Leading UX Audit Services

A UX audit agency is hired to establish where an existing product loses people, work out which of those losses actually costs money, and hand back an order of work a team can start on Monday. Finding problems is the cheap half. Deciding which nine of eighty deserve a quarter of your calendar is the half you are paying for.

In plain terms, somebody sits with your product the way a stranger would. They walk the paths where your revenue happens, hold the interface against usability principles the field settled decades ago, read the behavioural data you already own to see how many people meet each fault, and then throw most of it away. What survives gets a severity, an effort estimate, a value estimate and a place in a sequence.

The daily work looks like this.

• Reading funnel and session data to locate the steps where people stop
• Attempting your critical tasks end to end, on a phone and on a desktop, with every hesitation recorded
• Holding each screen against usability principles and against the accessibility standard
• Reading every label, error and empty screen for whether it says what it means
• Comparing your patterns against what your category has already taught buyers to expect
• Writing it up so a person who attended nothing can act on it

Why this work comes to us. Brand Vision started in 2018. More than 500 projects across roughly 150 brands have gone through it since, so most of what we find is a pattern we have met before and can price quickly. On the independent side, Clutch shows a 5.0 built out of 64 or more interviews with clients, conducted by people who do not work here, and the design shelf holds four Awwwards, a Webby, and RGD and CGD certification for the practitioners. Past programs here have raised client lead volume by 250% on average. Treat that as history and never as a forecast, since it covers programs starting from wildly unlike positions and none of them was yours. The people who write the report are the same ones who would implement it, which is why findings here read as specifications and not suggestions. Reading a funnel, running a walkthrough and judging accessibility call for three unrelated skills, so the roles we recruit for are mixed on purpose, and the Toronto studio keeps those seats beside one another.

Paying for the diagnosis puts you under no obligation to buy the repair from us.

Nine reviews, and each one is only worth what it is worth because it gets read against the other eight. A confusing label on a page nobody visits and the same label inside your checkout are one finding and two completely different decisions.

1. Heuristic evaluation. The interface held screen by screen against established usability principles, meaning the rules the field agreed on long ago about telling people what the system is doing, preventing mistakes, staying consistent and leaving somebody in control. Covered in full below, including what it cannot see.
2. Analytics and funnel review. Your own behavioural data read for the steps where people stall, reverse, loop or leave, which paths complete and which do not, and how that differs by device and by where the visit came from.
3. Task-based walkthroughs. The handful of paths your revenue depends on, attempted the way a first-time visitor would, with every dead end and ambiguity recorded in the order it arrives.
4. Form and checkout analysis. Field by field. What is mandatory and whether it needs to be, what the validation does when somebody gets it wrong, what happens on a declined payment, and how much of the abandonment belongs to the form instead of to the offer. On hosted platforms some of this is fixed by the vendor, so Shopify builds get a note on what is genuinely changeable.
5. Accessibility review against WCAG. Keyboard operation, focus order, contrast, labelling and what a screen reader announces, with its own answer further down.
6. Mobile and cross-device review. Where the experience quietly degrades on the screen most of your traffic is using.
7. Content and microcopy clarity review. Microcopy means the small text doing the work, so buttons, labels, error messages, help text and empty screens, judged on whether it uses your customer's words and says what happens next.
8. Competitive and convention review. What your category has trained people to expect, and where you have broken one of those expectations at a cost instead of a gain.
9. The severity-ranked findings report and prioritized roadmap. Everything above reduced to a set of decisions, which is the only section that commits to anything.

Two boundaries. This is expert review of a product that already exists, so where the real question is what people want or why they behave as they do, that is UX research and a different purchase. And where the finding is that a flow needs rebuilding instead of repairing, that becomes interaction design work with a scope of its own.

Three of the nine is a real and cheaper engagement. On plenty of products the honest answer is the funnel, the checkout and the phone.

A heuristic evaluation is an expert walking your interface against a fixed set of usability principles and recording every place it breaks one. It is the quickest diagnostic in this discipline, it costs the least, and it is the one most often oversold.

The principles are decades old and they have held up well. Whether the system tells you what it is doing. Whether it speaks your language and not the company's. Whether a mistake can be undone cheaply. Whether the same action looks the same everywhere it appears. Whether people can choose from what is visible instead of remembering something they were told two screens ago. Whether the design stops the error before it happens at all. An evaluator holds every screen against that list and writes down each violation with a severity beside it.

What the method genuinely buys you is coverage and speed. One reviewer gets through a whole product in days, catches the large majority of obvious faults, and produces something a team can begin fixing the same week. Two or three reviewers working separately catch considerably more than one alone, because evaluators miss different things, and the overlap between them is a rough signal of how serious a given problem is.

Now the limits, since this is where the technique gets misrepresented.

• It cannot tell you what your users want. The review judges an interface against rules. It has no access to intent, motivation, or what somebody arrived to do.
• It produces false alarms. Reviewers flag things that never trouble a real person, and from inside the review there is no way to know which flags those are.
• It misses problems specific to your world. Where the difficulty is that customers do not understand a term your industry uses hourly, an outsider reads straight past it.
• It weighs nothing. A broken principle carries no information about how many people hit it or what it costs when they do, which is exactly why the funnel data has to sit alongside it.
• It is one person's judgment. A careful judgment, produced by a method, and still a judgment. A technique that outputs an opinion should be described as outputting an opinion.

Which is why it never runs on its own here. It runs against your behavioural data, against the walkthroughs, and against a read of what your category has already standardized, because a pattern that violates a principle in the abstract can still be the pattern every buyer of software has spent five years learning. Breaking a convention is occasionally differentiation and more usually a tax, and telling those two apart is closer to category research than to any checklist.

Four audits get sold into the same moment of worry, they answer four different questions, and picking the wrong one is how most diagnostic budgets get wasted. Each of them opens your analytics, each of them will flag a slow page, and a customer experiences none of these divisions, which is precisely why buyers confuse them.

The search audit asks whether people can find you and pick you before the click. Indexing, coverage, site architecture, authority, demand, competitors, and what assistants say about your category when somebody asks. Buy it when search visibility is dropping, or never appeared at all, or trails whoever you compete with. That is an SEO audit, and it stops the moment a person lands.

The website audit is wide and shallow by design. It reads the whole property as a commercial asset. Whether the page delivers what the ad promised, how it loads on an average phone, whether navigation makes sense, whether the reassurance a wavering buyer looks for is on screen at the moment they waver, and whether the numbers can be trusted at all. Buy it when you cannot yet point at where the loss occurs, and it sits inside a wide read of the whole property.

The usability audit is narrow and deep. It follows the small number of paths your revenue depends on, one step at a time, from first screen to confirmation, and puts a severity on every place a reasonable person would give up. Buy it when people are clearly arriving and almost nobody finishes anything.

The marketing audit sits above all three. Wider again, it asks whether the plan holds together across audience, message, offer, channel mix and measurement. Its typical case is a company where each channel looks acceptable on its own dashboard and the business as a whole has not grown in a year, and that is a marketing audit.

A rough self-check, usually enough on a first call. Cannot be found, start with search. Found, and people leave without acting, start here. Found, converting, and the business not growing, then the problem is strategic and sits above all four.

The honest complication is overlap. A real share of what turns up in a usability review is a performance or a search fault wearing a design costume, and we hand those over instead of claiming them. We have also finished week two of this work and told a client the screens were fine and the audience arriving was wrong, which moved part of their budget to a team we do not sit in. So the brief gets written around the question you walked in with, and not around whichever service name it happens to resemble.

This is a fixed-price piece of work with an end date on it, never a monthly retainer, and most take two to three weeks between access being granted and the session where it gets handed over. Four weeks is realistic where several user roles or markets are involved. Nothing about a diagnosis needs to be ongoing, so buying one should not commit you to what it recommends.

Days one to three. Access and starting numbers. Analytics, any session recording already running, the content system, support tickets, and half an hour with whoever hears the complaints. Opening measurements get written down before any judgment is made, so the same ones can be repeated in ninety days.

Week one. Heuristic pass and funnel read. The interface against the principles, and your data for where people stop. These run together deliberately, because each says where to look harder in the other.

Weeks one to two. Walkthroughs, forms, devices and access. The revenue paths attempted end to end, the forms field by field, the keyboard and screen reader pass, then the same tasks on a real phone.

Week two. Prioritization. Each surviving finding is weighed against how many people meet it, what recovering them is worth, what the fix costs, and whether you can ship it this quarter. A long list turns into a short order of work here.

Week three. Delivery. The document, preceded by a live session with whoever holds sign-off. Findings read out loud to the people holding the budget survive the week after delivery.

What moves the number.

• How many journeys are in scope, since each is a separate walkthrough across two classes of device
• How many distinct roles the product serves. An administrator, a manager and an end user are three audits behind one login
• Whether the analytics can be trusted, because getting it reliable enough to judge anything has to be scoped and quoted on its own
• Whether accessibility is a full conformance review or a risk read, which are different amounts of work
• Markets and languages, since each adds screens and a separate set of conventions
• Whether a rebuild is already on the table, in which case the output leans toward a specification for the new build

Two commercial points. Where this becomes a larger piece of work with us, we take the fee off the first phase of it, and that credit does not lapse, since a credit with a clock on it is a pressure tactic wearing a discount. And the audit produces no results. Execution does. Send us the product and you will hear back with a scope and an opinion on whether buying it makes sense.

An audit is expert judgment applied to a product that already exists. A study puts real people in front of it and reports what they actually did. Both produce ranked findings, and only one of them can tell you why somebody behaved the way they did.

What the audit is. People who have examined hundreds of products walk yours, measure it against principles and conventions, read your behavioural data, and commit to a judgment. It is quick, it asks nothing of your customers, it can start the day access arrives, and it will find most of the visible damage. Its ceiling is that all of it is inference. An experienced evaluator predicting where people will struggle is right most of the time and confidently wrong some of the time, and the report cannot tell you which findings fall into which group.

What the study is. Participants screened against your real customer profile, given genuine jobs to do, watched by somebody whose main discipline is saying nothing. It answers why. It settles arguments an audit can only take a side in. It costs more and it takes longer, because finding the right participants sets the pace, and it needs incentives and scheduling. That work lives in structured user research.

The sequence that wastes the least money is audit first, study second. The audit clears out every problem nobody needed to be told about, which is most of them, and it converts a vague brief into three specific unknowns. A study aimed at three named questions costs a fraction of a general discovery exercise, and the participants spend their hour on what is genuinely still in dispute.

Two situations flip that order. Where the product is still being built and there is nothing yet to review, observation is the only instrument available. And where the argument inside your company is about who the customer is instead of how a screen behaves, an audit will not settle it and should not be asked to.

The last thing worth saying is what happens when the two disagree, because they sometimes will. A finding we were sure of gets contradicted by five people attempting the task, and the five people win. Observed behaviour outranks expert opinion every time, which is the honest reason we would rather sell you a small study after the audit than a larger audit on its own, and what we have shipped has turned on that order more than once.

Every finding carries a figure for what fixing it should be worth, and that figure is the difference between a report that gets executed and one that gets filed. Nobody signs off on a line item with no number attached, so those items queue behind a budget meeting that never gets scheduled.

Here is the arithmetic, printed in the document where you can attack it. Take the step where the loss happens and the volume moving through it. Take the share of people the finding affects, which comes from the funnel and not from a feeling. Take a realistic recovery, meaning a portion of the loss and never the whole of it. Multiply by what one completed action is worth to your business. All four are assumptions, so all four sit beside the result. A range with its assumptions on show beats a single confident figure with nothing behind it, and being argued with is the point of showing them.

Three things keep the arithmetic honest.

1. The value of a completed action comes from you and never from a benchmark. A recovered cart in an online store and a recovered enquiry on a development project are separated by three orders of magnitude, so an identical fix lands in completely different places on two roadmaps. In property and development the enquiry also has to be discounted by a close rate before any of it means anything.
2. The recovery estimate is deliberately pessimistic. Removing a friction point wins back some of the people it was costing you and never all of them, because a share of that drop-off was people who were never going to buy from anybody.
3. Effort is written beside the severity, in whose week it consumes. A wording change somebody ships this afternoon, a component change waiting on the next release, and a flow rebuild needing a quarter of engineering time are three separate currencies, and only the last of them has to win an argument against your product roadmap.

Two habits matter more than the model does. Findings get delivered live to the people who can approve them, so the uncomfortable one gets argued about while everybody is still present and not quietly dropped once they have all left. And ninety days after the fixes ship we ask to see the same measurements again, whoever did the building, because a value estimate nobody ever checks is marketing.

The ceiling on all of it. These are estimates made two weeks into knowing your business and some will be wrong. When one is, we correct it in the open and say which assumption broke, which is not the same thing as defending the original number.

Four situations where the honest advice is to keep your money and spend it on the repair. Naming them costs us work, and a version of this page that named none of them would not be worth reading.

1. You already know what is broken and can prove it. If somebody in the building can name the step, show the data behind it and describe the fix, you are buying a receipt for knowledge you already hold. Spend it on the change. Say on the first call what you can already prove, and you will hear back whether that alone is enough to act on. This is the most frequent reason we talk somebody out of this purchase.
2. A rebuild is funded and starts inside two quarters. A product with a deletion date on it is not worth diagnosing, because the document stops being true the day the replacement launches. The useful purchase is a review of the flows and templates arriving, held while they are still open for argument, which belongs inside a replatform or redesign and what comes out of it is a set of specifications and not a list of faults.
3. A capable audit from the past year is sitting there unexecuted. Another document will not create the capacity that was missing, because analysis was never the bottleneck. Pay for execution instead, or pay somebody for two hours to turn the report you own into a sequence with owners and value estimates against it. A scoped advisory conversation is the cheaper route to that.
4. The traffic is the problem. Where the people arriving were never going to buy, the highest-converting interface in your category will not rescue it. That is a targeting question and no amount of interface work touches it, so a search diagnosis or a media review is the correct spend.

One product gets mistaken for this one. An automated usability score, generated by software pointed at an address and sold cheaply, is a real thing and a different thing. It will catch contrast failures and missing labels, which is genuinely useful, and it holds no opinion whatsoever about whether your checkout makes sense.

There is a smaller version of this for nearly every situation. One journey instead of three. A heuristic and funnel read alone, inside a week. Or a single working session on the one decision actually in front of you. And the conclusion is allowed to be that your interface is not the ceiling, in which case we write that down and hand it over. Describe the situation and somebody here will tell you which of those fits.

Analytics locates the problem to a step and a segment. It has nothing to say about what the person was thinking when they left, and a fix depends on that part. Which is why a Brand Vision audit runs the funnel review and the walkthroughs beside each other instead of one after the other.

Where the numbers are strong. Scale, so you learn that four thousand people met the same obstacle and not that one customer emailed about it. Precision of location, down to the step and often to the field. Device and browser splits, which is where a surprising share of the answer usually hides. Source splits, so you can see that the failure belongs to one paid campaign and not to the page it lands on. Comparison across a change. And an opening measurement worth retesting against later.

Where they run out. Explaining the loss. Choosing between four plausible causes. Telling you what somebody expected instead. Saying what they did next, including buying the same thing elsewhere. Or saying anything at all about the people who never arrived, which is a question for where the traffic comes from and not for the interface.

The pattern repeats on almost every engagement. A funnel shows most of the traffic leaving at one step. Five explanations get proposed internally, the loudest one gets built, and the step performs exactly as it did before. Causes at that level of detail are close to unguessable from a dashboard. A mandatory field asking for a reference number that lives in an email nobody kept. A total that changes at the final step for a reason no earlier page mentioned.

So the two halves run together. The data narrows eighty screens to four. The walkthroughs go to those four and read them the way a stranger would, in sequence, with only the information a stranger would have. That is what turns a location into an explanation.

Two cautions about the data itself. Consent choices and browser restrictions mean the numbers describe a portion of your visitors, so reading them as a full census makes everybody more certain than the evidence allows. And recordings show what a hand did and nothing about why it did it, which makes forty of them an excellent source of questions and a poor source of answers. Where the measurement itself turns out to be wrong, and on most products we inherit some of it is, getting enough of it corrected to judge anything is scoped openly and said out loud.

The small number of paths your revenue depends on, and nothing else, because auditing everything produces a document with no order inside it. Which paths those are gets agreed in the first conversation, and it is usually not the list a team walks in expecting.

How the journeys get chosen. From the money first, meaning whatever ends in a purchase, an enquiry, a booking, a signup or a renewal. Then from volume, so the busiest entry points are included even where nobody internally thinks of them as a journey. Then from the support queue, since a question your team answers forty times a week is a design finding you have been paying salary to absorb.

How a walkthrough runs. The task is written as a goal and never as a route, so work out whether this covers my situation and not open the third tab. Then it gets attempted with only what a first-time visitor would know, in the order a first-time visitor meets it. Every hesitation, backtrack, dead end and ambiguity is logged in sequence with the screen attached. The output is the journey itself with friction marked at the point it occurs, which is more useful than a pile of screens with problems listed under them.

Forms and checkout get a separate pass, field by field, because the most recoverable money in most products is sitting there. What is mandatory and whether it has any business being mandatory. What a field asks for against what the customer actually has in front of them. What the validation does on a mistake, and whether the message appears next to the field or at the top of a page already scrolled past. Whether an error preserves what was typed, since a form that empties itself is a cart thrown away. What a declined payment looks like. And whether the total, the delivery promise and the return terms are visible before the step where somebody commits, because a surprise at the end is the most expensive kind there is. On stores, the category and product pages feeding that checkout get judged too, and where the fix sits upstream it becomes ecommerce SEO or a merchandising decision.

The wording along the route is read as part of the same pass. Labels, buttons, errors, help text and empty screens, against whether they use your customer's vocabulary and say what happens next. Where the answer is that whole pages are answering the wrong question, that is content work and not a microcopy repair, and an implementation estimate is attached wherever the fix turns out to be structural.

Every journey gets attempted again on a real phone and again with a keyboard alone, and the two belong together for one reason. Each is a version of your product that the people who own it almost never look at. The desktop experience gets inspected by your team daily. The phone and the keyboard-only path get inspected by your customers.

The mobile and cross-device pass. The same journeys, attempted again on a real handset and not in a narrowed browser window. What surfaces is rarely a layout fault. Tap targets sitting too close together for a thumb. A fixed header eating a third of a small screen. A field summoning the wrong keyboard. A date picker nobody can operate one-handed. A sticky element covering the button it was meant to promote. A payment step that loses everything when somebody leaves the app to fetch a code. Load behaviour on a mid-range device over a poor connection belongs here too, since the fastest usability fix available is sometimes a performance fix. Where one account is used across a phone and a laptop, the handover between them gets walked as well.

The accessibility review against WCAG. WCAG is the standard the regulations point at, and the practical version of the question is whether somebody can operate your product without a mouse, without sight, or without hearing. Every control has to be reachable and usable from a keyboard, in a sensible order, with something visible showing where you are. A screen reader has to announce something meaningful at each control, since an unlabelled icon button reads out as "button" and nothing more. Contrast gets measured on text and on interface elements, including the state changes people depend on. Errors have to be announced and not signalled by colour alone. Motion has to be reducible and time limits extendable. And focus must not be trapped anywhere, which modals do constantly.

Three honest notes. Automated checking catches a minority of the real issues, so the manual keyboard and screen reader pass is where the substance is. Meeting the standard is a defensible position and it is not evidence that anybody can use the thing, which stays a separate finding in the report. And we are not lawyers. Sectors carrying their own obligations, public institutions and regulated health services in particular, should have the legal read done by counsel, and what Brand Vision can do is tell you where the exposure sits and document it. Remediation that changes how controls behave becomes focus and state work and gets scoped that way.

A severity-ranked findings report, the evidence under every finding, and a roadmap with an order and a skip list, all readable by a person who attended none of it. Nine components, and the last two get quoted back to us more than the first seven.

1. An executive summary for whoever reads one page and no more. The three things costing you most, the largest opportunity, and what happens first.
2. The opening measurement. Every relevant number as it stood on day one, written down so the same measurement can be taken again once the work ships.
3. The findings, each rated for severity. Severity here means what it costs a person. Blocking, so they cannot finish. Serious, so most people who hit it lose time or lose confidence. Minor, so it is untidy and nobody stops. Then a fourth grade kept separate on purpose, cosmetic, so it never competes with the first two.
4. Effort written beside every severity. In whose week it consumes and not in hours, so a blocking issue somebody fixes this afternoon sits next to a minor one needing a quarter.
5. A value estimate on each finding, with its assumptions printed beside it.
6. Annotated screens. The screen as your customer sees it, the problem marked on it, the principle it breaks written alongside. A visual problem described in prose gets argued about for a week.
7. Clips. Where behaviour is the evidence the moment itself is attached, from a session recording or a walkthrough capture. Thirty seconds of somebody failing ends a debate faster than any paragraph.
8. The prioritized roadmap. Surviving findings in a sequence, grouped into what ships next week, what waits for a release, and what is honestly a project. Dependencies marked, so you can reorder without breaking a chain.
9. The skip list. Each entry carries the reasoning underneath it. This is the one section where the author can be proved wrong in public, which is why most reports drop it, and clients quote ours more than anything else we write.

Two deliberate choices. The whole thing assumes somebody else may carry it out, which rules out recommendations depending on a conversation with us or on software we license. And anything that is really a project instead of a repair gets named as one and priced on its own, because a rebuild buried in a bullet point is a sale nobody agreed to. Execution can stay in house, come to the wider design practice, or be split, and the range under one roof is there if you want it. You can see what the fixes looked like afterward before deciding whether the format suits.