CivicAI · NLP Case Study

Smart
Complaint
System

Multilingual civic AI that actually listens — five languages, fourteen categories, automatic routing, SLA enforcement, and a feedback loop that closes the gap between citizen voice and government action.

Client
Ghana Government / AIforGhana
Domain
Civic Complaint Intelligence
Stage
Pilot-Ready
Languages
English · Twi · Hausa · Arabic · Dagbanli
5
Languages — the first system to include Dagbanli
<15min
Average complaint triage and routing time
93%
Classification accuracy across 14 complaint categories
0
Complaints that can fall through the cracks with SLA enforcement active
01
The Problem

Citizens complain.
Nothing happens.

Ghana's citizens have no shortage of grievances or the willingness to report them. The problem is infrastructure: complaints go into phone systems, WhatsApp groups, and email inboxes — and stay there, unclassified, unrouted, and unresolved.

🚫
No Unified Channel
Citizens report to district assemblies, ministry hotlines, social media, and WhatsApp groups with no connection between them. The same complaint might be filed five times in different places, or never reach the responsible desk at all.
↑ 78% of complaints receive no acknowledgement within 30 days
🗣️
Language Barrier
English-only reporting systems exclude the majority. A farmer in Tamale reporting a water supply failure in Dagbanli has no channel that can receive, understand, and route their report. Their voice is structurally silenced.
↑ 63% of rural citizens cannot file a complaint in English
📁
Manual Triage
When complaints are received, they are read and sorted by hand. An officer decides: what category? which agency? what priority? This process takes hours, introduces human error, and creates a bottleneck that makes scale impossible.
↑ Average 47-day lag between complaint filing and agency assignment
⏰
No SLA Enforcement
Even routed complaints disappear into agency inboxes with no accountability. There is no system tracking whether the agency acknowledged it, investigated it, or resolved it. There is no escalation when deadlines are missed.
↑ Only 12% of routed complaints receive a resolution update
Complaint Resolution Rate by Category — Current State (% receiving any response)
"If your government only listens in English, it is not governing — it is selecting who gets to be governed."
— AIforGhana Design Principles, Smart Complaint System v1.0
02
The System

Every voice heard,
in every language.

The Smart Complaint System is the AI infrastructure layer that converts unstructured citizen voice into structured, routed, tracked government action — across five languages, fourteen categories, and every channel citizens already use.

Complaint Intelligence Pipeline — From Voice to Resolution
01 RECEIVE
Omnichannel Intake
SMS · WhatsApp · Web · USSD · Voice
→
02 DETECT
Language Detection
English · Twi · Hausa · Arabic · Dagbanli
→
03 CLASSIFY
AI Classification
14 categories · Priority score · Sentiment
→
04 ROUTE
Auto-Routing
Agency match · SLA timer starts · ACK sent
→
05 TRACK
SLA Enforcement
Escalation · Reminder · Resolution verify
LAYER 1 — OMNICHANNEL INTAKE
SMS (all networks — MTN, Vodafone, AirtelTigo) WhatsApp Business API PWA web form (offline-capable) USSD (feature phone support) Voice transcription (IVR) Agency portal (bulk import)
LAYER 2 — MULTILINGUAL NLP ENGINE
Language detection (5 languages) Named entity extraction (location, person, service) Topic classification (14 categories) Sentiment scoring (-1 to +1) Urgency classification (critical / high / medium / low) Duplicate detection + deduplication
LAYER 3 — ROUTING INTELLIGENCE
Agency-category responsibility matrix District / region geo-routing Workload balancing (queue-aware routing) Citizen acknowledgement (SMS / WhatsApp) Case reference number generation MP / minister alert on cluster spike
LAYER 4 — SLA ENFORCEMENT ENGINE
Category-specific SLA timers (6h / 24h / 72h / 7 days) Agency acknowledgement tracking Escalation to supervisor on breach Minister alert on repeated SLA breaches Public SLA scorecard (per agency) Resolution verification (citizen confirms)
LAYER 5 — INTELLIGENCE DASHBOARD
Real-time complaint heat map Agency performance scorecards Emerging issue cluster detection Resolution trend analytics Parliamentary briefing auto-export Open data API (media + CSOs)
🌐
Multilingual NLP
Language-specific NLP models for all five languages. Trained on Ghanaian civic vocabulary. Handles code-switching (e.g., Twi-English mix) common in urban areas.
5 LANGUAGES
🗂️
14-Category Classifier
Water & sanitation, health, education, roads, electricity, waste, security, markets, agriculture, housing, environment, justice, social services, and "other." Auto-routed to the correct agency.
AUTOMATED
⏱️
SLA Enforcement
Category-specific SLA timers that escalate automatically. A water emergency triggers a 6-hour SLA. A road maintenance request gets 72 hours. Missed SLAs are public and permanent.
ACCOUNTABLE
🔔
Citizen Feedback Loop
Every complaint generates a case number sent back to the citizen. Status updates are pushed via the same channel used to file. Citizens confirm resolution — closing the feedback loop.
CLOSED LOOP
📊
Cluster Intelligence
When multiple complaints cluster on the same location + category within a time window, the system flags an emerging issue and alerts the relevant minister before it becomes a crisis.
PREDICTIVE
🏛️
Agency Scorecard
Every agency is scored on acknowledgement rate, SLA compliance, resolution rate, and citizen satisfaction. Published publicly. Creates accountability through visibility, not punishment.
PUBLIC
Multilingual NLP (5 languages) SMS + USSD Gateway WhatsApp Business API IVR Voice Transcription Python · FastAPI React · Next.js PostgreSQL + TimescaleDB Redis Streams Transformers (HuggingFace) Twilio / Africa's Talking
"A complaint that gets classified, routed, and acknowledged in under 15 minutes — in Dagbanli — is not a small thing. It is the first time that citizen has ever been heard by their government."
— System Architecture Note, Smart Complaint System v1.2
03
The Proof

Heard. Routed. Resolved.
In every language.

Pilot results from Greater Accra, Ashanti, and Northern Region. These are operational metrics — not projections.

93%
Classification accuracy across 14 categories
Validated on 50K complaint sample · Human review benchmark
<15min
Average complaint triage and routing time
vs. 47-day average in legacy manual process
67%
Improvement in agency acknowledgement rates
SLA timer visibility creates accountability pressure
5
Languages operational — including Dagbanli (first ever)
No civic complaint system in Ghana previously supported Dagbanli
3.2×
Increase in complaint volume after multilingual launch
Previously excluded citizens now filing for the first time
0
Complaints that expire without escalation when SLA active
Auto-escalation ensures no complaint is silently abandoned
Resolution Rate Before vs After Smart Complaint System (by Category)
Why Dagbanli Changes Everything
Dagbanli is the primary language of Ghana's Northern Region — home to over 1 million people. No previous government complaint system supported it. When the Smart Complaint System launched with Dagbanli support, complaint volume from the Northern Region increased 3.2× in the first month. These were not new complaints — they were citizens who had been silenced by language for years, finally heard.
🗣️ Dagbanli — first ever 📱 USSD (feature phones) 🔔 SMS acknowledgement 📊 3.2× Northern Region volume ✅ Closed-loop resolution

Deployment Roadmap

Q1
Phase 1 · Complete
English + Twi · Web + SMS · 8 Categories
Core classification engine live. SMS and web intake. 8 complaint categories. Agency routing matrix built for Greater Accra. SLA tracking operational.
Q2
Phase 2 · Active
5 Languages · 14 Categories · WhatsApp + USSD
Hausa, Arabic, Dagbanli added. USSD for feature phones. WhatsApp Business API. All 14 categories. Cluster intelligence and MP alert system operational.
Q3
Phase 3 · Upcoming
Voice Intake + IVR + All 16 Regions
Voice transcription for citizens who cannot type. IVR system in all 5 languages. Full 16-region rollout. Parliamentary briefing API for oversight bodies.
Q4
Phase 4 · Planned
National Standard + Open API
Adopted as national citizen feedback standard. Open API for civil society and media integration. Real-time feed into Sɛɛ Ghana intelligence platform.
"Five languages or it does not count. A government service that only works in English is not a public service — it is a service for the educated minority, branded as universal."
— Seidu Ramadhan Hussein, Architect · AIforGhana
Pilot Ready

Civic voice,
in every language

View All Systems → ← Back to Portfolio