MakeMyPoll AI Caller gives campaigns booth-level voter intelligence at scale — proxy questions, overnight calling, and a live war room for MLAs, MPs, and municipal leaders. Deploy in any constituency; explore how it works in our North Mangalore case study below.
North Mangalore is our reference deployment — a coastal Karnataka constituency with fishing belts, minority pockets, cashew workers, and Tulu-speaking hinterlands. These six stories show the intelligence MakeMyPoll surfaces in a real campaign environment.
STORY 01
Civic Issue
The Fishermen's Jetty Neglect
MLA campaigns on a new ring road. But AI calls to Old Bunder and Thokkottu fishing households reveal 60% are furious about a rotting jetty that's destroying their nets and income. The highway means nothing to them.
"Campaign has pivoted to the wrong issue. Emergency MLALAD intervention at jetty required before week 3."
STORY 02
Trust Deficit
The Corrupt Contractor Signal
Approval drops 18% in Konaje panchayat over 3 weeks. AI caller using proxy questions in Tulu uncovers that a road contractor affiliated with the MLA's office is pocketing funds while leaving roads unpaved. Villagers blame MLA.
"Identify contractor name, initiate public audit. Hold Janata Darbar in Konaje. Approval recoverable within 6 weeks."
STORY 03
Caste Arithmetic
The Forgotten Kudubi Community
Three booths in Mulky-Moodubelle are historically 70% opposition. AI demographic scan + calls reveal Kudubi tribal community (28% of those booths) feel completely abandoned by both sides. Zero political outreach in 5 years.
"Private meeting with Kudubi elders. Promise a forest rights facilitation camp. Appoint one youth leader. Flip 28% → neutralize booths."
STORY 04
Laabharthi Economy
The Housing Scheme Middleman
PM Awas scheme beneficiaries in Surathkal and Panambur are receiving their keys, but AI calls reveal that 34% believe the ward corporator (not the MLA) delivered it. Credit attribution failure. Corporator is actually opposition-aligned.
"MLA must personally do key-handover ceremonies with video. Reframe narrative — beneficiary gratitude is leaking to wrong party."
STORY 05
Silent Voter
"Angry but Loyal" Minority Belt
Muslim voters in Ullal and Talapady show 0% swing intention in direct questions. But proxy questions ("what did your neighbor say about last week's speech?") reveal deep anger about a delayed masjid land dispute — loyalty is thin ice.
"Don't assume this vote is safe. Send a trusted community liaison to resolve land dispute quietly, 8 weeks out."
STORY 06
Booth Intelligence
The Teacher Vote Bloc
AI calls to government school teachers (a dense cluster in Deralakatte) uncover uniform frustration: pending DA arrears unpaid for 14 months. Teachers are highly networked opinion leaders. Their anger is spreading to 4 surrounding booths.
"Escalate DA arrears to state finance through MLA channel. Acknowledge publicly at a teachers' event. 1 action, 6-booth impact."
02 — Demo Modules
Four AI Caller Demo Types
Four calling modes used in the North Mangalore case study — each with its own conversational logic, voice persona, and intelligence output. They can run simultaneously across voter segments in any constituency.
📞
Pulse Poll Caller
Mode: Rapid Sentiment Tracking
Short 3-minute calls to ~500 voters/day across all booths. Measures real-time approval movement with proxy questions in Tulu/Kannada/Beary. Tracks the "angry but loyal" gap and detects sudden drops before they show up in field reports.
pulse_poll_tulu.wav — LIVE TRANSCRIPT
Namaskara. Naanu North Mangalore constituency voter survey maaḍuttidde. 2 nimisha maatanaaDaLaaguttada?[Tulu/Kannada response accepted]Nimma areaalli yaar MLA kaaryakke jaasthi mannaneyitidaare anta janaru heLtaare?PROXY: "who do people say" — not "who will you vote for"Nimma oodina haadiyalli keLe issues yaavalaadaroo ive? (road / water / power / other)ISSUE MAP → Booth cluster flagged if 3+ same issue in 48hr
Edge Cases to Capture
Voter says "I don't want to talk" → log as "reluctant," flag booth if >20% in area
Voter mentions a name unprompted → high-signal, route to issue caller
Angry tone detected → escalate to human follow-up within 24h
Language switches mid-call (Tulu → Beary Basha) → AI adapts, logs community identity signal
🔍
Deep Issue Excavator
Mode: Qualitative Focus-Group Style
Longer 8-12 minute calls to 80-100 targeted voters per week. Mimics a door-to-door conversation — asks layered follow-up questions, lets the voter vent, and extracts the real underlying grievance behind generic complaints.
deep_excavator_konaje.wav
You mentioned the road near your house has a problem. Can you tell me what exactly happens when it rains?"The drain overflows, water comes into my shop, I've complained 3 times—"Who did you complain to? Did anyone visit after that?"To the contractor fellow. He came once, took photos, nothing happened."FLAG: Contractor name captured. Cross-ref MLALAD records. Possible misappropriation.Do you feel the MLA knows about this problem?BRANCH: If "no" → trust recoverable. If "yes, but does nothing" → serious crisis.
Edge Cases to Capture
Voter mentions "contractor took money" → auto-log as corruption signal, escalate immediately
Voter expresses fear of retaliation → AI acknowledges anonymity, logs as "suppressed grievance" zone
Voter brings up a rival party's local work → competitive threat alert
Voter hangs up mid-complaint → call back in 3 days with different intro
📊
Laabharthi Gratitude Mapper
Mode: Welfare Credit Attribution
Targeted calls to confirmed beneficiaries of PM Awas, Ujjwala, Ayushman, and state schemes. Measures whether voters attribute the benefit to the MLA, the party, the PM, or a middleman — and calibrates campaign messaging accordingly.
laabharthi_awas_surathkal.wav
I understand you recently received your PM Awas housing keys. Congratulations! Who handed over the keys to you?"The ward officer came with the papers."ALERT: MLA not present at handover. Credit gap detected. Recommend direct MLA ceremony.Do you feel like the MLA had a role in getting this approved?"I'm not sure, I heard it's the central government scheme."FINDING: "Right-not-gratitude" psychology. MLA messaging must humanize the last-mile.
Edge Cases to Capture
Voter attributes scheme to opposition — find who told them this, potential misinformation campaign
Voter reports pending issues with scheme (not received full amount) → urgent grievance, route to MLA office
Voter "feels obligated" (gratitude framing) vs "it was my right" → two different GOTV messages needed
Voter mentions a middleman took fees → corruption signal, escalate to dedicated investigation track
🗳️
Booth-Level Caste Sensor
Mode: Demographic & Social Cohesion
Calls to micro-targeted sub-communities within specific booths. Maps inter-caste tensions, neglected sub-groups, and influential elders. Feeds directly into booth management strategy — identifying who to meet, where, and with what ask.
booth_sensor_mulky_kudubi.wav
In your area, is there any community that feels they haven't received enough government attention?"Our Kudubi people have been asking for forest land rights for years. Nobody comes."MATCH: Kudubi population 28% in Booth 47-51. Zero political contact logged. High flip potential.If an MLA representative came to meet your community elders, what would be the main request?"Forest rights certificate and one tube well for Kumari colony."ACTION: Prepare targeted ask package. Schedule elder meeting. One appointment = 5 booths shifted.
Edge Cases to Capture
Voter describes inter-community conflict → avoid triggering that community in same campaign event
Voter mentions "we already met the opposition MLA" → immediate competitive threat; escalate timeline
Community leader identified by name → add to VIP contact list for direct MLA call
Voter suspicious of survey → note as "politically sensitized area," adjust future call scripts
03 — Proxy Intelligence
The Questions That Get Real Answers
Silent voters don't tell you who they'll vote for. These proxy questions — used by India's top pollsters — are built into every AI Caller script.
"Nimma area alli yaara rally ge jaasthi jana bandidru?"
Who had more crowd at their rally in your area? — Measures perceived momentum without asking voting intention. Crowd preference is 83% correlated with actual vote in Coastal Karnataka data.
Momentum Proxy
"Your neighbors — do they feel the MLA listens to them?"
Displaces the question from self to community. Voters freely criticize via "my neighbor's view" what they won't say directly. Projects true sentiment through social deflection.
Social Projection
"Ek kaam jo MLA ne nahin kiya — woh kya hoga?"
One thing the MLA hasn't done — open-ended, backward-looking. Surfaces the top unarticulated grievance that focus groups consistently show is the real vote decider.
Grievance Excavation
"If voting were tomorrow, what would make you stay home?"
Identifies the "angry abstainer" — a voter who won't switch parties but will simply not show up. In close constituencies, abstainer suppression is as important as vote conversion.
GOTV Risk Signal
"Yaar heLidru ondu kelasa aaytu anta?"
Who did you hear got a scheme benefit recently? — Maps the social spread of laabharthi information. If a voter says "neighbor got a house," the MLA's credit attribution work is succeeding.
Laabharthi Diffusion
"Rate the energy in your colony — 1 (defeated) to 5 (excited)"
Measures emotional temperature of a booth cluster. "Energy" language gets honest answers — voters describe their colony's mood more freely than their own. Predicts turnout better than stated intention.
Turnout Predictor
04 — Why AI Beats Human Pollsters
The Intelligence Gap
Human polling agencies in India have well-documented failure modes — Pradhan filters, social desirability bias, fatigue, limited booth coverage. Here's where the AI Caller structurally outperforms.
Intelligence Dimension
Human Agency
In-House Team
AI Caller
Scale per day Voters contacted in 24 hours
80–120
150–200
800–1,200
Pradhan/Middleman Filter Does data pass through local gatekeepers?
Always filtered
Often filtered
Zero filters — voter to MLA direct
Social Desirability Bias Do voters tell you what they think you want to hear?
Very high
High
Low — proxy questions designed in
Caste/Language Adaptation Can it switch between Tulu, Beary, Kannada mid-call?
Requires 3 surveyors
Usually only Kannada
Automatic, real-time
Night/Weekend Coverage Reaches working-class voters who are home only after 9pm
Rarely available
No
24/7 — runs at 11pm, 6am
Corruption Signal Detection Catches "cut money" or contractor fraud mentions
Suppressed by loyalty
Staff fears MLA's people
Auto-flagged in 48 hours
Speed to MLA desk How fast does raw intelligence reach decision-makers?
7–14 days (report cycle)
3–5 days
Real-time dashboard alerts
Cost per survey (₹)
₹180–320/voter
₹90–140/voter
₹8–22/voter
Recall & Audit Trail Can you replay what a voter actually said?
No — notes only
No
Full transcript + audio stored
Booth-level granularity Can data be sliced by individual polling booth?
Ward level at best
Ward level
Individual booth ID tracking
05 — The Math
What This Pays For
In our North Mangalore pilot (~1.8 lakh voters, ~200 booths), here is what a 90-day AI Caller programme looks like in numbers — representative of what your constituency could achieve.
54K
Voters Reached
30% of constituency contacted in 90 days. Statistically sufficient for booth-level predictions in every ward.
₹12L
Total Campaign Cost
vs. ₹85–95L for equivalent human agency coverage. 7× cost reduction with 12× scale advantage.
48H
Crisis Response Window
A contractor scandal detected Friday is resolved by Sunday's Janata Darbar — before it becomes a Monday headline.
200+
Booths Mapped
Every polling booth gets its own sentiment score, issue index, and community profile — updated weekly through the campaign.
06 — How It Works
From Call to Campaign Decision
The intelligence pipeline from AI voice call to MLA desk — in five steps, fully automated except for the final human decision.
01
Voter List Segmentation
Booths ranked by priority. Sub-caste, scheme beneficiary, and language clusters identified from electoral rolls + welfare data.
02
AI Calls Go Out
Pulse Poll, Deep Issue, Laabharthi, or Booth Sensor mode selected per segment. Tulu/Kannada/Beary scripts deployed automatically.
03
Real-Time Transcription
Every call transcribed. Sentiment scored. Corruption signals, grievance keywords, and caste mentions auto-tagged and time-stamped.
04
Booth Dashboard Updates
MLA's war room sees live booth-level maps. Red booths need attention. Rising issues flagged before they go viral.
05
MLA Takes Action
Targeted intervention: MLALAD funds diverted, Janata Darbar held, community elder called. Action timestamped against booth score improvement.
07 — Case Study Deep Dive
How We Configured North Mangalore
Every constituency has its own socio-linguistic fabric. This case study shows how MakeMyPoll was tuned for one seat — the same playbook adapts to yours.
Language Configuration
Tulu · Kannada · Beary Basha · Konkani
The AI Caller detects the caller's language in the first 5 seconds and shifts automatically. No voter is asked to repeat themselves in a language they're uncomfortable with.
Tulu: Primary for Bunt, Billava, Mogaveera communities in coastal belt
Beary Basha: Deployed in Ullal, Talapady Muslim pockets
Kannada: Government scheme discussions, Deralakatte interiors
Konkani: GSB community nodes in Mangaluru city fringe
Critical Voter Segments
Community Intelligence Nodes
Each community has a different activation trigger — the AI caller script adjusts its issue probes accordingly.
Fishing communities (Old Bunder, Thokkottu): Jetty / auction house issues
Cashew belt workers (Konaje, Bajpe): MNREGA delays, wage disputes
Auto/taxi drivers: Mangaluru city permit issues, traffic police
Teacher + ASHA worker bloc: DA arrears, workload complaints
Kudubi / tribal belt (Mulky fringe): Forest rights, land pattas
Minority pockets: Wakf land, masjid approvals, scholarship delays
Threat Intelligence
What to Detect Early
Competitive dynamics in this case study seat require specific early-warning sensors built into the calling scripts.
Rival MLA "invisible sops" — cash or saree distribution in target weeks
Opposition community meeting held without MLA knowledge
Viral WhatsApp forwards blaming MLA for a scheme failure
Local newspaper (Udayavani / Prajavani) narrative divergence from ground
Ward corporator defection signal — corporator helping opposition quietly
Campaign Pivot Triggers
When AI Tells You to Change Course
The dashboard has automatic alert thresholds that trigger strategic reviews.
Any booth drops 15%+ approval in 7 days → Emergency issue investigation
Single issue mentioned by 40%+ of callers in one panchayat → Issue pivot
"Contractor" or "cut money" mentioned 10+ times in week → Corruption audit
Scheme credit attributed to wrong party in 30%+ calls → Handover ceremony
Neglected community identified in swing booth → Elder outreach within 7 days
08 — Begin
Ready to Run AI Caller in Your Constituency?
Start with a 200-call pilot in 72 hours — two priority booth clusters of your choice. See a live dashboard with real voter intelligence before committing to the full programme.