The first time we sat down with a maternal-health team to design their voice line, someone pulled up a flowchart. It had thirty-eight branches. "Press 1 for antenatal. Press 2 for postnatal. Press 3 for danger signs. Press 4 for…" and so on for two pages of PDF. It looked like a call-centre script, because that's the reference material every IVRS RFP in the sector was written against.

Here's the problem: the person calling this line is a first-time mother in her second trimester who has ten minutes on a shared phone in a courtyard. She doesn't know what "antenatal" means. She doesn't know what her risk category is. She knows something feels wrong, or she has a question she doesn't want to ask anyone in the courtyard. A phone tree with thirty-eight branches makes her hang up in the first forty-five seconds.

Voice is a medium. Design for it.

A phone tree is a database lookup with a keypad. That works when the caller already knows the taxonomy — press "billing" because you know you have a billing problem. It fails when the caller doesn't know what to call the thing that's wrong. In our sector, that's most of the time.

The lines that actually get used are the ones that behave like a person. They start with a welcome that acknowledges what week of pregnancy the caller is in (we know, because she registered). They ask one plain-language question at a time. They repeat things. They let her go back. They don't punish silence.

Three design principles from the field

1. One question per level, always.

If a level of the tree has more than three options, it's not a level. It's a wall. Break it into two levels of two, or three of two. The caller's short-term memory is one option. Give her one.

2. Voice acting matters more than the flow.

We've watched two versions of the same script — same questions, same order, same routing — perform completely differently. The one recorded by an ASHA from the district got 3× the completion rate of the one recorded by a professional voice actor from the capital. Accent, cadence, warmth. That's the actual product.

3. Design for the drop-off, not the completion.

Most callers won't finish the flow. That's fine. The question is: what did they get before they hung up? If the first thirty seconds carried the most important piece of information — a danger sign, a helpline number, a reassurance — you did your job even if she never got to level five.

"The IVRS line reached people in three states our outreach workers had never been able to call. Honestly the most useful partnership we've had in five years."

The uncomfortable question

Voice is expensive to build well. It needs local recording, multiple language passes, careful QA on rural connections, an on-call fix for the moment a level breaks. If your programme isn't willing to invest in that, don't build the line. A bad voice line is worse than none — it burns the trust of the caller, and she doesn't come back.

But when it's built right, voice reaches people that no other channel does. Not the app. Not SMS. Not the website. The person who has ten minutes and a shared phone and a question she doesn't want to ask aloud. That person is worth building for.