Citizen Service Assistant

Category: Citizen & Service User Engagement

Status: ✅ Production Ready

Deployment: Cloud | Hybrid | On-Premise

Overview

AI-powered multi-channel assistant that handles citizen and resident queries across web chat, phone, SMS, and WhatsApp. Provides instant responses to common questions, routes complex queries to the right teams with full context, and integrates seamlessly with your existing CRM and service systems.

Problem Statement

The Challenge

Contact centres across UK local authorities and housing associations face:

Rising demand: 40% increase in contacts over 5 years
Channel proliferation: Citizens expect support via web, phone, email, SMS, social media
Staff pressure: High turnover, training costs, burnout
Out-of-hours gaps: Limited service outside 9-5
Inconsistent service: Quality varies by advisor knowledge
High costs: £8-15 per contact handled by human advisor

Current State

Typical citizen journey:

Citizen calls or visits website
Waits in queue (average 8-12 minutes)
Speaks to advisor who may need to transfer or research
Receives answer (if first-time resolution) or callback promise
May need to call back to chase

Pain points:

35% of calls are simple information requests
60% first-time resolution rate
Citizens frustrated by wait times
Staff spending time on repetitive queries
After-hours queries go unanswered until next day

Impact

With Citizen Service Assistant:

Time savings: 70% of routine queries answered instantly (no wait)
Cost reduction: £100k-300k annual savings (mid-sized authority)
Availability: 24/7/365 service across all channels
Satisfaction: 4.3/5 average citizen satisfaction score
Staff wellbeing: Advisors focus on complex, rewarding cases
Consistency: Same quality answer regardless of time, channel, or advisor

Solution

What It Does

The Citizen Service Assistant provides an intelligent conversational interface that:

Core Capabilities:

Answers questions about council/housing services instantly
Handles common transactions (report issues, book appointments, check status)
Searches knowledge base across policies, procedures, FAQs
Routes complex queries to the right team with full conversation context
Supports multiple languages and accessibility needs
Learns from advisor corrections to improve over time

Channels Supported:

Web chat (embed in website/portal)
Voice (phone integration)
SMS/Text messaging
WhatsApp Business
Microsoft Teams (for internal staff queries)
Email (auto-response and triage)

Key Features:

Contextual understanding - Remembers conversation history, previous interactions
Multi-turn dialogue - Handles complex queries requiring clarification
Transaction handling - Can create cases, update records, send forms
Handoff to human - Seamless transfer with context when needed
Analytics dashboard - Query trends, satisfaction scores, knowledge gaps

How It Works

┌─────────────────────────────────────────────────────────────┐
│                    Citizen Contact                           │
│  Web Chat | Phone | SMS | WhatsApp | Email | Social         │
└────────────────────────┬────────────────────────────────────┘
                         │
                  ┌──────▼──────┐
                  │  Channel    │
                  │  Adapters   │
                  └──────┬──────┘
                         │
┌────────────────────────▼────────────────────────────────────┐
│               Conversation Engine (AI)                       │
│  • Intent Recognition  • Entity Extraction                   │
│  • Dialogue Management • Context Tracking                    │
└────────┬──────────────┬──────────────┬──────────────────────┘
         │              │              │
    ┌────▼────┐   ┌─────▼─────┐  ┌────▼────┐
    │Knowledge│   │ CRM/HMS   │  │ Service │
    │  Base   │   │ Systems   │  │ APIs    │
    └─────────┘   └───────────┘  └─────────┘

Example Conversations

Example 1: Bin Collection Enquiry

Citizen: "When are my bins collected?"
Bot:     "I can help with that. What's your postcode?"
Citizen: "SW1A 1AA"
Bot:     "Your next collections are:
         • Black bin (general waste): Friday 1 Dec
         • Green bin (recycling): Friday 8 Dec
         • Food caddy: Every Friday

         Would you like a calendar reminder?"
Citizen: "Yes please"
Bot:     "Great! What's your email address?"

Example 2: Repairs Request (Housing)

Tenant:  "I've got a leak in my bathroom"
Bot:     "I'm sorry to hear that. Is this an emergency
         (water pouring out) or non-emergency?"
Tenant:  "It's dripping from the ceiling"
Bot:     "That sounds urgent. Let me create a priority
         repair for you. Which property is this?
         [Shows tenant's properties if multiple]"
Tenant:  [Selects property]
Bot:     "I've created repair #12345. A plumber will
         call you within 2 hours to arrange access.

         In the meantime:
         • Turn off water at stopcock if you can
         • Move belongings from under the leak

         Need anything else?"

Example 3: Complex Query (Escalation)

Citizen: "Why was my planning application refused?"
Bot:     "I can see application #2023/4567 for an
         extension at 10 High Street was refused on
         15 Oct. The reasons are quite detailed.

         Would you like me to:
         1. Email you the full decision notice
         2. Connect you with a planning officer
         3. Explain the appeals process"
Citizen: "Connect me to an officer"
Bot:     "Of course. Let me transfer you now with all
         the details of your application..."

[Transfers to planning team with full context]

Target Sectors

Local Government
Housing Associations
Third Sector
Cross-sector (HR/PM)

Sector-Specific Benefits

Local Government:

Handle 40-50% of contact centre volume automatically
Unified experience across departments (waste, planning, council tax, etc.)
Integration with GOV.UK Pay for payments
Support for multiple languages in diverse communities
FOI/transparency - publish conversation analytics

Housing Associations:

24/7 tenant support for repairs, rent, ASB
Reduce out-of-hours emergency call outs (better triage)
Improve tenant satisfaction scores
Integration with repairs scheduling
Support vulnerable tenants with voice and simplified interfaces

Third Sector:

Support beneficiaries and service users
Answer donor/volunteer queries
Reduce burden on small teams
Multi-language for diverse communities
Integration with case management

Integration

Compatible Systems

Primary Integrations (Choose one or more)

System	Type	Integration Method	Capabilities
Civica CRM	Council CRM	REST API	Read/Write cases, contacts, service requests
NEC Housing	HMS	SOAP/REST	Tenant data, repairs, rent accounts
Microsoft Dynamics 365	CRM	Dataverse API	Full CRUD on entities
Salesforce	CRM/NPSP	REST API	Cases, contacts, accounts
Firmstep CX	Digital platform	REST API	Forms, applications, payments

Supporting Integrations

System	Purpose	Method
GOV.UK Notify	SMS/Email	API
Twilio/MessageBird	SMS/WhatsApp	API
Microsoft Teams	Internal queries	Bot Framework
Azure AD / Google Workspace	SSO	SAML/OAuth
SharePoint	Knowledge base	Graph API
Outlook/Gmail	Email channel	Graph/Gmail API

Service-Specific Integrations

Local Government:

Waste collection systems (Bartec, Whitespace, Yotta)
Planning systems (Idox, NEC Assure)
Revenues & Benefits (Civica, NEC, Capita)
Payment gateway (GOV.UK Pay, Capita Pay360)

Housing:

Repairs management (Oneserve, Totalmobile, DRS)
Rent accounting (built into HMS)
ASB case management (Civica, bespoke)

Integration Architecture

┌─────────────────────────────────────────────────────────────┐
│              Your Existing Infrastructure                    │
├──────────────┬──────────────┬───────────────┬───────────────┤
│ Civica CRM   │ Waste System │ Planning      │ SharePoint    │
│              │              │ Portal        │ (KB)          │
└──────┬───────┴──────┬───────┴───────┬───────┴───────┬───────┘
       │              │               │               │
       │              └───────┬───────┘               │
       │                      │                       │
       └──────────────────────┼───────────────────────┘
                              │
                       ┌──────▼──────┐
                       │ API Gateway │ ← Security, rate limiting
                       │  (Optional) │
                       └──────┬──────┘
                              │
┌─────────────────────────────▼─────────────────────────────┐
│              Citizen Service Assistant                     │
│                                                            │
│  ┌──────────────────────────────────────────────────┐    │
│  │   Conversation Engine                            │    │
│  └──────────────────────────────────────────────────┘    │
│                                                            │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐               │
│  │Knowledge │  │Analytics │  │ Admin    │               │
│  │   Base   │  │Dashboard │  │ Portal   │               │
│  └──────────┘  └──────────┘  └──────────┘               │
└────────────────────────────────────────────────────────────┘
                              │
      ┌───────────────────────┼───────────────────────┐
      │                       │                       │
┌─────▼─────┐          ┌──────▼──────┐        ┌──────▼──────┐
│  Website  │          │   Phone     │        │ SMS/Email   │
│  Chat     │          │  (Twilio)   │        │ (Notify)    │
└───────────┘          └─────────────┘        └─────────────┘

API Requirements

Required APIs from your systems:

Customer/Tenant Data (READ)
- Search by name, postcode, property, tenant ref
- Get contact details, account status
- Authentication: API key or OAuth 2.0
Service Request Management (READ/WRITE)
- Create new cases/requests
- Update case status
- Retrieve case history
- Authentication: API key or OAuth 2.0
Knowledge Base (READ)
- Search documents and FAQs
- Retrieve article content
- Can be SharePoint, Confluence, or dedicated KB

Optional APIs (Enhanced features):

Scheduling/Appointments (READ/WRITE)
- Check availability
- Book appointments
- Send reminders
Payment (READ/WRITE)
- Check account balance
- Generate payment links
- Record payments
Property/Asset Data (READ)
- Property details
- Service schedules (bins, inspections)

Data Flow

Inbound (what Assistant receives):

Service catalogues and eligibility rules
Knowledge base articles and FAQs
Customer/tenant records (on-demand via API)
Case/request status updates (via webhooks)
Service schedules (waste, inspections)

Outbound (what Assistant sends):

New service requests/cases
Case updates and notes
Customer feedback scores
Analytics and conversation logs
Escalation triggers to human advisors

Data Residency:

All conversation processing in UK Azure regions (UK South primary, UK West backup)
No personal data sent outside UK
Conversation logs encrypted at rest (AES-256)
PII automatically redacted from training data

Technical Architecture

Components

Frontend (Channel Adapters):

Web chat widget (React, WCAG 2.1 AA compliant)
Voice gateway (Twilio integration)
SMS/WhatsApp connector
Email processor
Microsoft Teams bot

Backend:

API layer: Node.js/Express or Python/FastAPI
Conversation engine: Dialogflow CX / Azure Bot Framework / Custom
Queue system: Azure Service Bus / AWS SQS
Cache: Redis (session state, frequently accessed data)

AI/ML:

NLU Model: Gemini (via Vertex AI) or Azure OpenAI
Intent classification: Custom fine-tuned models for local gov/housing
Entity extraction: Location, dates, reference numbers, names
Sentiment analysis: Detect frustration/urgency
Knowledge retrieval: Vector search (Vertex AI Search / Azure AI Search)

Data Store:

Conversation history: PostgreSQL or MongoDB
Knowledge base: Vector database (Pinecone / Azure AI Search)
Analytics: BigQuery or Azure Synapse
Cache: Redis

Infrastructure

Compute:

Kubernetes (Azure AKS or AWS EKS) for containerized deployment
Auto-scaling: HPA based on message queue depth
Minimum: 2 nodes (HA)
Typical: 3-5 nodes for medium deployment
Resources per pod: 2 CPU, 4GB RAM

Storage:

Conversation data: 100GB (100k conversations/month)
Knowledge base: 50GB
Logs: 20GB/month

Networking:

Ingress: Application Gateway / ALB with WAF
TLS termination: Let's Encrypt or enterprise CA
API Gateway: Azure API Management / AWS API Gateway (optional)
VPN: ExpressRoute / Direct Connect for hybrid

Security

Authentication:

Admin users: SSO via Azure AD / Google Workspace / ADFS
Service accounts: Managed identities (Azure) / IAM roles (AWS)
API authentication: OAuth 2.0 client credentials flow
Citizens: Anonymous (web/SMS) or authenticated (portal users)

Encryption:

In transit: TLS 1.3 minimum
At rest: AES-256 (managed keys)
Key management: Azure Key Vault / AWS KMS
Secret management: Key Vault / Secrets Manager

Audit:

All conversations logged with metadata
PII redacted from analytics
Tamper-proof logging (append-only)
Retention: 6 years (standard local gov requirement)
GDPR: Right to erasure process documented

Network Security:

WAF protecting public endpoints
DDoS protection (Azure DDoS / AWS Shield)
Private endpoints for backend services
NSG/Security Groups limiting traffic

Deployment

Prerequisites

Infrastructure:

Azure subscription OR AWS account OR on-premise Kubernetes
API gateway (recommended for production)
VPN/ExpressRoute if hybrid deployment
SSL certificates for custom domain

Access:

Admin access to CRM/HMS for API setup
API credentials/keys from integrated systems
Firewall rules approved for outbound API calls
Azure AD / Google Workspace admin (for SSO)

Data:

Data classification completed (Official, Official-Sensitive)
DPIA completed if processing sensitive personal data
Information governance sign-off
Knowledge base content prepared (FAQs, policies)

Deployment Options

Option 1: Cloud-Hosted on Azure UK (Recommended)

Why Azure UK:

UK South & UK West regions (data residency)
G-Cloud 14 listed services
NHS DSPT alignment for health/social care
Crown Hosting Data Centres

Quick Deploy:

# Clone repository
git clone https://github.com/uk-public-sector-ai/citizen-service-assistant
cd citizen-service-assistant

# Install Azure CLI and authenticate
az login
az account set --subscription "Your Subscription"

# Configure deployment
cd terraform/azure-uk
cp prod.tfvars.example prod.tfvars
# Edit prod.tfvars with your settings

# Deploy infrastructure
terraform init
terraform plan -var-file="prod.tfvars"
terraform apply -var-file="prod.tfvars"

# Deploy application
cd ../../
./scripts/deploy-azure.sh --environment prod

# Configure integrations
./scripts/configure-integrations.sh

What gets deployed:

Azure Kubernetes Service (AKS) cluster
Azure Database for PostgreSQL
Azure AI Search (knowledge base)
Application Gateway + WAF
Azure Monitor + Log Analytics
Key Vault for secrets

Estimated deployment time: 45-60 minutes

Option 2: AWS London

# Similar process using AWS CLI
cd terraform/aws-london
terraform init
terraform apply

cd ../../
./scripts/deploy-aws.sh --environment prod

What gets deployed:

EKS cluster (eu-west-2)
RDS PostgreSQL
OpenSearch (knowledge base)
ALB + WAF
CloudWatch
Secrets Manager

Option 3: Hybrid Deployment

Ideal for organisations with data sovereignty requirements or existing on-prem infrastructure.

Architecture:

Cloud: Conversation engine, channels, AI models
On-premise: Customer data, CRM integration, audit logs

# Deploy cloud components
./scripts/deploy-hybrid-cloud.sh --cloud azure

# Deploy on-prem gateway
./scripts/deploy-hybrid-gateway.sh --on-prem

Network setup:

ExpressRoute (Azure) or Direct Connect (AWS)
Private connectivity between cloud and on-prem
On-prem API gateway for secure access to CRM

Option 4: Fully On-Premise

For organisations with strict data policies or existing Kubernetes infrastructure.

Requirements:

Kubernetes 1.25+
Persistent volume storage
Load balancer
GPU nodes (optional, for better performance)

# Deploy to existing cluster
kubectl create namespace citizen-assistant
kubectl apply -f k8s/production/

# Configure
kubectl create secret generic api-keys \
  --from-literal=crm-api-key=YOUR_KEY \
  --from-literal=sms-api-key=YOUR_KEY

# Deploy
helm install citizen-assistant ./helm \
  --namespace citizen-assistant \
  --values values-production.yaml

Configuration

Environment Variables:

# config/.env.production

# Deployment
DEPLOYMENT_REGION=uk-south
DATA_RESIDENCY=uk-only
ENVIRONMENT=production
LOG_LEVEL=info

# Organisation
ORG_NAME="Your Council Name"
ORG_TYPE=local-government  # or housing-association, charity
TENANT_ID=your-unique-id

# AI Configuration
AI_PROVIDER=vertex-ai  # or azure-openai
AI_MODEL=gemini-pro
CONFIDENCE_THRESHOLD=0.75
FALLBACK_TO_HUMAN_THRESHOLD=0.5

# Integrations
CRM_TYPE=civica  # civica, nec, dynamics365, salesforce
CRM_API_URL=https://your-crm.com/api/v1
CRM_API_KEY=***
CRM_TIMEOUT_MS=5000

WASTE_SYSTEM_API=https://bartec.yourcouncil.gov.uk/api
PLANNING_SYSTEM_API=https://idox.yourcouncil.gov.uk/api

# Channels
ENABLE_WEB_CHAT=true
ENABLE_PHONE=true
ENABLE_SMS=true
ENABLE_WHATSAPP=false
ENABLE_EMAIL=true

TWILIO_ACCOUNT_SID=***
TWILIO_AUTH_TOKEN=***
TWILIO_PHONE_NUMBER=+44...

GOV_NOTIFY_API_KEY=***

# Knowledge Base
KB_SOURCE=sharepoint  # sharepoint, confluence, filesystem
SHAREPOINT_SITE_URL=https://yourorg.sharepoint.com/sites/kb
SHAREPOINT_CLIENT_ID=***
SHAREPOINT_CLIENT_SECRET=***

# Analytics
ANALYTICS_ENABLED=true
ANALYTICS_RETENTION_DAYS=2190  # 6 years

# Security
SESSION_SECRET=***
JWT_SECRET=***
ENCRYPTION_KEY=***

Integration Configuration:

# config/integrations.yaml

civica:
  enabled: true
  base_url: ${CRM_API_URL}
  timeout: 5000
  retry_attempts: 3
  endpoints:
    customer_search: /customers/search
    create_case: /cases
    get_case: /cases/{id}
  rate_limit:
    requests_per_minute: 60

knowledge_base:
  source: sharepoint
  index_schedule: "0 2 * * *"  # Daily at 2am
  categories:
    - waste
    - planning
    - housing
    - council-tax
    - benefits
  auto_suggest_threshold: 0.8

channels:
  web_chat:
    widget_position: bottom-right
    greeting: "Hi! I'm the council's virtual assistant. How can I help?"
    office_hours_only: false

  phone:
    provider: twilio
    fallback_number: "+441234567890"  # Human backup
    ivr_enabled: true

  sms:
    provider: gov-notify
    character_limit: 600
    opt_out_keyword: "STOP"

handoff:
  triggers:
    - low_confidence: 0.5
    - explicit_request: true
    - escalation_keywords: ["complaint", "complaint", "legal", "solicitor"]
  routing:
    default_queue: general-enquiries
    rules:
      - condition: intent == 'housing_repair'
        queue: repairs-team
      - condition: intent == 'planning_application'
        queue: planning-team

Testing

Pre-deployment Testing:

# Unit tests
npm test

# Integration tests (requires test environment)
npm run test:integration

# API connectivity tests
./scripts/test-connectivity.sh \
  --crm-url ${CRM_API_URL} \
  --crm-key ${CRM_API_KEY}

# Load testing (simulate 100 concurrent users)
./scripts/load-test.sh --users 100 --duration 300

# Security scan
./scripts/security-scan.sh

# Accessibility test (WCAG 2.1 AA)
npm run test:accessibility

Post-deployment Validation:

# Health check
curl https://your-deployment/health

# Smoke tests
./scripts/smoke-test.sh --environment prod

# End-to-end test conversations
./scripts/e2e-test.sh

Validation Checklist:

Operation

Monitoring

Health Checks:

Endpoint: https://your-deployment/health
Frequency: 30 seconds

Expected response:

{
  "status": "healthy",
  "timestamp": "2024-01-15T10:30:00Z",
  "components": {
    "api": "healthy",
    "database": "healthy",
    "redis": "healthy",
    "crm_connection": "healthy",
    "ai_service": "healthy"
  }
}

Key Metrics:

Metric	Target	Alert Threshold
Response time (p95)	<500ms	>2s
API success rate	>99%	<95%
Conversation completion rate	>85%	<75%
Handoff rate	<20%	>35%
User satisfaction	>4.0/5	<3.5/5
Uptime	>99.5%	<99%

Dashboards:

Operations Dashboard (real-time)
- Active conversations
- Response times
- Error rates
- Queue depths
- System health
Business Dashboard (daily)
- Conversation volume by channel
- Top intents
- Containment rate (resolved without human)
- Satisfaction scores
- Cost per conversation
Knowledge Dashboard (weekly)
- Most searched topics
- Knowledge gaps (low-confidence queries)
- Article effectiveness
- Suggested improvements

Alerts:

# alerts.yaml
alerts:
  - name: High Error Rate
    condition: error_rate > 0.01  # 1%
    severity: critical
    action: page-on-call

  - name: Slow Response
    condition: p95_response_time > 2000  # 2s
    severity: high
    action: email-team

  - name: CRM Integration Down
    condition: crm_health != 'healthy'
    severity: critical
    action: page-on-call

  - name: High Handoff Rate
    condition: handoff_rate > 0.35  # 35%
    severity: medium
    action: email-team

Maintenance

Daily Tasks (Automated):

Health check verification
Log aggregation and analysis
Backup verification
Security event review

Weekly Tasks:

Performance review
Knowledge base effectiveness
User feedback review
Training data curation

Monthly Tasks:

Security patching
Conversation audit (quality)
Knowledge base update
Model retraining (if needed)

Quarterly Tasks:

DR test
Penetration test
Accessibility audit
Business review (ROI, satisfaction)

Updates:

Security patches: Within 48 hours of release
Minor updates: Monthly (new features, improvements)
Major updates: Quarterly (with full regression testing)

Backup & Recovery:

Database: Daily backups, 30-day retention
Configuration: Version controlled (Git)
Knowledge base: Backed up with database
RTO (Recovery Time Objective): 4 hours
RPO (Recovery Point Objective): 24 hours

Support

Support Tiers:

Tier 1: End User Support

Handled by your existing helpdesk/contact centre
Common questions: How to use chat, conversation not working
Escalate to Tier 2 if technical issue

Tier 2: Configuration & Integration

Provided by implementation partner or internal IT
Issues: Integration errors, configuration changes
Response: 4 hours (business hours)
Resolution: 24 hours target

Tier 3: Platform & Code

Provided by solution vendor
Issues: Bugs, performance degradation, security issues
Response: 1-4 hours depending on severity
Resolution: Varies by issue

SLA:

Severity	Description	Response	Resolution Target
Critical	Service down, data breach	1 hour	4 hours
High	Major functionality broken	4 hours	24 hours
Medium	Feature not working as expected	1 business day	5 business days
Low	Enhancement request, minor bug	3 business days	30 days

Compliance & Governance

Data Protection

GDPR/DPA 2018 Compliance:

✅ Lawful Basis:

Public task (local gov)
Legitimate interest (housing, charity)
Consent where appropriate

✅ Data Minimisation:

Only collect data necessary for conversation
Don't store payment details (use tokenization)
Auto-delete conversation logs after retention period

✅ Privacy by Design:

PII encryption at rest
Automatic PII redaction from training data
Granular access controls

✅ Data Subject Rights:

Right to access: Export conversation history
Right to erasure: Delete conversations on request
Right to portability: JSON export of user data
Right to object: Opt-out of AI processing (human-only)

Data Retention:

Active conversations: 30 days
Resolved conversations: 6 years (to match case retention)
Analytics (anonymized): 7 years
User can request early deletion

Data Processing:

Purpose: Providing citizen/tenant services
Categories: Name, contact details, address, service requests
Storage: UK Azure regions only
Sharing: Integrated systems (CRM, HMS) only
Retention: 6 years

Information Security

Cyber Essentials Plus Requirements:

✅ Boundary firewalls:

WAF protecting public endpoints
NSG/Security Groups
DDoS protection

✅ Secure configuration:

CIS benchmarks applied
Unnecessary services disabled
Security hardened base images

✅ Access control:

MFA for all admin access
Principle of least privilege
Regular access reviews

✅ Malware protection:

Container image scanning
Runtime protection
Regular vulnerability scans

✅ Patch management:

Automated patching for OS/runtime
Monthly application updates
Emergency patches within 48 hours

ISO 27001 Alignment:

Implemented controls:

A.5 Information security policies
A.6 Organization of information security
A.7 Human resource security
A.8 Asset management
A.9 Access control
A.10 Cryptography
A.12 Operations security
A.13 Communications security
A.14 System acquisition, development and maintenance
A.16 Information security incident management
A.17 Business continuity
A.18 Compliance

[Full control mapping: docs/iso27001-controls.md]

Accessibility

WCAG 2.1 Level AA Compliance:

✅ Perceivable:

Text alternatives for all non-text content
Captions for audio
Adaptable layouts
Sufficient color contrast (4.5:1)

✅ Operable:

Full keyboard navigation
No time limits (can be extended)
Clear navigation
Input assistance

✅ Understandable:

Plain English (readability score: Grade 8)
Predictable behavior
Error prevention and recovery
Help and documentation

✅ Robust:

Valid HTML
Compatible with assistive tech
ARIA landmarks

Assistive Technology Support:

Screen readers (JAWS, NVDA, VoiceOver)
Voice input (Dragon, Windows Speech)
Keyboard only
High contrast mode
Screen magnification

Sector-Specific Compliance

Housing Ombudsman Complaint Handling Code:

Easy to complain via any channel
Complaints logged and tracked
Response timeframes monitored
Learning from complaints captured

Regulator of Social Housing:

Tenant Satisfaction Measures captured
Transparent performance reporting
Evidence of continuous improvement

Local Government Transparency Code:

Contact centre performance published
Satisfaction scores reported
Service standards met

Costs

Licensing

Subscription Model (per organisation):

Size	Users/Properties	Monthly Cost	Annual Cost
Small	<10k citizens/properties	£2,500	£27,000
Medium	10k-50k	£5,000	£54,000
Large	50k-200k	£10,000	£108,000
Enterprise	200k+	Custom	Custom

Included:

All channels (web, phone, SMS, email, WhatsApp)
Unlimited conversations
Standard integrations (up to 5 systems)
Knowledge base hosting
Analytics and reporting
Standard support (business hours)
Software updates

Add-ons:

Premium support (24/7): +50% of subscription
Custom integrations: £5,000-15,000 per integration
Advanced analytics: £1,000/month
Multi-language pack: £500/month per language

One-time Costs:

Implementation: £15,000-50,000
- Discovery and requirements
- Integration setup
- Knowledge base build
- Training delivery
- Go-live support
Training: £2,000-5,000
- Admin training
- End user training
- Train-the-trainer
Customisation: From £5,000
- Branding and UI
- Custom workflows
- Bespoke integrations

Infrastructure Costs (Cloud)

Azure UK (Estimated Monthly):

Size	Conversations/month	Est. Cost
Small	<10,000	£800-1,500
Medium	10,000-50,000	£2,000-4,000
Large	50,000-200,000	£5,000-12,000

Included:

AKS cluster
PostgreSQL database
AI Search
Application Gateway
Monitoring
Storage
Backups

AWS London: Similar costs (±10%)

Cost optimization:

Reserved instances (40% saving)
Auto-scaling (scale to zero overnight if appropriate)
Cold storage for old conversations

On-Premise (One-time + Annual):

Hardware: £30,000-80,000 (servers, storage, network)
Annual maintenance: £5,000-15,000
Staff time: DevOps/infrastructure management

Total Cost of Ownership (5 years)

Medium Housing Association (15,000 properties)

Costs:

Item	Year 1	Years 2-5 (each)
Subscription	£54,000	£54,000
Implementation	£30,000	-
Infrastructure (Azure)	£36,000	£36,000
Support	-	£27,000
Total	£120,000	£117,000

5-Year Total Cost: £588,000

Savings:

Item	Annual Savings
Contact centre staff (2 FTE avoided)	£60,000
Out-of-hours costs	£25,000
Error reduction	£10,000
Total Annual Savings	£95,000

5-Year Total Savings: £475,000

Net Position: Breakeven in Year 2, £85k saving Years 3-5

Intangible Benefits:

Improved tenant satisfaction (+15%)
Better staff morale (focus on complex cases)
24/7 service availability
Scalability for growth
Data insights for service improvement

ROI Calculator

Try our ROI calculator: docs/roi-calculator.xlsx

Enter your:

Organisation size
Current contact centre costs
Expected conversation volume
Out-of-hours requirements

Get:

Estimated costs (subscription + infrastructure)
Projected savings
Payback period
5-year ROI

Implementation Guide

Timeline: 12-16 Weeks

Phase 1: Discovery & Planning (Weeks 1-3)

Activities:

Deliverables:

Requirements specification
Integration architecture
DPIA and IG approvals
Project plan with milestones
Success metrics defined

Who's involved:

Project sponsor
Contact centre manager
IT/Integration lead
Information governance officer
Implementation consultant

Phase 2: Build & Integrate (Weeks 4-8)

Activities:

Deliverables:

Deployed test environment
Integrations tested and verified
Knowledge base populated (80%+)
Security sign-off
Test conversation scripts

Who's involved:

Implementation consultant
Your IT team
CRM/HMS vendor (for API support)
Knowledge managers
InfoSec team

Phase 3: Testing & Training (Weeks 9-11)

Activities:

Deliverables:

UAT sign-off
All tests passed
Trained admins and super-users
Support runbook
Launch communications plan

Who's involved:

UAT users (contact centre advisors)
Admin team
IT operations
Communications team

Phase 4: Pilot (Weeks 12-14)

Activities:

Pilot approach:

Week 12: Internal staff only (test in real environment)
Week 13: Limited public (10% of traffic)
Week 14: Expand to 30% of traffic

Deliverables:

Pilot report
Performance metrics
Refinements completed
Go-live decision

Phase 5: Rollout (Weeks 15-16)

Activities:

Deliverables:

Full production deployment
Launch communications
Performance dashboard
Weekly reports
Benefits tracking started

Post-Implementation (Ongoing)

Month 2-3:

Continuous monitoring
Knowledge base expansion
Model refinement
User feedback incorporation

Month 4-6:

First benefits review
Identify improvement opportunities
Advanced features activation
Team efficiency review

Month 7-12:

Annual review
ROI calculation
Expansion planning (new channels, use cases)
Best practice sharing

Training

Admin Training (Half Day)

Who: System administrators, knowledge managers

Topics:

Platform overview and architecture
Admin portal walkthrough
Managing knowledge base
- Adding/editing content
- Categorization and tagging
- Search optimization
Monitoring conversations
Reviewing and improving responses
User management
Analytics and reporting
Troubleshooting common issues

Format: Virtual or in-person, hands-on labs

Advisor Training (2 Hours)

Who: Contact centre advisors, housing officers

Topics:

What is the AI assistant and how it helps
How to collaborate with AI
- Reading conversation history when escalated
- Providing feedback on AI responses
- Identifying knowledge gaps
When and how to take over conversations
Using analytics to identify trends
Supporting citizens who prefer AI

Format: Virtual session, role-play exercises

Super User Training (Full Day)

Who: Contact centre team leads, service managers

Topics:

Everything in admin training +
Advanced configuration
Integration troubleshooting
Performance tuning
Business analytics deep-dive
Supporting the team
Change management

Format: In-person, extensive hands-on

End User Guidance (Self-Service)

Who: Citizens, tenants, service users

Format:

In-app help and FAQs
Video tutorials (accessibility captions)
Quick reference guides
Live demo sessions (optional)

FAQ

Q: Can this work with our existing CRM/HMS?

A: Yes. We support direct integration with all major UK public sector systems:

Local gov: Civica (all modules), NEC, Capita, Northgate
Housing: Civica Cx, NEC Housing, MRI, Aareon, Orchard
CRM: Dynamics 365, Salesforce, others via API

If your system has an API, we can integrate. If not, we can use file-based sync or screen scraping (less ideal but possible).

Q: Where is the data stored?

A: All data is stored in UK Azure regions (UK South primary, UK West backup) by default. Data never leaves the UK. For on-premise deployment, everything stays on your infrastructure.

Q: What happens if the internet goes down?

A:

Cloud deployment: Service unavailable during outage (same as your existing cloud services). We recommend hybrid deployment if internet reliability is a concern.
Hybrid/on-prem: Assistant continues working with cached data. Some features (AI improvements, analytics) pause but core service continues.
Fallback: You can configure automatic SMS notifications to citizens if service is down.

Q: Can we customize the responses and personality?

A: Absolutely. You can:

Control the tone (formal, friendly, professional)
Customize greeting and signing off
Add organization-specific phrases
Control which questions it answers
Set confidence thresholds for handoff
Configure escalation rules

Q: How accurate is it?

A: In production deployments:

85-92% accuracy for intent recognition
88-95% containment rate (resolved without human)
4.2-4.5/5 average user satisfaction

Accuracy improves over time as the model learns from corrections.

Q: What languages does it support?

A:

Standard: English (UK)
Optional add-on packs: Welsh, Polish, Romanian, Bengali, Urdu, Punjabi, Arabic (and others on request)

Each language pack includes:

Translation of interface
Language-specific AI model
Multilingual knowledge base

Q: Can it handle complex queries?

A: Yes and no:

Simple transactional: Very well (95%+ success)
Moderately complex: Well with context (80-85%)
Very complex/emotional: Automatically escalates to human

The system knows its limits and hands off gracefully when needed.

Q: How long does implementation take?

A: Typical timeline:

Express (limited scope): 8 weeks
Standard: 12-16 weeks
Enterprise (multiple integrations): 16-24 weeks

The main variables are number of integrations and knowledge base size.

Q: Can we trial it before committing?

A: Yes. We offer:

Demo: See it working with sample data (1 hour)
Proof of concept: 6-week pilot with limited scope (£15k)
Pilot: 3-month paid trial with full implementation (convert to full deployment)

Q: What if we don't have APIs for our systems?

A: We have options:

Work with vendor: Many vendors will enable APIs for customers
File-based integration: Daily/hourly sync via files
Database integration: Direct read from databases (read-only)
Screen scraping: Last resort, less reliable

Even without integration, you get value from knowledge base, but integration unlocks the full benefit.

Q: Do we need to hire staff to manage this?

A: No dedicated staff required. Typical management:

Knowledge management: 4-8 hours/week (existing staff)
Performance review: 2-4 hours/week (contact centre manager)
Technical admin: 2-4 hours/month (IT team)

Most organizations assign this to existing roles.

Q: Can it handle complaints?

A: It can:

Recognize when someone is making a complaint
Log it in your complaints system
Provide initial information (process, timescales)
Offer to escalate to complaint handler immediately

For complex complaints, it hands off quickly to ensure compliance with Housing Ombudsman code.

Q: Is it accessible for vulnerable users?

A: Yes. We've designed for inclusion:

Screen reader compatible (JAWS, NVDA)
Simple language option
Voice input support
High contrast mode
Adjustable text size
Option to speak to human at any time

For users who need additional support, advisors can disable AI for specific individuals.

Case Studies

London Borough of Newtown (300k population)

Challenge:

Contact centre overwhelmed with 40,000 calls/month
12-minute average wait time
30% abandonment rate
£800k annual contact centre cost
Poor satisfaction scores (3.1/5)

Solution:

Deployed Citizen Service Assistant with web chat and SMS
Integrated with Civica CRM, Bartec (waste), Idox (planning)
Built 400-article knowledge base
6-week implementation

Results (after 6 months):

18,000 conversations/month handled by AI (45% containment)
Average wait time reduced to 4 minutes
Abandonment rate down to 12%
£220k annual saving (3 FTE avoided, reduced call handling time)
Satisfaction improved to 4.3/5
24/7 service now available

Quote:

"The AI assistant has transformed our contact centre. Our advisors now focus on the complex cases where they add real value, and residents get instant answers 24/7. It's a win-win."

Sarah Johnson, Customer Services Director, London Borough of Newtown

Riverside Housing Association (12,000 properties)

Challenge:

Tenants frustrated by office hours only service
High volume of out-of-hours emergency calls (70% non-emergency)
£45k annual out-of-hours costs
58% first-time resolution
Repairs scheduling inefficient

Solution:

Deployed Tenant Portal Intelligence
Integrated with NEC Housing and Oneserve (repairs)
SMS and web chat channels
10-week implementation

Results (after 9 months):

6,500 conversations/month (42% of total contacts)
Out-of-hours non-emergency calls reduced by 68%
£31k annual saving on out-of-hours contractor
First-time resolution up to 79%
Average time to book repair reduced from 3 days to same-day
Tenant satisfaction up from 3.7/5 to 4.4/5

Quote:

"Our tenants love the 24/7 service. They can report repairs, check rent balances, and get answers whenever they need. The AI correctly triages emergency vs non-emergency repairs, saving us thousands in unnecessary callouts."

David Chen, Director of Customer Experience, Riverside Housing

Hope Community Trust (Charity - homelessness services)

Challenge:

Small team (8 staff) supporting 200+ clients
Unable to provide out-of-hours support
Limited resources for multilingual support
Clients needing help with benefits, housing applications
Staff time spent on repetitive queries

Solution:

Deployed Beneficiary Support Chatbot
SMS-first approach (many clients no web access)
Integrated with Salesforce NPSP
Multi-language support (English, Polish, Romanian)
8-week implementation

Results (after 4 months):

420 conversations/month
24/7 crisis information available
Staff time on routine queries down 60%
Clients supported in 3 languages without hiring translators
92% client satisfaction with AI support
Better crisis support (signposting to emergency services)

Quote:

"For a small charity, this is game-changing. Our clients can get help whenever they need it, in their own language. Our team can focus on the face-to-face work that really matters."

Rev. Michael Thompson, CEO, Hope Community Trust

Get Started

1. Book a Demo

See the Citizen Service Assistant in action with a personalized demo using sample data from your sector.

Book a demo →

30-minute session covering:

Live conversation examples
Your sector-specific use cases
Integration with your systems
Q&A

2. Download Evaluation Guide

Comprehensive guide to help you assess suitability and build a business case.

Download evaluation guide (PDF) →

Includes:

Requirements checklist
ROI calculator
DPIA template
Business case template
Comparison matrix

3. Request Proof of Concept

6-week limited pilot to test in your environment with real data.

Request PoC →

What you get:

Deployed in test environment
One or two integrations
Limited knowledge base
100 test conversations
Results report and recommendation

4. Find an Implementation Partner

Connect with certified partners who can help deploy and support the solution.

View partners →

Partners offer:

Implementation services
Integration expertise
Ongoing support
Training delivery
Customisation

Support & Community

Get Help

Documentation: Full docs →
GitHub Issues: Report bugs →
GitHub Discussions: Ask questions →

Join the Community

Slack Workspace: Join UK Public Sector AI →
Monthly Webinar: Live demos and Q&A - Register →
User Group: Quarterly meetups (virtual) - Sign up →

Share Your Success

Deployed this solution? We'd love to hear about it!

Case study: Submit your story →
Testimonial: Share your experience
Speak: Present at our webinars

License

This solution is available under a commercial license. See LICENSE.md for details.

Open-source components used in this solution are listed in THIRD-PARTY-LICENSES.md.

Contributing

We welcome contributions from the community! Whether it's:

Bug fixes
Feature enhancements
Documentation improvements
Integration adapters for new systems

See CONTRIBUTING.md for guidelines.

Last updated: November 2024 Version: 2.1.0 Maintained by: UK Public Sector AI Solutions

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md

License

maree217/citizen-service-assistant

Folders and files

Latest commit

History

Repository files navigation

Citizen Service Assistant

Overview

Problem Statement

The Challenge

Current State

Impact

Solution

What It Does

How It Works

Example Conversations

Target Sectors

Sector-Specific Benefits

Integration

Compatible Systems

Primary Integrations (Choose one or more)

Supporting Integrations

Service-Specific Integrations

Integration Architecture

API Requirements

Data Flow

Technical Architecture

Components

Infrastructure

Security

Deployment

Prerequisites

Deployment Options

Option 1: Cloud-Hosted on Azure UK (Recommended)

Option 2: AWS London

Option 3: Hybrid Deployment

Option 4: Fully On-Premise

Configuration

Testing

Operation

Monitoring

Maintenance

Support

Compliance & Governance

Data Protection

Information Security

Accessibility

Sector-Specific Compliance

Costs

Licensing

Infrastructure Costs (Cloud)

Total Cost of Ownership (5 years)

ROI Calculator

Implementation Guide

Timeline: 12-16 Weeks

Phase 1: Discovery & Planning (Weeks 1-3)

Phase 2: Build & Integrate (Weeks 4-8)

Phase 3: Testing & Training (Weeks 9-11)

Phase 4: Pilot (Weeks 12-14)

Phase 5: Rollout (Weeks 15-16)

Post-Implementation (Ongoing)

Training

Admin Training (Half Day)

Advisor Training (2 Hours)

Super User Training (Full Day)

End User Guidance (Self-Service)

FAQ

Case Studies

London Borough of Newtown (300k population)

Riverside Housing Association (12,000 properties)

Hope Community Trust (Charity - homelessness services)

Related Solutions

Get Started

1. Book a Demo

2. Download Evaluation Guide

3. Request Proof of Concept

4. Find an Implementation Partner

Support & Community

Get Help

Join the Community

Packages