# Constituency survey phone calls: how to run a fair one

> Constituency survey phone calls ask residents a short set of neutral questions about local problems, such as water, roads or drainage, and record each answer as a field you can count. To get honest results, sample fairly across wards and age groups, ask questions that do not lead, tell people who is calling and why, and let anyone decline. An AI caller can complete hundreds of calls a day in Telugu, Hindi or English.

Source: https://agents.elevatebox.in/blog/constituency-survey-phone-calls  
Updated: 2026-10-15

## Why phone surveys beat guesswork

An MLA's office, a municipal corporator or a civic group usually knows the loudest complaints. What it rarely knows is how common each problem is. Is the water supply issue in three colonies or thirty? Do older residents care about the same things as young families? A short, well-run phone survey answers that in a week.

The catch is that a careless survey is worse than none. If you only call party members, or ask "Don't you agree the roads are terrible?", you will get a confident answer that is wrong. This post is about doing it fairly.

## Sample design in plain words

You cannot call everyone, so you call a sample that looks like the whole constituency. Four simple rules cover most of it:

1. **Spread across wards.** Decide how many completed calls you want from each ward, roughly in proportion to its population.
2. **Balance age and gender.** Set rough quotas, for example a target share of women and of people over 50, and keep calling until each is met.
3. **Pick numbers at random from your list.** Do not start from the top of a sheet sorted by booth or by surname.
4. **Call at different times.** People who are home at 11 am are not the same as people who are home at 7 pm.

How many calls? A common rule of thumb is that about 400 completed interviews give a margin of error of around plus or minus 5 percentage points for the whole area. If you want to compare wards, each ward needs a meaningful sample of its own. Fewer than 100 in a ward and the numbers swing too much to trust.

Phone surveys also miss people without phones or who never pick up unknown numbers. You cannot fully fix that, but you can say so when you report, and you can cross-check big findings with a few ward visits.

## Writing questions that do not lead

The wording matters more than the technology. Compare these pairs:

| Leading (avoid) | Neutral (use) |
| --- | --- |
| Don't you think the corporator has ignored your colony's drainage? | How would you rate drainage in your colony: good, average or poor? |
| Water supply is the biggest problem here, right? | What is the one problem in your area you would most like fixed? |
| Would you support the new flyover that will end traffic jams? | Have you heard about the planned flyover? If yes, do you support it, oppose it, or are you not sure? |

- Keep it to six to eight questions. After four minutes people start giving any answer to finish.
- Always offer "not sure" or "don't know". Forcing a choice adds noise.
- Ask the open question early, before your list of options plants ideas.
- Never ask about caste, religion or who someone voted for unless you have a strong reason and clear consent.

## What the call sounds like

An opening in Telugu might be: "Namaskaram, nenu ward samasyala survey kosam call chestunna AI assistant ni. Rendu nimishalu matladagalara?" (Hello, I am an AI assistant calling for a survey on ward problems. Could you spare two minutes?) In Hindi: "Namaste, main aapke ward ki samasyaon par ek chhote survey ke liye AI assistant bol rahi hoon." (Hello, I am an AI assistant calling for a short survey on your ward's problems.)

If the person says they are busy, the AI employee offers a better time and calls back then. If they say no, it thanks them and marks the number so it is not called again. That one rule protects your reputation more than any script.

## Turning answers into fields you can count

Before you start, define each answer as a captured field. The AI employee listens to the reply, in whatever words the person uses, and fills the field. "Neellu vaaraniki okasari vastayi" (water comes once a week) becomes Water supply: poor, with the detail saved in the transcript.

- Ward or colony
- Top problem (from a fixed list, with "other" and the detail)
- Rating of water, roads, drainage and street lights
- Age group and gender, if the person chooses to share
- Willing to be contacted again: yes or no

Export these to a sheet and you have counts by ward the same evening. Listen to a sample of recordings each day to check the fields match what people actually said. If your team also needs volunteer and booth calls, read [voter calling in Telugu](https://agents.elevatebox.in/blog/voter-calling-software-telugu).

## Reading the results honestly

Numbers from a survey look precise, so people over-read them. A few habits keep you honest:

- **Report the sample with the result.** "42% rated water supply poor (312 calls, Ward 14)" tells the reader how much weight to give it. "42% are unhappy" does not.
- **Check who did not answer.** If most pickups were men over 40, your result describes men over 40. Weight the result or call more of the missing group.
- **Small gaps are noise.** With a few hundred calls, 38% against 41% is not a real difference. Look for gaps of ten points or more.
- **Read the open answers.** The fixed fields count what you expected. The open question finds what you did not, like a new garbage dump nobody had raised at meetings.
- **Repeat, do not compare apples and oranges.** If you run the survey again in three months, keep the same questions in the same order so the change means something.

Share the method along with the findings when you present them to residents or the press. A survey that hides its method gets dismissed, even when it is right.

## Consent, numbers and timing

Use numbers you are allowed to call: people who registered at a meeting, residents' welfare association lists shared with consent, or your own helpline callers. Respect the DND register. Call within sensible hours; ElevateBox defaults to 9 am to 8 pm and checks the do-not-call list before every dial. The details are in [calling policy](https://agents.elevatebox.in/help/calling-policy).

If an election has been announced, a survey that asks for votes or praises a candidate is campaigning and falls under the Election Commission's rules, including the 48-hour silence period before polling. Keep surveys about problems, not parties.

## Cost of a 1,000-call survey

Say you want 1,000 completed interviews and each takes about three minutes. That is 3,000 minutes, or about ₹23,940 of talk time at ₹7.98 a minute, before GST. Unanswered calls are not charged, so a list with many non-pickups does not add to the bill. Use [bulk campaigns](https://agents.elevatebox.in/help/bulk-campaigns) to upload the list and set retries.

> Run a small pilot of 50 calls first, listen to every one, fix the wording, then scale. New accounts get ₹50 free to try. [See how it works](https://agents.elevatebox.in/how-it-works).

## Frequently asked questions

### How many people should a constituency survey call?

About 400 completed interviews give a margin of error near 5 percentage points for the whole constituency. If you want to compare wards, each ward needs its own sample of at least around 100, so plan more calls in total.

### Can an AI caller run a survey in Telugu and Hindi?

Yes. The AI employee asks the questions in natural Telugu, Hindi or English, understands everyday replies including mixed language, and fills each answer into a field. You get a recording and transcript of every call to check accuracy.

### How do I stop a phone survey from being biased?

Sample across wards, ages and genders on purpose, pick numbers at random, call at different times of day, and word questions neutrally with a "not sure" option. Review recordings daily to catch questions people misunderstand.

### Is a ward problems survey allowed during elections?

A survey about local problems is different from campaigning, but once a survey praises a candidate or asks for votes it is campaigning and the Election Commission's rules apply, including the silence period. Keep it about issues, and check with your election agent if unsure.