Predictive AI/ML for Frontline Maternal Care
Predictive AI/ML for Frontline Maternal Care
Predictive AI/ML for Frontline Maternal Care
How I built explainable, human-first UX for frontline healthcare workers
How I built explainable, human-first UX for frontline healthcare workers
How I built explainable, human-first UX for frontline healthcare workers
Role
UX designer and Researcher
Industry
Public Health
OVERVIEW
I led the UX design for an AI-powered intervention that helps India’s frontline health workers identify which pregnant women are at risk of dropping out of care. Integrated into the government’s ANMOL MP app in Madhya Pradesh, the tool had to work offline, earn trust, and be used by ANMs (Auxiliary Nurse Midwives) with limited time and tech training. This case study covers how we translated predictive machine learning into something explainable, actionable, and usable without losing the human behind the data.
I led the UX design for an AI-powered intervention that helps India’s frontline health workers identify which pregnant women are at risk of dropping out of care. Integrated into the government’s ANMOL MP app in Madhya Pradesh, the tool had to work offline, earn trust, and be used by ANMs (Auxiliary Nurse Midwives) with limited time and tech training. This case study covers how we translated predictive machine learning into something explainable, actionable, and usable without losing the human behind the data.
I led the UX design for an AI-powered intervention that helps India’s frontline health workers identify which pregnant women are at risk of dropping out of care. Integrated into the government’s ANMOL MP app in Madhya Pradesh, the tool had to work offline, earn trust, and be used by ANMs (Auxiliary Nurse Midwives) with limited time and tech training. This case study covers how we translated predictive machine learning into something explainable, actionable, and usable without losing the human behind the data.
In India’s rural health system, every missed checkup can mean a missed chance to save a life. Auxiliary Nurse Midwives (ANMs) are responsible for hundreds of pregnant women across villages, often working with no network, limited support, and no way to know who needs them most.
Dropouts after the second antenatal visit are common. While tools like ANMOL app exists, they weren’t helping ANMs act in time.
At the same time, government incentives under JSY and PMMVY in some states are only paid when all four checkups are completed. Most women don’t reach the fourth, and ANMs had no way to track or support them effectively.
So we asked a single question:
problem AND OPPURTUNITY
The problem
The problem
The problem
In India’s rural health system, every missed checkup can mean a missed chance to save a life. Auxiliary Nurse Midwives (ANMs) are responsible for hundreds of pregnant women across villages, often working with no network, limited support, and no way to know who needs them most.
Dropouts after the second antenatal visit are common. While tools like ANMOL app exists, they weren’t helping ANMs act in time.
At the same time, government incentives under JSY and PMMVY in some states are only paid when all four checkups are completed. Most women don’t reach the fourth, and ANMs had no way to track or support them effectively.
So we asked a single question:
In India’s rural health system, every missed checkup can mean a missed chance to save a life. Auxiliary Nurse Midwives (ANMs) are responsible for hundreds of pregnant women across villages, often working with no network, limited support, and no way to know who needs them most.
Dropouts after the second antenatal visit are common. While tools like ANMOL app exists, they weren’t helping ANMs act in time.
At the same time, government incentives under JSY and PMMVY in some states are only paid when all four checkups are completed. Most women don’t reach the fourth, and ANMs had no way to track or support them effectively.
So we asked a single question:
"How might we enable ANMs to proactively identify and act on high-risk pregnancy dropouts before it's too late?"
"How might we enable ANMs to proactively identify and act on high-risk pregnancy dropouts before it's too late?"
Understanding the User
Understanding the User
Understanding the User
I conducted in-depth interviews with 50 Auxiliary Nurse Midwives (ANMs) across rural blocks in Jabalpur, Madhya Pradesh, and shadowed 10 ANMs during Village Health and Nutrition Days (VHSND). I observed how they conducted home visits, performed antenatal checkups, entered data into paper registers and the ANMOL app, and coordinated with ASHAs and families, often in weather conditions, poor network, and with little systemic support.
I conducted in-depth interviews with 50 Auxiliary Nurse Midwives (ANMs) across rural blocks in Jabalpur, Madhya Pradesh, and shadowed 10 ANMs during Village Health and Nutrition Days (VHSND). I observed how they conducted home visits, performed antenatal checkups, entered data into paper registers and the ANMOL app, and coordinated with ASHAs and families, often in weather conditions, poor network, and with little systemic support.
I conducted in-depth interviews with 50 Auxiliary Nurse Midwives (ANMs) across rural blocks in Jabalpur, Madhya Pradesh, and shadowed 10 ANMs during Village Health and Nutrition Days (VHSND). I observed how they conducted home visits, performed antenatal checkups, entered data into paper registers and the ANMOL app, and coordinated with ASHAs and families, often in weather conditions, poor network, and with little systemic support.
Sunita Yadav, 34, is an Auxiliary Nurse-Midwife (ANM) who serves 4 villages in Jabalpur, Madhya Pradesh. She is the first point of healthcare in these villages. She manages 200+ pregnancies every cycle, walking long distances, carrying a register with her medical kit, and doing it all with no signal, no real tech support, and no system in place to help her prioritise better.
Her Responsibilities:
Track all registered pregnancies in her area
Conduct ANC checkups, maintain digital and paper records
Counsel families on nutrition, danger signs, institutional deliveries
Ensure eligible women receive JSY/PMMVY benefits
Refer and accompany high-risk cases to health facilities
Coordinate with ASHAs, AWWs, and Medical Officers
Travel for immunization days, deliveries, and emergencies
Paint points:
📓 Tracks patients using paper registers
📶 ANMOL app fails to sync in low network, low trust in digital tools
❌ No way to know who’s missed ANC 2 or 3
🤯 Overburdened with work “Too many cases to remember”
🤷♀️ Doesn’t trust new tools without clear value
System Level Challenges
System Level Challenges
System Level Challenges
No Integration with Policy Milestones
Payments to patients only come after completing all four ANC checkups in Madhya Pradesh. But most women drop out after the second or third.
No Integration with Policy Milestones
Payments to patients only come after completing all four ANC checkups in Madhya Pradesh. But most women drop out after the second or third.
No Integration with Policy Milestones
Payments to patients only come after completing all four ANC checkups in Madhya Pradesh. But most women drop out after the second or third.
Lack of Prioritization Support
The ANMOL app displayed a flat list of all pregnancies with no risk signals or intelligent sorting. ANMs had no tool to help them decide whom to prioritise.
Lack of Prioritization Support
The ANMOL app displayed a flat list of all pregnancies with no risk signals or intelligent sorting. ANMs had no tool to help them decide whom to prioritise.
Lack of Prioritization Support
The ANMOL app displayed a flat list of all pregnancies with no risk signals or intelligent sorting. ANMs had no tool to help them decide whom to prioritise.
Sync Failures in Low Network Areas
Many rural zones had patchy or zero signal. ANMOL’s sync often failed silently, leading to data loss and loss of trust in the system.
Sync Failures in Low Network Areas
Many rural zones had patchy or zero signal. ANMOL’s sync often failed silently, leading to data loss and loss of trust in the system.
Sync Failures in Low Network Areas
Many rural zones had patchy or zero signal. ANMOL’s sync often failed silently, leading to data loss and loss of trust in the system.
Key Design Insights
Why ANMs distrust digital tools (especially due to sync failures)
Why ANMs distrust digital tools (especially due to sync failures)
Why ANMs distrust digital tools (especially due to sync failures)
Respect the ANM's judgment, not replace it
Respect the ANM's judgment, not replace it
Respect the ANM's judgment, not replace it
Flag who’s most at risk for better prioritisation
Flag who’s most at risk for better prioritisation
Flag who’s most at risk for better prioritisation
The solution must work offline with clear sync feedback
The solution must work offline with clear sync feedback
The solution must work offline with clear sync feedback
The solution must fit within the ANMs existing workflow
The solution must fit within the ANMs existing workflow
The solution must fit within the ANMs existing workflow
A disconnect between AI potential and field usage
A disconnect between AI potential and field usage
A disconnect between AI potential and field usage
The emotional burden of managing care for so many with so little
The emotional burden of managing care for so many with so little
The emotional burden of managing care for so many with so little
Design Goals
Trustworthy & Explainable AI
Trustworthy & Explainable AI
Trustworthy & Explainable AI
Simple UI for low literacy users
Simple UI for low literacy users
Simple UI for low literacy users
Offline-first functionality
Offline-first functionality
Offline-first functionality
Actionable insights and prioritisation (not just predictions)
Actionable insights and prioritisation (not just predictions)
Actionable insights and prioritisation (not just predictions)
User Flow
Initial Wireframes
Designing With ANMs, Not Just For Them
Usability Testing & Co‑Design
Usability Testing & Co‑Design
To validate the AI‑driven workflows and ensure our solution fits real‑world constraints, we conducted usability testing with 8 Auxiliary Nurse Midwives (ANMs) across 18 live antenatal care (ANC) visits in Jabalpur district, Madhya Pradesh.
These were not controlled lab sessions they took place during actual patient interactions, where connectivity fluctuated, time was limited. This field‑first approach revealed not just usability issues, but also deep behavioural insights about trust, perception, and decision‑making under pressure.
In parallel, we collected feedback and facilitated co‑design workshops with ANMs to iteratively refine language, flow, and feedback systems.
Every design choice from AI validation to core features was shaped by real user feedback, not assumptions.
To validate the AI‑driven workflows and ensure our solution fits real‑world constraints, we conducted usability testing with 8 Auxiliary Nurse Midwives (ANMs) across 18 live antenatal care (ANC) visits in Jabalpur district, Madhya Pradesh.
These were not controlled lab sessions they took place during actual patient interactions, where connectivity fluctuated, time was limited. This field‑first approach revealed not just usability issues, but also deep behavioural insights about trust, perception, and decision‑making under pressure.
In parallel, we collected feedback and facilitated co‑design workshops with ANMs to iteratively refine language, flow, and feedback systems.
Every design choice from AI validation to core features was shaped by real user feedback, not assumptions.
Voices from the field
Voices from the field
“Yeh list sabse kaam ki cheez hai isse mujhe yaad rehta hai kisko follow up karna hai”
“This due list is the most useful. It helps me remember who needs follow‑up.”
️️💡 Elevated the Due List to the app’s home screen as the primary interaction point.
“Agree/Disagree’ sunke lagta hai jaise maine kuch galat kar diya ho.”
“Agree/Disagree’ makes it feel like I’ve done something wrong..”
️️💡 Reframed the AI validation flow to Yes/No to reduce judgment and build psychological safety.
“Agree/Disagree’ sunke lagta hai jaise maine kuch galat kar diya ho.”
“Agree/Disagree’ makes it feel like I’ve done something wrong..”
️️💡 Reframed the AI validation flow to Yes/No to reduce judgment and build psychological safety.
“Prediction delay mein lagta hai kuch gadbad hai.”
“When the prediction takes long, I feel something’s wrong.”
️️💡 Introduced a clear progress bar and reassurance messaging during AI processing.
The solution
ANC Visit Form – Patient Data Entry
The visit begins with a streamlined form to capture patient vitals and demographic details. Fields are ordered to match the natural ANC workflow, reducing mental load and ensuring data completeness.
🧭 Logical field order
Follows the natural ANC workflow: LMP → vitals → tests → parity — no unnecessary scrolling or switching.
🧭 Logical field order
Follows the natural ANC workflow: LMP → vitals → tests → parity — no unnecessary scrolling or switching.
🧭 Logical field order
Follows the natural ANC workflow: LMP → vitals → tests → parity — no unnecessary scrolling or switching.
📱💻 Optimized for mobile and tablet views
All components are responsive, ensuring usability on different screen sizes during home visits (mobile) or facility-based work (tab).
📱💻 Optimized for mobile and tablet views
All components are responsive, ensuring usability on different screen sizes during home visits (mobile) or facility-based work (tab).
📱💻 Optimized for mobile and tablet views
All components are responsive, ensuring usability on different screen sizes during home visits (mobile) or facility-based work (tab).
🖐️ Large tap targets + minimal typing
Prioritizes drop-downs, toggles, and auto-fills to reduce fatigue and improve speed.
🖐️ Large tap targets + minimal typing
Prioritizes drop-downs, toggles, and auto-fills to reduce fatigue and improve speed.
🖐️ Large tap targets + minimal typing
Prioritizes drop-downs, toggles, and auto-fills to reduce fatigue and improve speed.
🗣️ Local language support
Hindi labels improve confidence and comprehension, especially for newer ANMs or those working in low-literacy areas.
🗣️ Local language support
Hindi labels improve confidence and comprehension, especially for newer ANMs or those working in low-literacy areas.
🗣️ Local language support
Hindi labels improve confidence and comprehension, especially for newer ANMs or those working in low-literacy areas.
Consent Screen – Shown After Saving Visit Data
Before the AI prediction is triggered, the ANM is prompted to seek consent from the patient. This ensures transparency and aligns with WHO guidelines and IRB requirements for AI use in healthcare.
📅 Shown only during the first ANC visit
To reduce repetition and friction, consent is requested just once per pregnancy cycle.
❌ Optional consent with cross (“X”) button
Manage your expenses, budgets, and investments all in one place with an intuitive and user-friendly interface.
✔️Bilingual communication
Supports local languages to ensure clarity for both ANMs and patients across regions.
🔄 Works offline with deferred sync
Manage your expenses, budgets, and investments all in one place with an intuitive and user-friendly interface.
🛡️ Compliant with WHO guidelines & IRB requirements
Consent language is concise, readable, and based on ethics review protocols.
📅 Shown only during the first ANC visit
To reduce repetition and friction, consent is requested just once per pregnancy cycle.
❌ Optional consent with cross (“X”) button
Manage your expenses, budgets, and investments all in one place with an intuitive and user-friendly interface.
✔️Bilingual communication
Supports local languages to ensure clarity for both ANMs and patients across regions.
🔄 Works offline with deferred sync
Manage your expenses, budgets, and investments all in one place with an intuitive and user-friendly interface.
🛡️ Compliant with WHO guidelines & IRB requirements
Consent language is concise, readable, and based on ethics review protocols.
📅 Shown only during the first ANC visit
To reduce repetition and friction, consent is requested just once per pregnancy cycle.
❌ Optional consent with cross (“X”) button
Manage your expenses, budgets, and investments all in one place with an intuitive and user-friendly interface.
✔️Bilingual communication
Supports local languages to ensure clarity for both ANMs and patients across regions.
🔄 Works offline with deferred sync
Manage your expenses, budgets, and investments all in one place with an intuitive and user-friendly interface.
🛡️ Compliant with WHO guidelines & IRB requirements
Consent language is concise, readable, and based on ethics review protocols.
AI Prediction Result – Validation Flow (Yes/No)
After the ANC visit is saved and consent is captured, the AI model predicts the patient’s risk of ANC dropout. The ANM is then asked to validate the prediction based on her experience, while retaining control over decision-making.
🟥🟩 Simplified color scheme: Red (risk), Green (no risk)
Replaced the initial red-yellow-green scale to reduce confusion. In healthcare, any risk is important a binary color system proved clearer for low-literacy users.
🟥🟩 Simplified color scheme: Red (risk), Green (no risk)
Replaced the initial red-yellow-green scale to reduce confusion. In healthcare, any risk is important a binary color system proved clearer for low-literacy users.
🟥🟩 Simplified color scheme: Red (risk), Green (no risk)
Replaced the initial red-yellow-green scale to reduce confusion. In healthcare, any risk is important a binary color system proved clearer for low-literacy users.
❌ Explanation of prediction not shown
The AI model didn’t return interpretable reasons at this stage. While this limited transparency, the validation step helped retain trust by giving ANMs a voice.
❌ Explanation of prediction not shown
The AI model didn’t return interpretable reasons at this stage. While this limited transparency, the validation step helped retain trust by giving ANMs a voice.
❌ Explanation of prediction not shown
The AI model didn’t return interpretable reasons at this stage. While this limited transparency, the validation step helped retain trust by giving ANMs a voice.
👩⚕️ Gave ANMs a say in the decision-making process
The model explicitly asks for the ANM’s input (“Yes/No”) before taking any action, reinforcing trust and respect for their field knowledge.
👩⚕️ Gave ANMs a say in the decision-making process
The model explicitly asks for the ANM’s input (“Yes/No”) before taking any action, reinforcing trust and respect for their field knowledge.
👩⚕️ Gave ANMs a say in the decision-making process
The model explicitly asks for the ANM’s input (“Yes/No”) before taking any action, reinforcing trust and respect for their field knowledge.
✅ “Agree/Disagree” → “Yes/No”
Reworded based on field feedback — ANMs felt “Disagree” sounded like they were being blamed. “Yes/No” was neutral and more approachable.
✅ “Agree/Disagree” → “Yes/No”
Reworded based on field feedback — ANMs felt “Disagree” sounded like they were being blamed. “Yes/No” was neutral and more approachable.
✅ “Agree/Disagree” → “Yes/No”
Reworded based on field feedback — ANMs felt “Disagree” sounded like they were being blamed. “Yes/No” was neutral and more approachable.
Intervention Recommendation – Counselling + Action Flow
After the prediction is validated, the app presents context-specific intervention guidance, tailored to the woman’s ANC visit stage (1st, 2nd, or 3rd). This screen bridges the gap between insight and follow-up by surfacing counselling content + next steps, designed to support ANMs in the moment of care.
🔺 AI Prediction Banner (Risk Result)
At the top, the app clearly flags if the woman is at risk of dropping out before the next ANC visit allowing the ANM to shift focus accordingly.
🔺 AI Prediction Banner (Risk Result)
At the top, the app clearly flags if the woman is at risk of dropping out before the next ANC visit allowing the ANM to shift focus accordingly.
🔺 AI Prediction Banner (Risk Result)
At the top, the app clearly flags if the woman is at risk of dropping out before the next ANC visit allowing the ANM to shift focus accordingly.
🗣️ Counselling Cards Based on ANC Stage & Motivation
Messages are tailored to each ANC visit and designed to motivate women to return. Written in simple Hindi with visuals, they help ANMs counsel with ease.
🗣️ Counselling Cards Based on ANC Stage & Motivation
Messages are tailored to each ANC visit and designed to motivate women to return. Written in simple Hindi with visuals, they help ANMs counsel with ease.
🗣️ Counselling Cards Based on ANC Stage & Motivation
Messages are tailored to each ANC visit and designed to motivate women to return. Written in simple Hindi with visuals, they help ANMs counsel with ease.
🧍♀️ Supports Human-Centered Action
Recommendations are designed as nudges, not commands. The goal is to support ANMs, not automate them. The action steps act as a checklist.
🧍♀️ Supports Human-Centered Action
Recommendations are designed as nudges, not commands. The goal is to support ANMs, not automate them. The action steps act as a checklist.
🧍♀️ Supports Human-Centered Action
Recommendations are designed as nudges, not commands. The goal is to support ANMs, not automate them. The action steps act as a checklist.
🧍♀️ Human-Centered Language in Local Script
Messages are in simple Hindi, familiar clinical language for low-literacy comprehension and verbal delivery.
🧍♀️ Human-Centered Language in Local Script
Messages are in simple Hindi, familiar clinical language for low-literacy comprehension and verbal delivery.
🧍♀️ Human-Centered Language in Local Script
Messages are in simple Hindi, familiar clinical language for low-literacy comprehension and verbal delivery.
ANC 1 - Intervention Screens
Stage-Based Intervention Design for ANC Care
ANC 1
Focus: Investigations, Medicines, BPCR, CHO care
Actions:
• Start BPCR calendar
• Counsel for key tests
Counselling Cue:
Highlight tests needed + monetary benefits tied to completing all visits.
ANC 1
Focus: Investigations, Medicines, BPCR, CHO care
Actions:
• Start BPCR calendar
• Counsel for key tests
Counselling Cue:
Highlight tests needed + monetary benefits tied to completing all visits.
ANC 2
Focus: Fetal health, teleconsultation
Actions:
• Refer for fetal scan
• Schedule MO/CHO call
Counselling Cue:
“You’ll hear your baby’s heartbeat next time.” Motivate using emotional + financial incentives.
ANC 2
Focus: Fetal health, teleconsultation
Actions:
• Refer for fetal scan
• Schedule MO/CHO call
Counselling Cue:
“You’ll hear your baby’s heartbeat next time.” Motivate using emotional + financial incentives.
ANC 3
Focus: Birth preparedness and planning, family support
Actions:
• Plan for institutional delivery
• Involve family in prep
Counselling Cue:
Stress safe delivery + completion of visits for final scheme benefits.
ANC 3
Focus: Birth preparedness and planning, family support
Actions:
• Plan for institutional delivery
• Involve family in prep
Counselling Cue:
Stress safe delivery + completion of visits for final scheme benefits.
Due list
The Due List serves as the starting point of the app, showing ANMs a prioritized view of patients who are at risk, due for follow-up, or missed a visit. This screen helps frontline workers act quickly, stay organized, and reduce loss to care.
🔢 Due Count Cards
Shows the number of women needing follow-up (e.g., “7 DUE”), helping ANMs prioritise better.
🔢 Due Count Cards
Shows the number of women needing follow-up (e.g., “7 DUE”), helping ANMs prioritise better.
🔢 Due Count Cards
Shows the number of women needing follow-up (e.g., “7 DUE”), helping ANMs prioritise better.
🎯 Action-Oriented Labels
Subtext prompts clear next steps: “Ensure counselling by CHO since these women might not visit again.”
🎯 Action-Oriented Labels
Subtext prompts clear next steps: “Ensure counselling by CHO since these women might not visit again.”
🎯 Action-Oriented Labels
Subtext prompts clear next steps: “Ensure counselling by CHO since these women might not visit again.”
👩🏽 At-a-Glance Patient Info
Includes name, MP ID, last visit date, referral status, and risk flags (e.g., Low Height | Diabetes | HIV).
👩🏽 At-a-Glance Patient Info
Includes name, MP ID, last visit date, referral status, and risk flags (e.g., Low Height | Diabetes | HIV).
👩🏽 At-a-Glance Patient Info
Includes name, MP ID, last visit date, referral status, and risk flags (e.g., Low Height | Diabetes | HIV).
🔍 Smart Filters
ANMs can filter by village or ASHA, or search directly by name/ID to find patients fast.
🔍 Smart Filters
ANMs can filter by village or ASHA, or search directly by name/ID to find patients fast.
🔍 Smart Filters
ANMs can filter by village or ASHA, or search directly by name/ID to find patients fast.
IMPACT AND LEARNINGS
Key Usability Outcomes
Key Usability Outcomes
Tested with 20 ANMs across 40 ANC visits in Jabalpur and Mandla districts
Design for trust, not just usability: Clear visual cues, responsive devices, and predictable flows helped ANMs feel in control — and confident in their own ability to deliver care. Even the most hesitant users — especially older ANMs — adapted quickly when the tools fit seamlessly into their daily routine.
Field-first design means offline-first thinking: Connectivity can't be assumed. Every action — from data entry to device pairing — had to succeed without the internet. Designing offline-first wasn’t just a feature choice, it was a reality check. It ensured the tools worked exactly where they were needed most.
Emotional ease is part of good UX: Confidence grows when training is safe, feedback is supportive, and mistakes are part of learning — not reasons to disengage.
System alignment fuels sustainability: It wasn’t enough to help the ANM. The solution had to make sense for BMOs, CHOs, and dashboard viewers too. Designs that fit neatly into reporting structures, supervision systems, and government priorities are the ones that last beyond pilots.
Recognition fuels consistency: Motivational strategies like certificates and public appreciation had a real impact — reinforcing usage through pride, not pressure.
✅ What Worked Well
• 95% of ANMs completed the full ANC flow independently
• 18 out of 20 found Due List and Counselling Cards most helpful
• Switching to “Yes/No” improved clarity for 100% of users
• All 20 ANMs preferred Red/Green over multi-color risk labels
• 16 ANMs felt more confident using the Hindi interface
• Offline mode + auto-save worked in all ANC visits with zero data loss
📊 Field Outcomes (in pilot blocks):
• 90%+ device adoption by ANMs
• 40–60% reduction in data heaping (BP, Hb, weight)
• +7% increase in high-risk pregnancy detection
• Significant improvement in work-life balance (less late-night data entry)
🏛️ System-Level Outcome:
• NHM approved scale-up for scale in pilot state.
• National level Directive Order for Creation of standardized comprehensive kits for all health cadres.
🔧 What Could Be Improved
Model prediction time (20–35 seconds) caused wait-time anxiety
→ Needs faster inference or clearer progress feedback reduced to 5-10 seconds but needs further improvement.
No reason shown for prediction reduced AI transparency
→ Users wanted to understand why the patient was marked as “at risk”.
No notification/reminder system
→ ANMs requested reminders for follow-ups and overdue visits
📊 Field Outcomes (in pilot blocks):
• 90%+ device adoption by ANMs
• 40–60% reduction in data heaping (BP, Hb, weight)
• +7% increase in high-risk pregnancy detection
• Significant improvement in work-life balance (less late-night data entry)
🏛️ System-Level Outcome:
• NHM approved scale-up for scale in pilot state.
• National level Directive Order for Creation of standardized comprehensive kits for all health cadres.
What I Learned
What I Learned
Trust is the real deliverable: Especially in healthcare, users need to feel in control of AI and not overruled by it.
Design with, not for: Co-design led to better adoption than top-down instruction
Healthcare AI must support, not replace, human judgment: Designing human-in-loop validation gave ANMs confidence without removing their agency.
AI fails if it doesn’t fit into existing workflows: Success meant integrating seamlessly into how care is already delivered.
Designing for healthcare means designing for edge cases: You’re building for low network, low literacy, emotional stress, and system gaps all at once.
Explainability is a UX problem: Simplifying AI outputs made them usable
Design for trust, not just usability: Clear visual cues, responsive devices, and predictable flows helped ANMs feel in control — and confident in their own ability to deliver care. Even the most hesitant users — especially older ANMs — adapted quickly when the tools fit seamlessly into their daily routine.
Field-first design means offline-first thinking: Connectivity can't be assumed. Every action — from data entry to device pairing — had to succeed without the internet. Designing offline-first wasn’t just a feature choice, it was a reality check. It ensured the tools worked exactly where they were needed most.
Emotional ease is part of good UX: Confidence grows when training is safe, feedback is supportive, and mistakes are part of learning — not reasons to disengage.
System alignment fuels sustainability: It wasn’t enough to help the ANM. The solution had to make sense for BMOs, CHOs, and dashboard viewers too. Designs that fit neatly into reporting structures, supervision systems, and government priorities are the ones that last beyond pilots.
Recognition fuels consistency: Motivational strategies like certificates and public appreciation had a real impact — reinforcing usage through pride, not pressure.
IMPACT AND LEARNINGS
Key Usability Outcomes
Tested with 20 ANMs across 40 ANC visits in Jabalpur and Mandla districts
✅ What Worked Well
• 95% of ANMs completed the full ANC flow independently
• 18 out of 20 found Due List and Counselling Cards most helpful
• Switching to “Yes/No” improved clarity for 100% of users
• All 20 ANMs preferred Red/Green over multi-color risk labels
• 16 ANMs felt more confident using the Hindi interface
• Offline mode + auto-save worked in all ANC visits with zero data loss
🔧 What Could Be Improved
Model prediction time (20–35 seconds) caused wait-time anxiety
→ Needs faster inference or clearer progress feedback reduced to 5-10 seconds but needs further improvement.
No reason shown for prediction reduced AI transparency
→ Users wanted to understand why the patient was marked as “at risk”.
No notification/reminder system
→ ANMs requested reminders for follow-ups and overdue visits
What I Learned
What I Learned
Trust is the real deliverable: Especially in healthcare, users need to feel in control of AI and not overruled by it.
Design with, not for: Co-design led to better adoption than top-down instruction
Healthcare AI must support, not replace, human judgment: Designing human-in-loop validation gave ANMs confidence without removing their agency.
AI fails if it doesn’t fit into existing workflows: Success meant integrating seamlessly into how care is already delivered.
Designing for healthcare means designing for edge cases: You’re building for low network, low literacy, emotional stress, and system gaps all at once.
Explainability is a UX problem: Simplifying AI outputs made them usable
Other projects
Other projects
ANMOL Sakhi: Supporting Frontline Nurses with IoT-Based Maternal Care
UX Research & Design for Scale | From Pilot to National Adoption
Editage 2.0- Redesign for trust
Redesigned Editage’s digital experience with a focus on establishing trust and clarity for global researchers.
Editage 2.0- Improving the form funnel
Enhancing clarity and credibility for global researchers
The Antenatal Wheel: Bridging Knowledge and Care in Rural Pregnancy Health
A Nation-Scale Job Aid Empowering frontline healthcare workers
Copyright 2025 by Snehashri Panda
Copyright 2025 by Snehashri Panda
Copyright 2025 by Snehashri Panda