Comparison
Percelia vs. an accessibility agency
5 min read
In short
Both routes end with a human taking responsibility for the result. The difference sits before that. A manual audit spends person-hours on every additional page, so the sample stays narrow. Percelia assesses every page in scope across six analysis layers with the same yardstick: the scan in minutes to hours, the reviewed report within one to three working days, repeatable at the push of a button, and routes every ambiguous case to human review.
This is not „human versus machine“. Percelia has a human sign off too, and automation never settles an ambiguous criterion on its own. So the interesting question is not whether a human is involved, but how much was assessed before that human decides.
Agency and Percelia at a glance
| Agency / consultancy | Percelia | |
|---|---|---|
| Who settles ambiguous cases | A human | A human, the same responsibility |
| What an extra page costs | Person-hours, so a narrow sample | Compute time, so a wider scope |
| Checks applied per page | Weighted by experience and available time | Identical for every page: six analysis layers, in full |
| Consistency across pages and runs | Depends on person, order and time budget | Same assessment system, same thresholds, versioned rules |
| Time to result | Weeks | Scan in minutes to hours, reviewed report in 1 to 3 working days |
| The second run | Usually a new project | One click, plus continuous monitoring and a CI gate |
| Evidence per finding | Prose report, depth varies by provider | Element, screenshot marker, criterion, confidence, machine-readable |
| After the finding | A recommendation, implementation depending on scope | A concrete code fix plus a re-test on our copy of the page |
| Deliverable | An audit report | An audit report plus VPAT / EN 301 549 for sign-off |
Thorough means more checkpoints, applied evenly
Thoroughness is usually read as depth per page. It has a second axis that matters just as much in practice: how many pages were assessed at all, and whether the same yardstick applied to all of them. A manual audit has to compromise here, because every additional page costs real person-hours. For Percelia the limit is compute time. That shifts the economics: the scope can be wider, and every page inside it gets the same complete set of checks rather than whatever attention was left that day.
Reproducibility comes on top. Assessing the same page twice yields the same result here, because the same rules, thresholds and versions run. That is not a judgment about auditors, it is a property of the method: what runs deterministically does not drift. Our pilot study shows what that looks like in practice, applying one identical set of checks across many sites.
Six layers that run in full on every pass
Percelia does not assess with one rule engine but with six independent layers, five of them deterministic:
- Rules: axe-core plus our own contrast computation on the rendered DOM and CSS.
- Keyboard and interaction: real Tab, Enter and Escape operation driven through the browser, focus order, visible focus, keyboard traps, completion of key flows.
- AI visual review: judgment over screenshot and DOM where rules cannot decide, such as whether alt text is meaningful or text over an image stays legible, every finding carrying a confidence score.
- Adaptation and reflow: the page is shrunk to 320 CSS pixels, resized to 200 percent, overridden with the WCAG text-spacing values and checked in both orientations.
- Accessibility tree: reads the tree the browser itself resolved, meaning exactly the roles, names and states assistive technology receives, not the source markup.
- Dynamic content: safely triggers form validation and observes whether status and error messages are announced at all.
Layers four to six are where scope and economics show most clearly. Zoom levels, text-spacing overrides, the resolved accessibility tree and triggered error messages are laborious to check by hand on each page, which is why audits under time pressure often sample them. Here they run on every assessed page, every time.
Where the human decides, and why that stays
Every finding carries a confidence score. What is unambiguous is accepted automatically. What is ambiguous goes to human review, as does every criterion that cannot be settled by machine in principle. No automated tool can decide all WCAG A/AA criteria, and we do not claim otherwise. That is exactly why the human here is not a concession made for the comparison but the part of the method the overlay vendors left out.
After the finding: a fix and a re-test
An audit report says what is wrong. Percelia additionally produces a concrete before/after fix for that exact element and then verifies it: the fix is applied to our copy of the page, the detector that raised the finding runs again, and the outcome is measured rather than assumed. Nothing changes on the client's own site in the process, which would be the overlay we deliberately refuse to build.
Conformance is a state, not a date
An audit describes a moment. The next release can break it, often unnoticed. That is why the second run here is not a new project but a click: continuous monitoring reports regressions, and a CI gate can stop one before it ships. What a full audit costs is in the guide on audit costs.
For agencies: we are not competing for the same hour
Agencies offering accessibility work with us rather than against us: Percelia takes on the broad, repeatable part of the assessment and delivers the findings with their evidence, while the agency brings consulting, implementation and the client relationship. An agency that wants to scale assessments without hiring for it can put Percelia underneath.
Frequently asked questions
- Is an AI-assisted audit as thorough as a manual expert audit?
- Thoroughness has two axes: depth per page, and scope times consistency. On scope and consistency Percelia has a structural advantage, because an extra page costs compute rather than person-hours and every page gets the same complete set of checks. On depth in the individual case, human judgment stays part of the method, which is exactly where every ambiguous finding is routed.
- Does a human really review at Percelia?
- Yes. Every finding carries a confidence score, and everything ambiguous, plus every criterion that cannot be settled by machine, goes to human review. A human signs the result off. Without that step there would be no defensible conformance statement.
- Why is Percelia faster?
- Because the six analysis layers run automatically and in parallel, which concentrates human review on the cases that actually need it. Weeks of lead time become days without dropping assessment steps.
- Does Percelia work with agencies?
- Yes. Agencies use Percelia as assessment infrastructure and add consulting and implementation. For the agency it does not replace the client relationship, it replaces the manual assessment effort that scales badly.
Request a free scan
Tell us your domain - we run a free initial scan and show you the most important barriers.