WhatsApp is the interface, whether you designed for it or not

Every construction organisation in India has an official system for site reporting, and a WhatsApp group where the reporting actually happens.

The official system has forms, fields, validation and a login. The group has photographs, voice notes, half sentences and a great deal of information that never reaches the official system at all.

Most software projects in this space begin by trying to move people from the second to the first. It is a reasonable-sounding plan and it does not work.

Why the group wins#

Not because people are resistant to change. Because on the measures that matter to somebody standing in the sun, the group is genuinely better.

It is already open. No login, no app switch, no waiting for a page.

It works on a bad connection, and it queues and retries without being asked.

It accepts a photograph as a complete message. A picture of a delivery slip contains everything the form was going to ask for, and takes two seconds.

It accepts speech. A site engineer sends a voice note because his hands are dusty and the screen is in the sun.

Everybody is already in it, including the people who will never be given a licence for your software — drivers, contractors' supervisors, the fabricator.

It is forgiving. No field is mandatory. Nothing is rejected for being incomplete.

A form on a phone competing with that loses on every axis. The competition is not close, and no amount of training changes an outcome that is determined by physics and attention.

The move that works#

Stop competing. Read the group.

Treat the messages as the input and do the structuring afterwards, on your side, where somebody has a keyboard and a moment. The person on site changes nothing about what they do.

This is not a lowering of standards. The structured record still gets created, still gets validated, still gets confirmed by a person. What changes is who does the structuring and when, and moving that work off the site and into the office is almost always correct, because the site is the place with the least available attention on the whole project.

What we got wrong first#

Our software recognised a photograph as a delivery record only if its caption contained one of four words: challan, bilty, GRN, delivery.

Measured against four hundred real captions that the site team had actually sent, those four words matched exactly two.

What people really write is the material and a quantity — a grade and a volume and a weight. Or the vehicle and the fact that it has arrived. Or a plain list: a trader's name and then two numbered lines with quantities and units.

Reading those three shapes instead — the material named directly, the vehicle that brought it, or any quantity with a unit — took the match rate from two to thirty-four out of the same four hundred captions.

The vocabulary was never the site's problem. It was ours. A longer piece on that is the words the site already uses.

Fifteen hundred photographs had been sent from site before this was fixed, and not one had ever become a delivery record. That is why no quantity appeared anywhere. Nothing had failed loudly. The photographs kept arriving and kept producing nothing.

Voice notes are the same problem again#

Once the pictures were being read, the voice notes were still a labelled blob that nobody opened.

The comment in our own code had known why for over a year — a site engineer sends a voice note because his hands are dusty and the screen is in the sun — and we had done nothing about it.

Transcribing them was the obvious half. The half that mattered was what happens to the words afterwards: a spoken incident report classifies and registers, a spoken answer to an evening question is treated as that answer, a spoken question gets an answer grounded in the records. Same doors, same functions as a typed message takes. Not a parallel path, by construction — because a parallel path is a second implementation of every rule, and second implementations drift.

That piece has its own account: the voice note is the report.

The rules that keep it honest#

Reading a group chat as an input has real risks, and three rules cover most of them.

Everything the machine concludes is a suggestion. A photograph that looks like a delivery slip becomes a proposed record, matched to a proposed order, which a person confirms with one tap. Never an automatic entry. The reasoning is in the machine proposes, a person decides.

Refuse the obvious non-cases loudly rather than guessing. A cylinder going out for refilling, a photograph of a damaged divider, a weight dispute — these look superficially like delivery messages and are not. A system that files them as receipts is worse than one that files nothing, because somebody now has to find and remove them.

Privacy is a boundary, not a filter. This one we got wrong and had to fix. Our text extraction ran before the check that knew which rooms were private, so a personal photograph was being read and, if its text resembled a delivery slip, filed as a goods receipt. The correction was not to discard the result afterwards — it was to exclude private rooms at the query, so the bytes are never fetched at all. A privacy control that operates after processing is not a privacy control.

What you still need the real system for#

Reading the chat does not remove the need for a system of record. It changes what the system of record is for.

The chat is where events are reported. The system is where they are confirmed, structured, connected to money and orders, and kept. Signature, authority, audit trail and reconciliation all belong on the system side, and none of them can live in a group chat, for reasons set out in what an audit trail is for.

The division that works: the site reports in the medium it already uses; the office confirms in the system that keeps records. Nobody is asked to change their behaviour to suit software.

The direction back out carries a decision of its own — whether a message leaves from the company or from the colleague the other side already knows: one WhatsApp number, or one per person.

The short version#

The group chat is not a workaround people will grow out of. It is the interface, because it wins on every dimension that matters to somebody on a site.

Read it. Read it in the vocabulary people actually use, not the vocabulary your forms were written in. Treat everything the machine concludes as a proposal. And keep the record-keeping where records belong.