Representative · Votix

How a union parishad campaign replaced twelve spreadsheets with one field system

2026-11-208 min readBy Black & White StudioLast reviewed 2026-11-20
In short

Twelve descendant copies of one spreadsheet, silently overwriting each other across nine wards. Record-level writes, offline capture and role-scoped access replaced them; the payoff is a turnout list that can refresh through polling day.

About this study: This is an illustrative composite, not a documented client engagement. It is assembled from conversations with local campaign organisers to show how field data problems typically present. The worker counts, household numbers and the electoral outcome are illustrative, not measured results, and no real voter data was involved in preparing it. No claim is made that any software affects an election result.
38
Field workers
9
Wards
~11,000
Households canvassed
2 hrs
Turnout list refresh

Illustrative figures from a composite scenario — not measured client results.

A union parishad chairman campaign, nine wards, around 11,000 households in the roll. Thirty-eight field workers, most of them volunteers, nearly all working from Android phones on mobile data of varying quality.

They came to us eight weeks before polling day with a specific complaint: the campaign manager no longer believed his own numbers.

Twelve spreadsheets and no single truth

The setup had grown organically. One master spreadsheet of the roll, shared to nine ward organisers, each of whom had made a copy for their own area. Three organisers had further split theirs by worker. Twelve files in circulation, all descended from one original, none of them reconcilable.

The specific failures:

The manager's estimate was that somewhere between a fifth and a third of canvass returns were either duplicated or lost. He could not narrow it further, which was itself the problem.

What we put in

A web panel for the office and an Android app for the field, over one database.

Record-level writes

Each household is a record, not a spreadsheet row. Two workers updating two households in the same street both succeed. Two workers updating the same household produce a flagged conflict for the ward organiser rather than a silent overwrite — which surfaced something useful, since repeated conflicts on one street usually meant two workers had been assigned overlapping areas.

Offline capture with queued sync

The app writes locally and syncs when it has signal. This was the change workers noticed most, because it removed the reason they had retreated to paper. Entry happens at the door, while the conversation is still in their head.

Role-scoped visibility

A worker sees their assigned streets. A ward organiser sees their ward. The manager sees everything. Nobody carries 11,000 households on a personal phone, and access is revoked in one action when someone leaves.

A five-point support scale

Strong support, lean support, undecided, lean against, strong against, plus not-contacted. The previous spreadsheet had a free-text "notes" column where workers had written everything from "yes" to "maybe if brother agrees", none of which could be counted.

The argument about the support scale

Two ward organisers pushed hard for a ten-point scale. More precision, they argued, would make the targeting better.

We pushed back, and this is worth recording because it comes up every time. Thirty-eight people cannot apply a ten-point scale consistently. One worker's 6 is another's 8, and once the data is pooled those distinctions are noise dressed as precision. Five points, applied the same way by everyone, produces a number you can act on.

They agreed to trial five points for a week. The argument did not come back.

What the data showed in week three

Once attribution existed, two things surfaced within a fortnight.

First, one ward's canvass rate was roughly four times the campaign average. Investigated, it turned out one worker was recording households as contacted after knocking and getting no answer. Not fraud — a genuine misunderstanding of what "contacted" meant. It had been invisible for weeks and had inflated that ward's numbers substantially.

Second, three streets in the largest ward had never been assigned to anyone. They had fallen between two organisers' copies of the spreadsheet months earlier. About 400 households that nobody had knocked on, surfaced by a map view that showed not-contacted records as a block.

Polling day

This is the day the whole system exists for.

  1. Before opening, the panel generated a turnout list per ward: every household marked strong or lean support.
  2. Observers at each polling centre recorded voters as they arrived, from the app.
  3. Every two hours the list refreshed, subtracting those who had voted.
  4. Workers were dispatched to the remaining addresses, nearest first.
  5. By late afternoon the outstanding list in each ward was small enough to work door by door.

The step that was impossible before was the second one. Observers at nine centres writing into the same dataset at the same time is exactly the concurrent-write problem a shared spreadsheet cannot solve. Under the old system the campaign had done this once, around 2pm, on paper, and the reconciled list arrived too late to act on.

In the scenario the candidate wins, but that detail is set dressing and we would not want it read as a result. Local margins turn on candidates, parties and events far larger than any software, and no tool should be credited with an electoral outcome. The transferable point is narrower: a turnout operation can run five times through the day instead of once, because concurrent updates from nine polling centres no longer overwrite each other.

What we would do differently

Train on the support scale first, not the app. The app took workers about twenty minutes to learn. Agreeing what "lean support" means took considerably longer and mattered considerably more. We should have run that session before handing out logins.

Define "contacted" explicitly and in writing. The inflated-ward problem was entirely a definition failure, and it was predictable.

Set the data disposal plan at the start. It was agreed in the last week, when everyone was busy. It should be decided when the system is set up, while there is time to think about it.

The spreadsheet was never the problem in itself. The problem was that nobody could tell which version was true — and a campaign runs on exactly that.
Who wrote this, and on what basis

We build Votix. This is not legal advice; election and data-protection rules vary by jurisdiction. No real voter data was used in preparing this piece.

Written by the Black & White Studio product team, Dhaka. Last reviewed 20 November 2026; rates, rules and platform behaviour change, so check anything time-sensitive against a current source. Spotted an error? Tell us — we correct things.

Votix — campaign panel and field app

Voter records, household surveys, worker assignments, analytics, budgets and election-day turnout operations in one system. Web panel for the office, Android app for the field. Demo and pricing by enquiry.

See Votix →