🤖 AI-Powered Data Collection Lab
Master chatbot surveys, satellite imagery analysis, mobile AI features, and ethical data collection practices
💬 Chatbot & Conversational AI Practice
Design conversational surveys that feel like natural interactions rather than bureaucratic interrogations.
🎯 Key Chatbot Design Principles
🔄 Dynamic Adaptation
Skip irrelevant questions based on previous answers. If someone doesn't own land, don't ask about irrigation.
✅ Real-Time Validation
Flag illogical responses immediately ("Your household has 47 members?") and prompt correction.
🌐 Asynchronous Flexibility
Let respondents complete surveys at their convenience—during commutes, breaks, or evening downtime.
🗣️ Multilingual Intelligence
Automatically switch languages based on user preference or detect language from responses.
Design Challenge: COVID-19 Patient Monitoring
Context: A health NGO needs to monitor 15,000 COVID-19 patients in home isolation. Traditional phone-based monitoring is expensive ($300,000/month), slow (staff reach only 200-300 patients daily), and misses deteriorating cases between scheduled calls.
Your Task: Design a WhatsApp chatbot that checks in daily with symptom questions, flags severe cases, and alerts medical teams automatically.
🛠️ Build Your Chatbot Survey
Essential Chatbot Elements Checklist
Sample Conversation Flow
💡 Chatbot Best Practices
- Keep questions short and conversational (5-15 words)
- Use familiar platforms (WhatsApp, SMS) rather than custom apps
- Provide clear response options to reduce confusion
- Allow natural language where appropriate, not just rigid formats
- Include progress indicators ("Question 3 of 8")
- Offer help/clarification options throughout
- Test with target population before full deployment
📊 Expected Outcomes Comparison
| Metric | Traditional Phone Calls | WhatsApp Chatbot |
|---|---|---|
| Patients Monitored | 200-300 per day | 15,000 simultaneously |
| Cost | $300,000/month (50 staff) | $75,000/month (75% reduction) |
| Response Time | Hours between check-ins | Minutes for critical alerts |
| Completion Rate | 60% (many missed calls) | 87% (asynchronous convenience) |
| Data Quality | Manual entry errors | Validated in real-time |
| Scalability | Requires proportional staff | Marginal cost near zero |
🛠️ Popular Chatbot Platforms
Twilio
Best for: SMS-based surveys, works on basic phones, programmable, developer-friendly
Cost: Pay-per-message (starting $0.0075/SMS)
ManyChat
Best for: Facebook Messenger & WhatsApp bots, visual builder, no coding required
Cost: Free tier available, Pro from $15/month
Chatfuel
Best for: No-code bot builder, WhatsApp & Telegram support, AI responses
Cost: Free up to 1,000 conversations/month
🛰️ Remote Sensing & Satellite Imagery Practice
Learn to monitor vast areas, track changes over time, and extract actionable M&E insights from space.
🌍 What AI Can Detect From Space
🌾 Agricultural Monitoring
Crop health via NDVI, drought stress detection, harvest predictions, yield estimates (85-95% accuracy)
🏗️ Infrastructure Development
New buildings/roads detected automatically, construction progress tracked, displacement camps mapped
🌳 Environmental Changes
Deforestation rates quantified, water quality monitored, land degradation tracked over time
👥 Population Dynamics
Settlement patterns mapped, urbanization trends quantified, migration flows estimated
🌊 Disaster Impact
Flood extent mapped in real-time, earthquake damage assessed, wildfire spread tracked
📍 Program Verification
Verify activities without site visits (trees planted? buildings constructed?), detect fake GPS
Real-World Application: WFP Famine Early Warning
Challenge: World Food Programme needed to predict food security crises in East Africa 3-6 months in advance. Traditional field surveys took 2-4 months to collect and analyze—by then, famines were already developing.
Solution: AI-analyzed satellite imagery monitoring crop health (NDVI), rainfall patterns, vegetation greenness, and market activity. Combined with weather forecasts and economic indicators.
Results: Predicts food insecurity 3-6 months earlier, covers entire regions continuously, reaches remote/conflict areas, reduces warning-to-response time from months to weeks, prevents 30-40% of famine mortality through earlier intervention.
🎯 Interactive Satellite Analysis Exercise
Scenario: Reforestation Program Verification
Your organization funded a reforestation program claiming to plant 50,000 trees across 500 hectares in Madagascar over 2 years. You need to verify progress without expensive field visits. Select which satellite indicators would be most useful:
💡 Satellite Imagery Best Practices for M&E
- Establish baseline imagery BEFORE program starts
- Account for seasonal variations (don't compare rainy season to dry season)
- Cloud cover is a major limitation—use radar for cloudy regions
- Combine satellite data with ground truth validation samples
- Resolution matters: 10-30m for agriculture, 1-5m for infrastructure
- Free sources (Sentinel, Landsat) often sufficient for M&E needs
- AI can automate analysis but requires training data from your context
🛠️ Free Satellite Imagery Platforms
🌍 Google Earth Engine
Best for: Massive data catalog, powerful analysis, time-series comparisons
Data: Landsat (1970s-present), Sentinel, MODIS, weather data
Cost: Free for research/education (requires coding knowledge)
🛰️ Sentinel Hub
Best for: European satellite data, customizable visualization, API access
Data: Sentinel-1 (radar), Sentinel-2 (optical), Sentinel-3 (ocean/land)
Cost: Free tier available, commercial plans for high-volume use
🌎 NASA Worldview
Best for: Real-time views, natural disasters, environmental monitoring
Data: MODIS, VIIRS, multiple NASA satellites
Cost: Completely free, no account required
📊 Benefits of Satellite-Based M&E
🌍 Access Previously Inaccessible Areas
Monitor conflict zones, disaster areas, remote villages, or politically restricted regions safely from a distance. Collect data where human presence is impossible, dangerous, or prohibited.
🔄 Continuous Rather Than Episodic Monitoring
Track changes daily or weekly rather than baseline-midline-endline snapshots. Detect problems as they emerge rather than months later. Document gradual changes invisible to periodic visits.
💰 Cost-Effective at Scale
Cover entire regions for less than traditional survey costs for a fraction of sites. Marginal cost of monitoring additional area approaches zero. One analyst with satellite data can monitor what would require dozens of field teams.
🎯 Objective, Unbiased Observations
Satellites don't have preconceptions, don't tire, don't fabricate data. Provides ground truth free from human interpretation or reporting bias. Validates or challenges self-reported data.
📱 AI-Enhanced Mobile Data Collection
Transform mobile apps into intelligent assistants that improve data quality and field team productivity.
🚀 AI Features Transforming Mobile Data Collection
🖼️ Image Recognition & OCR
What it does: AI analyzes photos automatically—reads text from documents, identifies crop types, verifies products
Accuracy: 85-95% for clear images
🎤 Voice-to-Text Transcription
What it does: Converts spoken interviews to text instantly in 100+ languages with speaker identification
Time saved: 40-60% reduction in interview time
📍 Geolocation Verification
What it does: Flags suspicious GPS coordinates, impossible travel speeds, patterns suggesting GPS spoofing
Fraud detection: 70-85% of fabricated data caught
🔍 Smart Error Detection
What it does: Detects outliers in real-time, flags inconsistencies, prompts immediate correction
Error reduction: 60-80% fewer data cleaning issues
🔄 Adaptive Skip Logic
What it does: AI learns from patterns to optimize survey flow and reduce respondent burden
Completion improvement: 15-25% higher rates
📊 Real-Time Analytics
What it does: Aggregates data as it flows, flags quality issues, compares enumerator performance
Management: Enable adaptive decisions during data collection
Case Study: South Sudan Vaccination Verification
Challenge: Health workers needed to verify children's vaccination records from handwritten paper cards. Manual data entry was slow (5-7 minutes per card), error-prone (15-20% transcription errors), creating bottlenecks and long waits.
Solution: CommCare app with AI-powered OCR. Workers photographed vaccination cards. AI automatically extracted child's name, DOB, all vaccine types/dates, and lot numbers. System flagged missing vaccines and identified children behind schedule.
Results: Processing time reduced 85% (from 5-7 min to 45-60 sec), errors dropped from 15-20% to under 2%, throughput increased 400% (same workers processed 4x more children daily), created accurate digital records for 50,000+ children in 6 months.
🎯 Design Your AI-Enhanced Mobile App
Select AI Features for Your Use Case
Scenario: You're conducting household surveys to assess water access in rural Kenya. Households have varying literacy levels. You need to verify water sources, measure distances, photograph facilities, and record GPS locations. Select the most valuable AI features:
💡 Mobile AI Implementation Best Practices
- Test AI features with target populations BEFORE deployment
- Always allow manual override when AI makes mistakes
- Offline functionality is CRITICAL for rural/low-connectivity areas
- Train enumerators on when to trust AI vs when to question it
- Monitor AI accuracy throughout deployment, not just pre-testing
- Build in quality checks for AI-extracted data (random verification samples)
- Consider device requirements (camera quality, processing power, storage)
🛠️ Popular AI-Enhanced Mobile Platforms
📱 KoboToolbox
AI Features: Growing OCR support, image analysis plugins, advanced validation
Best for: Humanitarian sector, free for unlimited surveys, offline-first design
Cost: Free (open-source)
📊 SurveyCTO
AI Features: Advanced validation, audio audit, ML integration, high security
Best for: Research projects, enterprise needs, high-security requirements
Cost: From $149/month
🏥 CommCare
AI Features: Image recognition, OCR, AI decision support, voice collection
Best for: Health sector, case management, community health workers
Cost: Free tier available, paid plans from $100/month
📊 Impact Comparison: Traditional vs AI-Enhanced
| Task | Traditional Manual | AI-Enhanced | Time Savings |
|---|---|---|---|
| Transcribing 30-min interview | 90-120 minutes | 5-10 minutes (verify AI transcription) | 85-90% |
| Extracting data from vaccination card | 5-7 minutes | 45-60 seconds | 85% |
| Data cleaning 1,000 surveys | 40-60 hours | 5-10 hours (reviewing flagged items) | 80-85% |
| Verifying GPS locations | Manual spot checks (10-20%) | Automated 100% verification | Comprehensive coverage |
| Identifying crop diseases in photos | Expert review needed | Instant AI classification (85% accuracy) | Real-time field guidance |
⚖️ Ethical Considerations in AI Data Collection
Navigate privacy, bias, exclusion, and accountability challenges when deploying AI for data collection.
⚠️ Critical Ethical Challenges
🔒 Data Privacy & Security
AI tools collect sensitive information on cloud servers. Are you encrypting data, following GDPR/HIPAA, limiting access, and planning for breaches?
⚖️ Algorithmic Bias
AI trained on unrepresentative data produces biased results. Chatbots may fail for non-native speakers, image recognition for darker skin tones.
📵 Digital Divide
AI methods require smartphones, internet, digital literacy—excluding elderly, poor, rural, or less-educated populations systematically.
👁️ Surveillance Concerns
Continuous monitoring can feel invasive. GPS tracking reveals movements. Satellite imagery monitors private land. Trust, once lost, is hard to rebuild.
🎯 Informed Consent
Do respondents understand how AI processes their data? What happens to it? Where it goes? How long it's retained? Who accesses it?
🌐 Cultural Sensitivity
AI may not understand local contexts, customs, communication styles—leading to misinterpretation that's inaccurate or offensive.
🎯 Ethical Dilemma Scenarios
Scenario 1: Remote Monitoring Without Consent
Your organization wants to use satellite imagery to monitor deforestation in areas where your reforestation program operates. Communities haven't been informed this monitoring will occur. What's the ethical approach?
Scenario 2: Digital Exclusion in Vulnerable Populations
Your chatbot survey reaches 85% of the target population efficiently and cheaply. But it systematically excludes elderly women who don't use smartphones—a highly vulnerable group. What should you do?
Scenario 3: AI Bias in Data Collection
Your voice-to-text transcription AI works well for standard accents but has 40% error rate for local dialect speakers, who are also your most marginalized beneficiaries. What's the right response?
⚖️ The Community Test: Your Ethical Litmus Test
When facing difficult ethical decisions, apply the "community test": Would you be comfortable explaining your data collection methods, purposes, and safeguards to the people being studied in a public community meeting?
If the answer is no—if you'd feel defensive, embarrassed, or worried about their reaction—that's a red flag suggesting your approach needs rethinking.
Communities deserve data collection that serves their interests, respects their dignity, protects their privacy, and prioritizes their wellbeing over research convenience.
✅ Ethical Best Practices Checklist
💡 Remember: Ethics Aren't Optional
With great technological power comes great ethical responsibility. AI-powered systems can reach more people faster and extract more information—capabilities that amplify both benefits AND risks.
Getting the technology right is necessary but insufficient; getting the ethics right is what determines whether AI serves or harms the communities you aim to help.
Use AI to amplify voices, not exploit data. Prioritize community wellbeing over research convenience. Build trust through transparency and accountability.