People say more out loud than they will ever write down.

Written surveys lose exactly the people whose experience matters most — a nurse between rounds, a driver at a depot, a customer who is annoyed and has thirty seconds. They tick three boxes to make it stop, and the one field that could have told you something comes back empty.

• A rating is where the answer stops

  Four out of five tells you nothing you can act on this week. The sentence that comes after it does, and a form has no way to ask for it.

• Writing is a tax on the people you most need to hear

  A keyboard in a second language, on a cracked screen, at the end of a shift. Speaking costs the respondent nothing they were not already able to do.

• The best answer rarely fits the box you built

  The most useful thing a respondent says is usually a reply to a question nobody thought to ask. Only a conversation can go there.

One link, one page. The respondent reads what will be recorded and why, agrees or leaves, allows the microphone, and starts talking. The voice agent asks the questions your study is built from, listens, and probes where an answer was thin — the way a good interviewer would.

• Their language, not yours

  Hindi and English are live today. The speech models behind the agent cover all major Indian languages, and we verify them study by study rather than claiming the whole list at once. A respondent can switch language part-way through.

• It asks the follow-up

  When someone says the wait was “a bit long”, the agent asks how long, and where. That second answer is the one an operations team can do something with.

• About five minutes, then it ends cleanly

  A thank-you, and a way to opt out of ever being contacted again — honoured within a day, and against the person rather than the row, so it survives your next upload.

The respondent page is judged on the worst phone and the weakest connection in your list, not the best one. Nothing installs. It runs in the browser already on the device — Chrome on Android, Safari on iPhone — and it is built to hold a conversation on 3G.

• No app and no account

  An install step loses most of a list before the first question is asked. There is nothing to download, nothing to remember and no password to reset.

• Nothing else running in the background

  No third-party scripts on the page a respondent answers from. It is faster, and it means nobody else is watching your people answer.

Every finished conversation passes one fixed gate before it is counted: consent actually given, long enough to be an interview, enough of the required questions answered, speech the models could hear, and not a repeat of someone who already answered. A silent thirty seconds is never billed.

Every report opens with the same methodology block — sample source, field dates, contacted, valid and rejected counts, language and channel mix, and the line that findings are directional only unless quota controls were applied and met.

A live call is not always the right moment — a bad line, a shared phone, or simply a preference not to hold a conversation with someone waiting on the other end. When that happens, the respondent can record a voice message instead: the same questions, asked one at a time, answered whenever they are ready and at whatever pace suits them.

• No one is waiting on the line

  They record an answer, listen back if they want to, and move to the next question when they are ready — there is no agent pacing them through it in real time.

• The same transcription and analysis run either way

  A recorded answer is transcribed, scored and reported exactly like a live one. The methodology block states the mix of live calls and recorded messages, so nobody has to guess how the sample was collected.

• Still voice, not a separate channel

  Recorded and live sit under the same voice channel. A study built to accept voice accepts either; a respondent is never asked to install anything extra to use one over the other.