Skip to content
PoliticsPilotSTRATEGY. PEOPLE. PROGRESS.
Intelligence · Service

AI Voice Call Survey

A field enumerator costs ₹80–₹150 per completed interview and takes six to eight weeks to canvass an assembly seat. An AI voice bot completes the same interview for ₹9–₹14 in seventy-two hours, in the voter's dialect, and hands you a transcript you can re-analyse next week. It is not a replacement for a well-designed sample — it is what makes a well-designed sample affordable at scale.

Ask about this service +91 6366 82 9999
₹9–₹14
Cost per completed interview
18–26%
Completion rate on live pickups
72 hrs
Turnaround for a 40k-sample seat
01

What an AI voice survey actually sounds like

The voter answers the phone and hears a natural-sounding voice — not a menu, not a monotone recording — that identifies itself as an automated call from a research firm within the first five seconds and asks if the voter has two minutes. When they say yes, the bot conducts an eight-to-twelve question survey with real back-and-forth: it hears "the roads are terrible near the sugar factory", asks a follow-up about which specific road, and moves on. When they say no, it thanks them and hangs up. No pressure, no re-dial that day.

02

Where AI voice beats human enumerators, and where it does not

It wins on cost, speed, sample size and geographic reach — 40,000 completed interviews across a Lok Sabha seat in ten days is routine. It wins on consistency: the bot does not get tired, does not lead the respondent, and asks question seven the same way it asked question seven three hours ago. It loses on sensitive community questions and on voters over 65 in rural belts, who still respond better to a human enumerator. We usually mix the two — AI voice for the base 80% sample, field enumerators for the last 20% where the demographic is hard to reach digitally.

03

Languages, dialects and what the bot cannot yet do

We currently run production surveys in Kannada, Telugu, Tamil, Hindi, Marathi, Bengali and Malayalam with regional dialect tuning — Bidar Kannada is not Mysore Kannada, and the bot has to code-switch accordingly. Code-mixing (English words dropped into a Hindi sentence, or the reverse) is handled. Heavy sarcasm and idiomatic caste-community references are still hit-or-miss, which is why every batch has a 5% human back-check.

04

Compliance, consent and the DND question

Political survey calls are permitted to numbers on DND under TRAI rules, but the bot is briefed to hang up gracefully if the voter says they are on DND and did not expect this. The outbound number is DLT-registered under the campaign entity, the opening line names the survey firm and states the call is automated, and every completed interview is timestamped and audio-archived for the ECI expenditure register.

FAQ

AI questions, answered directly

Do voters realise it is an AI, and does that hurt response rates?

Yes and no. We are legally and ethically required to disclose it is an automated call in the opening line, and we do. In production this drops completion rate by 3–4 percentage points versus a human enumerator on the same script — a fraction of the cost saving. Voters who continue past the disclosure are engaged voluntarily and their data is cleaner for it.

How is this different from an IVR press-1 poll?

IVR gives you closed-ended, pre-scripted answers — press 1 for water, press 2 for roads. AI voice conducts a real conversation, hears open-ended responses in the voter's own words, and follows up. Response depth is 4–6x richer; response rate is comparable. Cost is 2–3x an IVR poll and one-tenth of a field enumerator.

Can the bot detect that the same voter is being surveyed twice?

The dial list is deduplicated at input, and the bot cross-checks against the last-90-days call log before dialling. Repeat calls within a campaign are capped at two, spaced at least six weeks apart.