Page count is a poor estimate of accessibility audit effort on its own. A thousand articles may share a small number of layouts. A ten-page application may contain complex forms, dialogs, account roles, and several ways to complete the same task. The estimate needs to describe that work, not just the number of addresses in a spreadsheet.
Build the estimate from the proposed sample, test conditions, reporting requirements, and known uncertainties. Make the arithmetic visible so the client or product team can change an assumption without having to start the conversation again.
Estimate activities before assigning hours
Break the work into activities that produce a recognizable result. Discovery produces a reconciled inventory. Sample selection produces a coverage rationale. Evaluation produces observations and evidence. Reporting produces a prioritized handoff. Retesting produces updated results against an identified release.
For evaluation, list the distinct templates, components, journeys, and states that need attention. Include the agreed browsers and assistive technologies where they change the work. Do not multiply every page by every possible combination automatically; explain the coverage strategy and the reason for each combination.
Automated checks can support this process, but they do not remove the need for human evaluation. WAI's guidance on selecting evaluation tools explains that no tool alone can determine whether a site meets accessibility standards. Estimate the manual work explicitly rather than leaving it as an unpriced assumption.
Separate repeated structure from unique behavior
Group pages that genuinely share a template, then inspect the differences inside each group. A common layout can still contain a unique chart, form, document, or media player. Record those additions as work items rather than counting the entire page as either completely new or completely repetitive.
Avoid double counting shared components. If the same navigation appears across several templates, the estimate can include a core evaluation plus checks that it behaves consistently in those contexts. If the implementation differs, treat that difference as additional work.
The representative sampling guide describes how to make these selections defensible. An estimate based on a visible sample is easier to challenge constructively than a flat rate derived from the crawl total.
Show a worked model with explicit assumptions
The table below is a fictional planning example. These hours are chosen to demonstrate the calculation; they are not market benchmarks, a quotation, or a prediction for your website.
| Activity | Illustrative assumption | Hours |
|---|---|---|
| Discovery and scope agreement | Inventory review and one scope discussion | 4 |
| Template evaluation | Four distinct templates at two hours each | 8 |
| Journey evaluation | Two processes at three hours each | 6 |
| Additional states | Error, empty, and expanded states | 3 |
| Evidence and report preparation | Reproduction details and review | 5 |
| Handoff discussion | Findings walkthrough with owners | 1 |
| Agreed retest | A defined set of fixes on one release | 3 |
| Planned total | Sum of the listed activities | 30 |
The value of this model is its transparency. If a second account role introduces a different approval workflow, the team can add that activity. If the report needs a separate executive presentation, it can be estimated as a deliverable instead of appearing as unexpected work later.
Keep rates separate from hours when discussing cost. Currency, procurement conditions, expertise, and contractual terms vary. An effort model should not imply that a universal price follows from the same number of pages.
Give uncertainty a named cause
An unexplained contingency percentage is difficult to review. List the conditions that could expand the work and describe the effect of each one.
For example, an unresolved legacy section might add a template and another navigation implementation. An unavailable test account might delay the work without increasing evaluation time, or it might reveal an additional role once access is granted. Distinguish schedule risk from effort risk.
Prepare a base scenario using confirmed information and an expanded scenario using specific unresolved items. In the fictional model above, discovering one additional template and a two-hour journey review would add four hours under the stated assumptions. That is a conditional calculation, not a statistical confidence interval.
Assign an owner and a decision date to each unknown. Reducing uncertainty before testing is often more useful than negotiating a smaller unexplained buffer.
Include the work after findings are discovered
Testing is only part of the engagement. Reviewers need time to verify observations, distinguish repeated defects from separate issues, attach useful evidence, and explain the impact. Developers need enough context to reproduce the problem without arranging another meeting for every ticket.
Define whether the handoff includes issue tracker entries, a report, a findings meeting, or a retest register. The evidence handoff guide provides a practical structure. State whether retesting covers all delivered findings, a specified subset, or a separately agreed change set.
Also define what happens when remediation introduces new behavior. Checking a repaired label may be straightforward. Reviewing a replacement checkout flow can require new evaluation rather than a narrow confirmation of the old finding.
Revisit the estimate when the boundary changes
Keep a dated record of scope changes and their effect on effort. Examples include a newly included language, a different target release, additional document formats, or a changed reporting requirement. The record should explain the decision, not merely show a revised total.
Start with a written audit scope, then use the public scoping report to see the kind of inventory and evidence that can inform early planning. The report makes discovery tangible. The estimate still depends on the people, interactions, and deliverables that the audit must cover.