Visionably Research · Report · 7 September 2026

Digital Banking Accessibility Report

We measured the public websites of 30 Indian scheduled commercial banks with one instrument, against WCAG 2.2, and asked of each whether a screen-reader user can reach three routine pages: find a branch, file a complaint, reach net-banking.

Download the report (PDF/UA)

31 pages. Version 1.1, public edition. Only the five best-scoring banks are named; every other bank is identified by rank code and type.

Headline numbers

23
banks measured, across 182 pages
0
banks with all three task pages usable; 20 of 23 with none
58.9 / 100
median confirmed site score
2,221
confirmed failures, 233 of them critical
35%
of confirmed failures found only by live testing

What we found

Only confirmed failures are counted. Anything that needs a person's judgement is listed in each bank's workbook and never scored.

The task result is the finding. Under a deliberately low bar, where a single confirmed keyboard, state or naming failure disqualifies a page, no measured bank clears the three pages a customer needs most often. The complaint page, which a customer reaches when something has already gone wrong, was usable with a screen reader at one bank.

The defects are shared, so the fixes are. Touch targets under 24 px and controls without a name, role or value fail at every measured bank; a skip link is missing at 22. These are properties of a design system, not of a page. A bank that fixes each once, in its component library, removes most of its confirmed failures everywhere. The keyboard traps, observed at 13 banks, are almost all carousels.

Live testing changes the picture. 35% of confirmed failures came from pressing keys, opening controls and re-measuring the page rather than from reading its code. A programme that relies on a static scanner will under-report the failures that most directly block keyboard and screen-reader users.

Ownership is not the variable. Public-sector and private banks are distributed throughout the ranking, and publishing an accessibility statement did not distinguish banks in this sample. The differences that matter are in the front-end platform and the review discipline behind it.

The five best-scoring banks

Not defect-free: they carry a mean of 67.8 confirmed failures across eight pages, against 104.6 for the other 18. What sets them apart is the absence of the shared, systemic defects.

Ranking of the five named banks by confirmed site score
# Bank Type Score Grade
1 Canara Bank Public sector 77.3 C
2 RBL Bank Private 74.6 C
3 Karur Vysya Bank Private 72.9 C
4 ICICI Bank Private 71.5 C
5 State Bank of India Public sector 69.3 D

Grades: A at 90 and above, B at 80, C at 70, D at 50, F below 50. No measured bank reached A or B, which is in line with what automated surveys of large consumer sites generally find. Every other bank appears in the report as a rank number and a type, with nothing that identifies it.

How it was measured

The full method, page lists, scoring definitions and tool versions are in the report, so that the measurement can be reproduced.

  1. Sample

    30 scheduled commercial banks, fixed before any measurement: the twelve public-sector banks, the largest private banks by assets, and one small finance bank. Eight pages per bank, seeded for the three tasks by a fixed procedure.

  2. Instrument

    One engine, run identically on every site: WCAG 2.2 A and AA rule evaluation, live keyboard traversal, interaction states opened with a virtual screen reader listening, reflow and zoom re-measurement, and a transcript reading reported for review and never counted.

  3. Scoring

    Only confirmed failures are scored. A task page is usable when it was reached and carries no confirmed keyboard, state or naming failure, a necessary condition for a screen-reader user, not proof that the task can be completed.

  4. Naming and disclosure

    Every bank received its own findings before publication and may reply; corrections are made on reply. Six banks whose sites could not be measured are listed with the reason and draw no conclusion. The same measurement is repeated in ninety days.

The confirmed score is a floor on the failure rate, not a ceiling. Criteria that need human judgement, media alternatives and cognitive accessibility were reported for review or not evaluated. This report is an evaluation, not a certification, and not legal advice.

Work at one of the banks?

Your complete workbook is free on request: every finding with its evidence and a suggested fix, the screen-reader transcript, the interaction states as heard, and a conformance table that states the method on every row.

Request your workbook

Press and researchers

The audit records, screen-reader recordings of the pages as a blind customer hears them, and the data behind every table are available on request, as is a demonstration on any bank site.

Write to us