How this works
A scorecard that names institutions has to show its working. Every weight, threshold and limitation is set out below, so any operator can recompute their own score and challenge it.
Automated testing detects roughly a third of the barriers a real person meets. It reliably finds a button with no accessible name. It cannot tell you whether the reading order makes sense, whether an error message is understandable, or whether somebody can actually finish paying a bill.
A score here is a floor, not a certificate. A service scoring 100 has passed the checks a machine can run, nothing more. A service scoring 20 has problems so basic that no manual testing is needed to confirm them.
This is why every service examined closely is also tested by hand by a screen reader user, and why those results sit alongside the automated ones.
Every failing element deducts from a starting score of 100, weighted by how much it blocks a person. An unlabelled button is silent to a screen reader, so the user cannot know it exists. Redundant link text is friction. Weighting them equally would flatter a service with many small problems and punish one with a single blocking failure.
−10
Critical
Blocks the task outright
−5
Serious
Task possible but obstructed
−2
Moderate
Confusing or slow
−0.5
Minor
Friction
Adjust the counts to see how a score is built. Four critical failures is enough to take a service below half marks on their own.
Score
53
grade D
100 − (2×10 + 4×5 + 3×2 + 2×0.5) = 53
| Grade | Score | Means |
|---|---|---|
| A | 90 and above | Few barriers found |
| B | 75 to 89 | Usable, with friction |
| C | 60 to 74 | Significant barriers |
| D | 40 to 59 | Largely unusable |
| F | 0 to 39 | Blocked |
Each service is loaded in a headless Chromium browser at 1366 by 900 pixels. Once the page settles, axe-core runs against WCAG 2.0 and 2.1 at levels A and AA, plus WCAG 2.2 at AA.
Only the public landing page is scanned. Pages behind a login are not covered, which means the real picture for authenticated banking is likely worse than what is shown here, not better.
Every example below is a violation type found on Nigerian services in this scan, with the counts from the committed results. Switch between the version as found and the corrected one to see both the markup change and what a screen reader announces.
Almost all of them are a single attribute.
image-alt43 found in this scanWhen an image is the only thing conveying something, and it has no text alternative, a screen reader has nothing to say. A logo, a promotional banner, a phone number set as an image: all silent.
One attribute to fix
Markup
<img src="/customer-care.png">What a screen reader announces
image
WCAG 2.0 A · 1.1.1 Non-text Content
link-name71 found in this scanLinks made only of an icon, or wrapped around an image with no alternative text, are announced as nothing. People who navigate by pulling up a list of links on the page get a list of blanks.
One attribute to fix
Markup
<a href="/transfer">
<svg class="icon-arrow"></svg>
</a>What a screen reader announces
link
WCAG 2.0 A · 2.4.4 Link Purpose
button-name3 found in this scanThis is the failure people mean when they say an app is unusable. Moving through the page, every control announces as the single word "button". There is no way to know which one submits the form.
One attribute to fix
Markup
<button class="btn-primary">
<i class="fa fa-search"></i>
</button>What a screen reader announces
button. button. button.
WCAG 2.0 A · 4.1.2 Name, Role, Value
color-contrast74 found in this scanLight grey on white looks refined on a designer’s monitor. On a phone in Nigerian daylight, or to anyone with reduced vision, it disappears. This was the single most common failure in the scan.
One colour value to fix
Markup
.help-text {
color: #9ca3af; /* 2.5:1 on white */
background: #ffffff;
}What a screen reader announces
Text is announced, but cannot be read by sight
WCAG 2.0 AA · 1.4.3 Contrast Minimum
label0 found in this scanA placeholder disappears the moment somebody starts typing, and many screen readers never announce it at all. The person is left in a field with no idea what belongs there.
One element to fix
Markup
<input type="text" placeholder="Account number">What a screen reader announces
edit, blank
WCAG 2.0 A · 3.3.2 Labels or Instructions
target-size10 found in this scanA control under 24 pixels is hard to hit for anyone with a tremor, limited dexterity, or simply large hands on a small phone. Missing repeatedly is not a minor irritation when the control transfers money.
Two CSS values to fix
Markup
.icon-btn {
width: 16px;
height: 16px;
}What a screen reader announces
Announced correctly, but difficult to activate
WCAG 2.2 AA · 2.5.8 Target Size
For each service examined in depth, a screen reader user attempts a short set of ordinary tasks: find the customer service number, locate the branch finder, begin an account opening. Each is recorded as completed or not completed. There is no partial credit, because from the user’s side there is no partial credit either.
Environment and date are recorded with every finding so results can be reproduced or challenged.
The purpose is repair, not exposure. Operators are notified of findings before results are published and given a window to respond. Where a service improves, the score improves with it, because scans run continuously rather than once.
Nigerian institutions have already shown this is achievable. UBA published a formal commitment to WCAG 2.1 Level AA in June 2025. OPay supports independent account setup for blind users. Airtel scores 98 in this scan with no critical or serious failures. The gap is not technical difficulty.
The scanner, the target list, the scoring code and the raw output are all in the project repository. Anyone can clone it and rerun the scan. Findings that cannot be reproduced should not be trusted, and that applies to ours.
Engine: axe-core 4.10 via @axe-core/playwright
Last run: 10 August 2026