Most conversations about a company website are conversations about how it looks, because appearance is the part everyone can have an opinion on. It is also the part that matters least to the person the site was built for.
A business website has four jobs. They are not a matter of taste, and each can be checked in a few minutes without any technical knowledge. This article describes the four, gives you a test you can run on your own site this afternoon, and explains how to tell the difference between a problem you can fix in a week and one that means the site needs rebuilding.
The four jobs
One: be findable by someone who does not know your name. Plenty of sites are only ever reached by people who already typed the company name into a search box. That is a business card with hosting costs. A working site is also found by someone searching for the thing you do, in the place you do it, using their words rather than yours.
Two: answer the questions that come before a decision. Every purchase has three or four questions standing in front of it. Do they work with companies my size. How long does it take. What does it cost, roughly. Do they do the specific thing I need. A site that answers those has done most of the selling. A site that answers none of them makes the visitor ask by email, and most people do not ask; they go and check the next result.
Three: make one action obvious. Not five. One primary thing you want a visitor to do, available from anywhere, phrased so it is clear what happens next. Sites that offer a call, a quote, a brochure, a newsletter and a chat window all at once are asking the visitor to make a decision about how to decide, which is a decision most people resolve by leaving.
Four: keep working without a developer. Prices change, staff change, you add a service. If every small edit requires a paid request to somebody who may or may not still be reachable, the site starts drifting out of date within months. The drift is invisible for a while and then it is the main thing wrong with it.
Everything else, including how it looks, is in service of those four. Good design is not decoration; it is what makes the four legible. But a beautiful site that does not answer the questions is failing at the job, and a plain one that does is not.
The 20-minute test
Run this on your own site. Be a stranger to it.
Open it on your phone, on mobile data, not on the office network. Count how long until you can read something. If it is more than a few seconds, that is the single largest problem you have, because most visitors will never see anything else. Turn off wifi to do this properly.
Read only the first screen. What does this company do, and for whom? If a visitor cannot answer that without scrolling, the top of the site is decorating rather than communicating. Headlines that could belong to any company in your industry are the usual cause.
Write down the four questions a customer asks you most often on the phone. Then find each answer on the site. This is the most useful five minutes in the exercise. The questions people repeatedly phone about are, by definition, the ones the site is not answering.
Try to make contact in one move from three different pages. Not from the contact page, from wherever you happen to be. If it takes a hunt, the site is losing enquiries quietly and there is no report that shows it.
Search for what you do, as a customer would phrase it, without your company name. Not to check a ranking, but to see what a stranger sees, and how your entry reads against the others on the page.
Ask someone in your team to change one sentence. Time it. If the answer is that they cannot, or that it needs a developer, note it. That constraint compounds over years.
Six checks, twenty minutes, no tools. Most sites fail two or three of them, and the failures are usually not the ones the owner expected.
What is worth measuring, and what is not
If you have analytics, a few things carry real information and several popular measures carry very little.
Worth attention. How many people arrive from search without your brand name, because that separates a working site from an online business card. Which pages people actually read. Where enquiries come from. How the site performs on a phone as against a desktop, since the gap is often large and always in one direction. And enquiry quality, which is not in any dashboard and needs a note kept by whoever answers them.
Worth less than its reputation. Raw visitor counts, which rise with irrelevant traffic. Bounce rate, which is high on pages that answer the question perfectly and send the visitor away satisfied. Time on page, which is high when text is confusing. And rankings for terms nobody searches, which are easy to win and worth nothing.
The honest summary: a site with 300 monthly visitors and eight useful enquiries is beating one with 9,000 visitors and two.
Four common failure patterns
| Pattern | What it looks like | What is actually wrong | Fix or rebuild |
|---|---|---|---|
| The brochure | Attractive, well photographed, says almost nothing specific. Every page could belong to a competitor | A content problem, not a design one. Nobody wrote down what the business actually does differently | Fix. Usually a fortnight of writing |
| The locked site | Simple changes require the developer who built it, and they are slow or unreachable | Either no content management at all, or one so awkward that nobody uses it | Depends. Sometimes a fix, often a rebuild |
| The owner's taste | Built around what the founder liked, ignoring how customers actually decide | Nobody asked customers what they needed to know before buying | Fix, if the underlying structure is sound |
| Fast in the office | Quick on desktop, painful on a mid-range phone away from wifi | Structural: heavy images, too many scripts, a page assembled slowly | Frequently a rebuild, because speed comes from how a site is built |
The pattern worth noticing: three of the four are content and decision problems wearing the costume of a design problem. Redesigning a site with the brochure problem produces a better-looking site that still answers nothing.
When the website is not the problem
Sometimes the site is fine and the disappointment is real. A few situations where redesigning will not help:
The problem is that nobody knows you exist. A site cannot generate demand on its own. If nothing links to it and nobody searches for what you offer, the fix is visibility, not layout.
The enquiries arrive and nobody follows them up quickly. A site that produces enquiries answered three days later has done its job. This one is common and nobody enjoys hearing it.
People need to do something, not read something. If visitors want to check an order, download a document, or see their own account, no amount of website will satisfy them. That is a different kind of build, and website, web application or custom software sets out the distinction.
The offer itself is unclear. If the business cannot state in one sentence what it does and for whom, the site will not resolve that. The sentence has to exist before it can be published.
An illustrative example
A regional services company with 25 staff has a site three years old. Enquiries have fallen. The instinct is a redesign.
Running the test finds something else. The site takes eight seconds to show anything on a phone because the homepage carries a full-width background video. Two of the four questions customers ask most, about coverage area and about how quickly they can attend, appear nowhere. The services page describes the work in the company's internal vocabulary rather than the words customers use, which is also why it does not appear in search results for those terms.
None of that is a design problem. Removing one video, writing two paragraphs and renaming three pages addresses most of it, and would cost a fraction of a rebuild. This example is hypothetical, but the shape recurs: the visible symptom is disappointment with appearance, and the causes are speed, missing answers, and vocabulary.
Common questions
How often should a website be rebuilt? Less often than people assume. A site built with clean structure and editable content can serve five years or more, with content updated throughout. Rebuilding on a schedule, because it feels dated, is an expensive habit. Rebuild when the structure genuinely stops fitting the business.
Do we need a blog? Only with a real commitment to publish. A blog with four posts and nothing for a year is worse than no blog, because it advertises abandonment on a page carrying your name. When it is maintained, it is one of the few durable ways to be found by people who do not know you exist.
How many pages should a site have? As many as there are things a customer needs to know, and no more. Eight well-written pages beat 30 thin ones, and thin pages compete with each other for the same searches, which makes both perform worse.
Should the site show prices? It filters enquiries in both directions, and whether that is good depends on your situation. If most enquiries are people who cannot afford you, publishing a starting figure saves everyone time. If pricing genuinely depends on scope, a range or a worked example does most of the job without committing you to a number.
The short version
- Four jobs: be findable by strangers, answer the questions that precede a decision, make one action obvious, and stay editable without a developer.
- Run the 20-minute test on a phone, on mobile data, as a stranger. Most sites fail two or three checks.
- The four questions customers ask you on the phone are the ones your site is not answering. Start there.
- Judge the site on enquiries from people who did not already know your name, not on visitor counts or bounce rate.
- Three of the four common failures look like design problems and are content problems. Speed is the one that is usually structural.
- Sometimes the site is not the problem: no visibility, slow follow-up, an unclear offer, or a need for software rather than a website.
Where to take this next
If the test suggested the structure itself is the problem, the next decision is whether that means a refresh or a rebuild, and the two differ far more in cost and risk than they do in conversation.
If it turned out that what your visitors need is to log in and do something, that is a different build entirely, and website, web application or custom software is the place to start.
We are happy to run this test on a site and tell you what we find, including when the answer is that it needs writing rather than rebuilding. Discuss your idea.
# Visual recommendations
Concepts only. Hairline drawing, ink and brass, per 10-DESIGN-SYSTEM.md. No stock photography, no device mockups.
Hero visual — the four jobs
Concept. Four hairline panels in a row, each carrying a small abstract mark rather than an icon: a magnifier outline for found by strangers, three short answer lines for answers the questions, a single brass mark for one obvious action, and a small pencil rule for editable without a developer. Beneath all four, a thin baseline labelled appearance sits underneath these, not above them.
Purpose. Establishes the article's framing before the reader begins, and quietly displaces the assumption that a website discussion is a design discussion.
Composition. Wide and shallow, roughly 4:1. Brass used on the third panel only, since it is the single-action job.
Alt text. Four jobs of a business website shown as equal panels: be found by strangers, answer the questions before a decision, make one action obvious, stay editable without a developer.
Supporting visual — the same page on two connections
Concept. Two identical page outlines side by side under a shared timeline. The left, labelled office wifi, fills in almost immediately. The right, labelled mid-range phone, mobile data, shows the same page filling in stages across a much longer bar, with the readable text arriving last. No numbers on the timeline.
Purpose. Supports the first and most important check in the test, and makes the gap felt rather than asserted.
Placement. Inside the 20-minute test, after the first check.
Alt text. The same page loading on office wifi and on a phone using mobile data, where the phone version takes far longer and the readable text arrives last.