose up to here ""Does not propose beyond here" CX designer verbalises the boundary, System prompt and Translated into output Guardrails. A bespoke design that frames Generative AI's degrees Of freedom within business rules.

Primary unitCX designer ＋ Business unit

OutputPersonalisationBoundary lineWrite

Tier 3

Personalisation-engine layer

Personalization Engine Layer

Generative AI (LLM) Core. LLM Customer context understanding Top Customer-service script / Recommendation / Assist / NotificationCopy DynamicGeneration, Four touchpoints Deliver.

Generative AI (LLM) Four touchpointsShared "Customer understandingLayer" and In motion. App PushText / E-commerce Recommendation explanation / Store staff-side Proposal memo / Call Centre OperatorAssist , That Customer Context MappedDynamicGeneration . Conventional recommendation model and Demand-forecast model , LLM Context and Integrated as prompt.

Core LLMLarge-ScaleLanguage model (Customer service / AssistGeneration)

Auxiliary modelRecommendation ＋ Demand forecasting (LLM Context)

Tier 4

Observability layer

Generative AI Output quality (Boundary lineCompliance / DiscomfortRate / OperationsMetric contribution) and Customer Satisfaction Continuous monitoring. LLM LiftMove Organisation Assured.

NPS Lagging metric and Semi-annual Confirmation, CVR / LTV As leading metrics Observed daily / weekly. FurthermoreGenerative AI As Specific metrics , LLM Output GuardrailViolation rate / CustomerComplaintRate / Prompt quality Continuous monitoring, Boundary line (Tier 2) and LLM prompt Review Periodic Feedback loop performed Was built.

Leading metricCVR / LTV / Inter-touchpoint movement rate

Lagging metricNPS / CSAT / Churn Rate

Engagement TimeLine

3 Phase × 9-month — Diagnosis / Design / Phased Rollout end-to-end pipeLine

Phase 3 leads with one Of the four functions (App / E-commerce / Store / CS), After Effect validation, design expands to a full rollout across all four functions. Each phase sets an executive-decision gate, with proceed / halt / return decided by the executive committee in a structured manner .

Phase 1

2 months

CustomerMoveLineDiagnosis / Four touchpointsInterview

App unit / E-commerce unit / Store business / CS Four functions 50+ Interviews. Customer 60 journey maps Created, Four touchpoints Fragmentation Customer experience GiveImpact Quantitative and qualitative visualisation .

Key Deliverables

Four touchpointsFragmentationReport
CustomerJourney map (60 minutes)
Four functionsGovernanceMap
NPS Structural-decomposition Report

Phase 2

3 months

4-Layer CX architecture design

Customer-ID Integration layer / Experience-design layer / Personalisation-engine layer / Observability layer Four layers Co-design. CX designer PersonalisationBoundary line Four touchpointsPer Definition, AI team ImplementationScope Articulation .

Key Deliverables

Four layersReference architectureDiagram
PersonalisationBoundary lineWrite (Four touchpointsSeparately)
KPI Dashboard design (Lead/LagBehaviour)
Per-function responsibility-boundary definition document

Phase 3

4 months

Phased Rollout / Effectiveness Validation

App / E-commerce 2 Touchpoint LeadIntegration Implementation, CVR / LTV Improvement Confirmation. That After, Store POS / Call Centre CRM 2 Touchpoint AddIntegration, Four touchpointsFull rollout target. Executive committee Enterprise-wideRollout plan Approval Acquisition .

Key Deliverables

Lead PoC validation Report (2 Touchpoint)
Four touchpointsFull rollout plan
Monthly KPI-Review taxonomy
executiveexecutive-committee-approved edition / Enterprise-wideRollout plan

CX ROI Projection

Three Domains × 3-year — Phased Rollout executive language Translation Investment return

Each Of three domains has Expected Effects: personalised customer service / demand-forecasting linkage / omnichannel integration , Year-1 PoC to Year 3 Full rollout Range presented. Executive impact structured in the language shared by both CFO and CMO .

Domain

Year 1 (2 Touchpoint PoC)

Year 2 (Four touchpointsRollout)

Year 3 (Enterprise-wideRollout)

Key KPIs

Personalised customer service

Recommendation / Customer service AI

CVR +1.0pt (PoC 2 Touchpoint)

CVR +2.0pt (Four touchpoints)

CVR +2.3pt (Enterprise-wide)

CVR / Click rate

Demand-forecasting linkage

Inventory / Replenishmentoptimisation

Stock-out rate -8% (PoC Scope)

Stock-out rate -15% (leadStore)

Stock-out rate -22% (all stores)

Stock-out rate / Inventory turnover

Omnichannel integration

LTV / NPS

LTV +6% (Lead 2 Touchpoint)

LTV +12% (Four touchpoints)

LTV +18% ／ NPS +28pt

LTV ／ NPS ／ TouchpointMovement rate

Figures This engagement Executive committeePresented At ExpectedRange. Actual values will be re-evaluated from Year 2 onwards in Line with phased-Rollout results.

Results

Executive impact / Operations Effect / Governance / Rollout plan — 4 axisValidation

At project completion, NPS +28pt / LTV +18% / E-commerce CVR 2.3pt Improvement / Enterprise-wideRollout plan Executive approval Achieved simultaneously. These are 4-Layer CX architecture from derived Chain outcome.

Executive impact

+28pt

NPS Industry-average+ to — Customer experience Structural Improvement

Fragmentation Four touchpoints 1 Customer-ID integration Result, NPS 28pt ImprovementIndustry average Top tierWater-level target. The C-suite committed to CX strategy on its own initiative; the regime was established.

Operations Effect

+18%

LTV Lift — Repeat purchase StructuralIncrease

Cross-touchpoint coherent experience , Customer churn rate LoweredRepeat purchase Embedment. 3-year LTV +18% Expected, Annual Revenue Tens-Of-billions-Of-Yen scale Effect Structural Structurally secured By design.

Governance

BoundaryExplicit

PersonalisationBoundary line Organisation Articulation

"AI What should not be proposed" Four touchpointsPer CX designer verbalises, Governance document and organisation Embedment. Conversion from personality-driven decisions to CX governance at the level Of organisational capability .

Rollout plan

Approved

Four touchpointsAllRollout plan — the following fiscal yearBudget Secured

2 Touchpoint PoC Outcome Based on Four touchpointsEnterprise-wideRollout plan Formulation, Executive committee Approval. the following fiscal year Enterprise-wideRollout budget Secured .

CX Design Principles

5 CX design principles — Retail / Service / Small-ticket finance ReproducibleStructuralDecision

Of the lessons distilled in this engagement, Consumer-facing business (Retail / Restaurants / Hotel / telecom / Small-ticket finance, etc.) Five reproducible structures for CX-AI design .

Customer-ID Integration Not by technology GovernanceProblem

Customer ID Is a Governance Issue

Unless the organisational structure that treats the same customer as 'four different people' is transformed, technical ID integration does not work. App unit / E-commerce unit / Store business / CS Responsibility boundary Redefinition , Comes before technical design.

Decide upfront the 'line that must not be crossed' for personalisation

Define What AI Will NOT Do First

What AI may proposerather than, What should not be proposed First Definition . CX designers articulate the boundary line in clear language, and the AI team implements within that scope . Boundary line Absent Customer discomfort and Brand damage Generate.

Cannot separate demand forecasting from store Operations

Demand and Operations Are Inseparable

When demand forecasting is owned by head Office and stores decide independently, forecast and operations dissociate and lost Opportunity is generated. At forecast-model design , Store Operations Constraints Embed required as input.

NPS is a lagging metric; pair it with CVR as a leading metric

Track Both Lagging and Leading Indicators

Looking at NPS alone , Improvement / Deterioration Sensed Half-year lag results. CVR / Inter-touchpoint movement rate are observed daily as leading metrics; a mechanism to forecast NPS Volatility is required.

AI customer service is not 'human substitute' but 'human-decision auxiliary'

Augment Human Staff, Don't Replace

Store staff / Call-Centre operator Substitute Aim and , FinalCustomer experience Quality settles. AI Customer understanding Immediately presents, Humans make the final decisionHybrid design , Highest customer Satisfaction Generate.

Nationwide Chain Retailer Omnichannel AI Integration — Four touchpoints 1 Customer ID LinkingCustomer-experience redesign NPS +28pt

Customer ID Fragmented — Four touchpointsSiloed organisation Sight Of the single Customer was lost

Four touchpoints Separately Customer ID

ConventionalRecommendation Limit / Precondition for leveraging Generative AIAbsent

Demand forecasting and Store Operations Disjunction

NPS Lagging metric, Frontline Cannot notice

Four touchpoints Role and FragmentationStructure Customer viewpoint decompose — AI Integration guidelines Verbalised on four axes

Omnichannel CX Four layers — Customer-ID Integration layer from Observability layer Until 1 DesignLanguage Connection

3 Phase × 9-month — Diagnosis / Design / Phased Rollout end-to-end pipeLine

CustomerMoveLineDiagnosis / Four touchpointsInterview

Key Deliverables

4-Layer CX architecture design

Key Deliverables

Phased Rollout / Effectiveness Validation

Key Deliverables

Three Domains × 3-year — Phased Rollout executive language Translation Investment return

Executive impact / Operations Effect / Governance / Rollout plan — 4 axisValidation

NPS Industry-average+ to — Customer experience Structural Improvement

LTV Lift — Repeat purchase StructuralIncrease

PersonalisationBoundary line Organisation Articulation

Four touchpointsAllRollout plan — the following fiscal yearBudget Secured

5 CX design principles — Retail / Service / Small-ticket finance ReproducibleStructuralDecision

Customer-ID Integration Not by technology GovernanceProblem

Decide upfront the 'line that must not be crossed' for personalisation

Cannot separate demand forecasting from store Operations

NPS is a lagging metric; pair it with CVR as a leading metric

AI customer service is not 'human substitute' but 'human-decision auxiliary'

This engagement Linkage DX Strategy Service

Your firm's Generative-AI leverage , Customer experience Form that reasons backward..

Nationwide Chain Retailer Omnichannel AI Integration — Four touchpoints 1 Customer ID LinkingCustomer-experience redesign NPS +28pt

Customer ID Fragmented — Four touchpointsSiloed organisation Sight Of the single Customer was lost

Four touchpoints Separately Customer ID

ConventionalRecommendation Limit / Precondition for leveraging Generative AIAbsent

Demand forecasting and Store Operations Disjunction

NPS Lagging metric, Frontline Cannot notice

Four touchpoints Role and FragmentationStructure Customer viewpoint decompose — AI Integration guidelines Verbalised on four axes

Omnichannel CX Four layers — Customer-ID Integration layer from Observability layer Until 1 DesignLanguage Connection

3 Phase × 9-month — Diagnosis / Design / Phased Rollout end-to-end pipeLine

CustomerMoveLineDiagnosis / Four touchpointsInterview

Key Deliverables

4-Layer CX architecture design

Key Deliverables

Phased Rollout / Effectiveness Validation

Key Deliverables

Three Domains × 3-year — Phased Rollout executive language Translation Investment return

Executive impact / Operations Effect / Governance / Rollout plan — 4 axisValidation

NPS Industry-average+ to — Customer experience Structural Improvement

LTV Lift — Repeat purchase StructuralIncrease

PersonalisationBoundary line Organisation Articulation

Four touchpointsAllRollout plan — the following fiscal yearBudget Secured

5 CX design principles — Retail / Service / Small-ticket finance ReproducibleStructuralDecision

Customer-ID Integration Not by technology GovernanceProblem

Decide upfront the 'line that must not be crossed' for personalisation

Cannot separate demand forecasting from store Operations

NPS is a lagging metric; pair it with CVR as a leading metric

AI customer service is not 'human substitute' but 'human-decision auxiliary'

This engagement Linkage DX Strategy Service

Related Case Studies

Your firm's Generative-AI leverage , Customer experience Form that reasons backward..