How do I tell whether I need a consultant or a tool?

Name the bottleneck first. If nobody knows what is wrong, that is a detection problem and the cheap half is automated — scan before you hire anyone. If you have a list and nobody knows whether an item is genuinely a failure or how to fix it in your framework, that is a judgement problem, which is what a consultant sells. If you know exactly what is wrong and what to do and there is simply more of it than your team can ship, that is a throughput problem, and the answer is contract development hours or a hire, not another audit. Most stalled accessibility programmes buy the wrong one of these three and conclude the category does not work.

What does an accessibility consultant actually cost?

Engagements are usually structured as a fixed-fee audit over a named sample, a day rate or hourly rate for advisory and remediation support, or a retainer for ongoing design and code review. Rates vary widely with the depth offered — a specialist doing screen-reader testing on a complex application is a different purchase from a generalist reviewing static pages — which is why comparing totals without normalising the sample is meaningless. The number that is almost never quoted and usually dominates: your own developers' time to implement the fixes. Ask any bidder to estimate remediation effort alongside the findings; a proposal that will not is a proposal that has not considered whether its own report can be acted on.

When does an in-house hire make sense?

When accessibility work is continuous rather than episodic — you ship frequently, you have several product teams, and design decisions with accessibility consequences are being made weekly without anyone qualified in the room. At that point a consultant is being asked to be present for decisions they cannot attend, and the retainer quietly becomes more expensive than a salary while delivering less. The other trigger is procurement load: if you are answering accessibility questionnaires and maintaining conformance reports for enterprise or public-sector customers on a rolling basis, that is a standing internal job. Below those thresholds, a specialist hire tends to be under-used and isolated, which is also how they leave.

Can a consultant help without touching our code?

Yes, and for many teams that is the higher-value engagement. Accessible component patterns reviewed once in the design system prevent the same defect across every screen that uses them; a design review catches contrast and focus-visibility decisions before they are implemented in forty places; and developer training turns a one-off report into capability that survives the contract. The rule of thumb is to buy consultancy at the leverage points — design system, component library, definition of done, code review checklists — rather than as a per-page remediation service, which is the most expensive way to consume expert time.

Is an overlay widget ever the right purchase?

Treat it as a distinct product from either option here, and evaluate it on its own evidence rather than as a shortcut to conformance. An injected script modifies the page after load; it does not repair the underlying markup, and disabled users' organisations have been consistently critical of overlays as a substitute for fixing the source. If a vendor bundles one with scanning, separate the two commercially and ask for scan results with the widget disabled — that reading is the state of your actual markup, which is what a remediation programme has to move and what an auditor will evaluate.

Hire an Accessibility Consultant or Buy Software? 2026 Decision Guide

Diagnose the Bottleneck Before You Price Anything

Accessibility programmes stall in three distinguishable places, and each has a different remedy. The first is detection: no inventory of what is broken. That half is mostly automatable and should never be the first thing you pay a person for — run a scan over your main templates and you have the machine-detectable layer for free.

The second is judgement. You have findings and nobody in the building can say whether this custom combobox actually fails 4.1.2, whether the alternative text on a chart is adequate, or what the right pattern is for an error summary in your framework. That is expertise, it is not in any tool, and it is what a consultant sells.

The third is throughput. The list is understood, the fixes are known, and the backlog does not move because the team is shipping other things. No audit fixes that, and a second audit definitely does not — the answer is contracted development capacity or a reprioritisation conversation, and it is worth naming as such rather than dressing it up as an accessibility purchase.

What Each Option Genuinely Delivers

A scanning platform gives you breadth, recurrence and a dated trail across the machine-checkable minority of WCAG. It is excellent at catching the same defects being reintroduced and incapable of telling you whether a journey can be completed with a screen reader. A consultant gives you the judgement criteria, a prioritised backlog, and — if you scope it that way — knowledge transfer into your team. An internal hire gives you presence at the moment decisions are made, which is the only intervention that stops defects being created in the first place.

Platform: breadth, recurrence, regression alerts, dated evidence. No judgement.
Consultant: depth on a sample, adjudication, patterns, training. Not continuous.
Hire: continuous presence in design and review. Expensive below a certain scale.
Contract developers: throughput against a known backlog. No detection, no judgement.
Overlay vendor: a different product entirely — evaluate separately, on its own evidence.

Buy Expertise at the Leverage Points

The most common way to waste a consulting budget is to spend it per page. Expert hours applied to a design system component are amortised across every screen that uses it; the same hours applied to one marketing page are spent once. If you are going to engage a specialist, point them at the places where a single decision repeats: the component library, the design review for new patterns, the definition of done, the pull-request checklist, and a training session for the developers who will be implementing.

The corollary for scoping: ask for deliverables that outlive the engagement. Accessible patterns documented in your own component documentation, a reusable checklist, recorded training, and a triage rubric your team can apply to the next scanner output without calling anyone. A report is a snapshot; a pattern library and a rubric are capability. Vendors who sell the second thing will say so when asked; those who do not will quote per page.

How to Evaluate a Consultant Without Being an Expert

You do not need to know WCAG to judge a bidder. Ask for a redacted sample finding from a previous engagement and look at its shape: does it name the success criterion, the element, the assistive-technology combination it was reproduced in, the observed and expected behaviour, and a suggested fix in the relevant framework? Ask which screen readers and browser pairings they test and how many hours of that are in the quote. Ask whether anyone on the team uses assistive technology daily. Ask them to name something a scanner would have missed on a site like yours — a good answer is specific and technical, and a bad one is a sales sentence about compliance.

Two answers should end the conversation. A guarantee of compliance or of immunity from complaints, which nobody can give about a site that changes weekly. And a proposal whose remediation strategy is to install a widget, which is a different product being sold under the name of the one you asked for.

The Hybrid Most Teams Arrive At

After a cycle or two, the arrangement that survives tends to look the same. Automated checks live in the build pipeline so component-layer regressions fail a pull request. A scheduled crawl covers what is published outside the pipeline — CMS pages, marketing builders, uploaded documents. A specialist is engaged periodically for the judgement criteria: a sampled audit of the journeys that matter, a review of new components before they proliferate, and a retest that closes the loop on the last backlog. And somebody internal owns the programme, even at a fraction of their time, because an unowned backlog is indistinguishable from no backlog.

The order matters more than the mix. Scan, so you know the cheap layer. Audit a named sample, so you know the expensive layer. Remediate and verify. Then, and only then, decide what recurring spend your publishing cadence actually justifies.

Questions Teams Ask Before Signing

We have a small marketing site. Do we need any of this?

Probably not a consultant and almost certainly not a platform. A handful of templates, a scan you read rather than file, a keyboard-only walk through your contact and conversion flows, and alternative text discipline on uploads will cover most of the realistic exposure. Where a small site does justify expert time is a single review of the template set — once, at build time or at redesign — because a defect baked into a header or a form component repeats on every page and costs nothing to avoid at the point the markup is written.

Our developers say they can handle it. Is that reasonable?

It is reasonable and usually incomplete. Developers can implement almost anything once they know what correct looks like; what they typically lack is exposure to how the interface behaves in a screen reader and which patterns are known traps. The cheap intervention is not to overrule them but to give them the missing input — training, one reviewed reference implementation of your hardest component, and access to real assistive technology on their own machines. Teams that get that input tend to stop needing per-page consulting, which is the outcome you want.

Can we use an agency that also built the site?

For remediation, yes, and they will be faster because they know the codebase. For the evaluation that goes into a conformance statement, get a second party, for the same reason you do not audit your own accounts. The practical split is an independent audit defining the backlog and the original agency doing the work, with the independent party performing the retest. Where budget allows only one, keep independence on the evaluation rather than on the remediation.

What should be in the contract if we hire a consultant?

The sample, stated as templates and journeys rather than a page count. The standard and version being applied. The testing combinations and how many hours of manual assistive-technology testing are included. The finding format, machine-readable, with stable identifiers. At least one retest inside the base fee. Developer access hours for questions during remediation. And ownership of the findings, plus whether you can be named as a client. Every one of those is a normal ask, and a bidder's reaction to them is itself information.

How do we keep this from regressing once the engagement ends?

Put accessibility into the two places work cannot bypass: the definition of done and the pull-request template. Automated rules in CI catch the mechanical layer; a short manual checklist on the review — keyboard reachable, focus visible, labels present, error announced — catches most of the rest at the point of creation. Then schedule one recurring thing you will actually do, whether that is a quarterly journey walk or a monthly scan someone reads. A modest habit that runs beats an ambitious programme that lapses after the consultant's invoice is paid.

Find Out Which Half You Are Missing First

The detection half is the cheap half and you can see it without signing anything. Read the machine-detectable layer on your own templates, then ask the only question that picks the right purchase: is the problem that we do not know, that we cannot judge, or that we cannot ship?

Each answer has a different price, and only one of them is a consulting engagement.