Use an automated audit when the question is what is wrong with this page. Hire an agency when the question is why are people leaving, what should we build instead, or can we prove this is accessible. The first is readable from the page itself. The other three are not on the page at all, at any price.
The money follows the same split. An agency audit runs $3,500–$25,000 and 40–80 expert hours over one to three weeks; automated tooling runs $0–$500 a month. Those are the same order of magnitude as the four other 2026 price guides cited on what UX audits cost. Below, 9 situations with the answer stated — 5 of them say hire the agency.
Most comparisons argue about quality. The real difference is what is in front of each of them when they look: an automated audit has the served document, and nothing else. An agency has whatever you give them access to, plus the ability to ask a person a question.
| The question | Automated | Agency | Answerable by |
|---|---|---|---|
| Is the primary action ambiguous on this page? | Yes — it is in the served document and countable. | Yes. | Either |
| Does the page contradict itself between title, heading and offer? | Yes — a text comparison over the same document. | Yes. | Either |
| Where in the funnel do people actually leave? | No. That is in your analytics, and an audit of a captured page is not connected to them. | Yes, when the engagement includes analytics access — drop-off per step is standard scope. [Groto] | Agency |
| Why did they leave? | No. A page can be observed; a motive cannot be read off one. Anything claiming otherwise is inferring. | Yes, with moderated sessions or interviews — that is what the research hours buy. [wearepresta] | Agency |
| Does the flow break for a screen-reader user? | Partly. An automated rule set flags what code can show — missing labels, names, contrast — but whether the flow works with a screen reader needs a person. No score here is a compliance position. | Yes, when an accessibility specialist is on the engagement. | Agency |
| Is the buying logic right for this category? | No. Category-specific judgement is not in the document. | Yes — this is the part you are paying a person for. [wearepresta] | Agency |
| Did last Thursday's deploy quietly break the signup page? | Yes, if you rerun it. Cheap enough to run on every release. | In principle, but nobody buys a $5,000 engagement to re-check one page. | Automated |
| How do forty landing pages compare to each other? | Yes — the same rules applied forty times, consistently. | Impractical. A reviewer cannot hold forty pages to an identical standard by hand. [wearepresta] | Automated |
| Which two of these problems should we fix first? | A ranking by a published formula, from severity, evidence and effort. | A judgement informed by your business, which is usually better and always costlier. | Either |
Find the row that matches. Take the answer, including when it is not the product this site sells.
You do not know whether there is a problem worth paying to fix.
Spending four figures to find out whether there is anything to find is the wrong order. Run the cheap thing first.
Conversion dropped and you do not know where.
The answer is in your funnel data, and finding it is analytics work before it is design work.
You ship weekly and want regressions caught.
The value is in repetition. A one-off engagement cannot watch every release.
A redesign is being argued over by a room of stakeholders.
You are buying facilitation and authority as much as findings. A score settles nothing in that room.
You must demonstrate accessibility compliance.
Compliance needs a person testing against the standard. No automated score is a legal position.
You have one page, a small budget and a specific worry.
The cheapest way to convert a worry into something written down and checkable.
The thing you need audited is behind a login.
An automated capture opens a public URL. A dashboard, an app or anything past a sign-in is not reachable that way at all — this is a hard boundary, not a quality difference.
You need the findings turned into designs someone can build.
A report ends at acceptance criteria. Somebody still has to draw the replacement, and that is design work you are buying by the hour.
Your product is live and complicated and the stakes are high.
Scan first to triage, then spend the expert hours where the scan and your data agree something matters.
The one dedicated manual-versus-automated guide published in 2026 lands on sequencing rather than picking a winner: “The winning play is both, sequenced correctly.” — scan first to triage, then spend the expert hours where the scan and your own data agree something matters. That is the right rule, and it is worth reading it from wearepresta rather than from a vendor.
One caution about the same guide, and about this genre generally: it also projects a conversion lift and a six-figure revenue recovery from applying its method. Those numbers are a model, not a measurement. Treat the sequencing advice as good and the revenue projection as an illustration.
For completeness, since this page is published by one of the two things it is comparing.
It is the automated column, and only that column
One page, opened in a real browser and read by published rules. No analytics, no users, no interviews.
Its mobile and speed numbers are one lab run, not your visitors
The page is re-laid out at 375px and checked for the viewport tag, sideways scrolling, tap-target size and text size; load timings come from one browser run in one location. That is a check, not a device lab and not field data — and if the browser pass fails, all three metrics read "not measured" rather than zero.
It will not estimate your lost revenue
That needs your traffic, conversion rate, margin and a causal link from a finding to a lost sale. Tools that print the number anyway are guessing in a confident font.
Its scoring is published
The severity weights, the nine ceilings, the hard 18–96 clamp and what each browser-measured metric deducts for are on the example report, with the arithmetic.
If the sequencing rule is right, the automated pass comes first and costs nothing to try. 3 audits per 24 hours without an account or email. A free account includes 3 audits per month. The scoring is published so you can judge the output rather than trust it.
Done for you instead: the Pro UX Audit is €249, delivered in 3 business days. It is still the automated column — it does not become an agency engagement by costing more.