Website strategy — five decisions taken before the first design is drawn

A site that does not work is rarely badly made. Usually it is well made and nobody established what it was for — after which every decision is settled by a vote on taste, for want of anything else to settle it with. That is how you get a project in which everybody is right and nobody knows whether it worked.

Below are five decisions taken before anyone opens a design tool, each with the criterion that settles it. If one has no answer the project still starts — the answer is simply taken later, by accident, by somebody who was not meant to take it.

Before the first one: what to check on the site you have

If a site already exists, all five decisions are easier: some answers are already lying about. Four checks, a quarter of an hour each, none of them paid for.

How many visits, and from where. Google Analytics, twelve months, acquisition report. One number: how many arrived from search and how many from everything else. If organic traffic is near zero, no change of appearance reverses it — the search engine is not showing the site, and that is different work from a redesign.

Whether Google shows the site at all. Search Console, performance report, three months. Many impressions and few clicks — the problem is the titles and descriptions. Almost no impressions — the problem sits earlier, in indexing. The two look identical in a visit-count report and need completely different work.

Where people drop out. In the pages report, find the addresses with the most visits and walk them on a phone. Not on a desktop — on a phone, in the conditions a customer uses.

Whether the form works. Send an enquiry through your own form and check it arrived, to whom, and how fast. The most frequently found fault: forms mailing an inbox nobody reads, or landing in a spam filter.

Those four usually settle whether a new site is needed at all, or work on the existing one. To go through it against a ready list, we have a self-audit checklist — no sign-up.

What to check at your competitors, and what not to

Competitor analysis usually ends in a list of things "they have and we do not", and a budget spent catching up. The sensible version is narrower and answers three questions.

What they answer and you do not. Open five competitors' sites and note the questions they answer outright: price, delivery time, who it is for, what happens after signing. Every question they close and you leave open is a reason somebody picks them.

What nobody says. More often than an advantage you find a gap — a question customers ask at every meeting that no site in the trade addresses. Material for content with nothing to compete against.

What not to check. Colours, animations, "how it looks" — the easiest to copy and the least connected to the result. A competitor's site can look superb and sell nothing; from outside you cannot tell.

Decision 1: one sentence saying why the site exists

Not "so the firm is on the internet". It needs a verb and a number.

The criterion: if in six months the site does that sentence and nothing else — was the project worth it? If yes, that is the goal. If you hear "and also…", it was a wish list, not a sentence.

A B2B services firm might want fifteen enquiries a month from companies of more than ten people. A shop — average basket value up 20%. A manufacturer — its sales team to stop answering the same questions by email. Three different sites and three different budgets, though each is "a company website".

One indicator, not twenty

A goal gets one number, looked at monthly. Not a set of charts nobody opens — that is the difference between a measurement and a dashboard.

The indicator needs a baseline recorded before the project. Without it, in six months nobody can say whether anything changed, and the discussion becomes an exchange of impressions.

This decision settles the four that follow. That is why it is first and cannot be skipped.

Decision 2: who reads it, and at what moment

Personas in the version that is any use have one field: what question this person arrives with. The rest — age, job title, favourite media — changes no design decision.

Three questions, three different sites:

• "do I even need this" — the reader is at the beginning and needs the problem organised, not a price list. Closing a sale at this stage frightens people off.
• "what does it cost and what moves the price" — needs a range and what shifts it. No figure at all reads as "it will be expensive".
• "why you and not them" — needs proof: a specific job, a specific result, a specific name.

A site answering only the third question — which is most company sites — works solely for people who have already decided.

Where to get those questions without guessing

The cheapest source is inside the firm and nobody uses it: the sales inbox and the meeting notes. Write out the questions from the last twenty sales conversations and count which repeat. That is your list of sections in order of importance, derived from data rather than a workshop.

The second source is five conversations with existing customers — what they looked for before they found you, and what held them back. The answers usually fail to match what the firm thinks about itself, which is the most valuable thing about them.

Inventing personas round a table produces a person who wants exactly what you were going to build anyway: the most expensive way to confirm your assumptions.

Decision 3: structure by the customer's questions, not by your departments

The commonest architectural mistake is organisational, not aesthetic: the menu mirrors the firm's structure instead of the buyer's path. "Our services / About us / Case studies / Blog / Contact" describes a firm. "What it costs / How it works / Who it is for" are answers.

A correction while we are here, because this rule has circulated in web guides for years: the "7±2 rule" does not apply to navigation. It is George Miller's 1956 work on working-memory capacity when recalling sequences — phone numbers, for instance. A menu does not have to be memorised; it stays on screen. Limiting items to seven may make sense for other reasons, but invoking Miller while doing it is a misunderstanding.

The criterion for architecture: does an outsider carrying the question from decision 2 land on the right page without reading the menu carefully?

What a website should contain — page by page

Each page answers one stage of the decision. Setting it out before the design work is cheaper than moving content after launch.

A simple rule for internal linking: every text from the first stage leads to one specific text from the second. Not to the home page, not to "contact us" — to the next step.

Search phrases: intent, not volume

What counts is not how many people search but what they are searching for. "Website" has enormous volume and almost no value: it is typed by firms wanting to commission one and by students writing an essay. "How much does a website cost" draws a fraction of that — about 720 a month in our own Keyword Planner pull — and is a buyer's search.

A checkable rule: if the first results for a phrase are definitions and encyclopaedias, it is informational. If they are service pages and price lists, it is a buyer's phrase. That is the one worth writing for.

Decision 4: what it runs on

One figure worth knowing. WordPress powers about 41% of all sites on the internet — but the share is falling: from 43.6% in mid-2025 to 41.2% in July 2026, according to W3Techs. Six consecutive months of decline.

Not an argument against choosing WordPress. An argument for justifying the choice instead of defaulting to it.

What that costs is easier to see through rates than through per-technology medians, because the technology is not what you pay for. A published German agency price list puts a freelancer at €50–80 an hour, an agency junior at €80–110, a senior or lead at €110–150. And rates in both large continental markets are falling: the French freelance barometer (Silkhom, from 20,000+ freelance profiles, 2019–2025) shows median day rates for JavaScript roles down 10.6% front-end, 12.9% back-end and 16.6% full-stack year on year, while technical project management rose 5.5%; the German freelancer study (freelancermap, fieldwork winter 2025/26) recorded a fall in its overall average — the first decline since the survey began.

The difference between quotes comes not from the technology but from what is bought: with WordPress usually a ready theme to adapt, with a custom stack a build from zero. The amount for your own scope is in the website cost calculator.

Performance is not an aesthetic category here. According to the Google/SOASTA study from 2017, the probability of leaving rises by 32% as load time goes from 1 to 3 seconds, and by 123% at 10. The study has years on it and covered mobile traffic — the direction is stable and confirmed by later measurement. The figure for your own address is in the website speed test, no sign-up.

Hosting: where saving money comes back as a cost

Hosting is often the only line the buyer understands well enough to economise on, which is why it is often the most expensive economy. Single-digit euros a month separate a cut-price shared plan from a decent one — European shared-hosting renewal rates run from roughly €2.69 to €11 a month — and that difference turns into seconds of load time, and those into people who do not wait.

Three questions settle the choice, in this order: where the server physically sits (for traffic concentrated in one country this is measurable), what the backup looks like (whether there is one, how often, and whether anyone has ever restored from it), and who answers when the site is down on a Saturday. The specifications in the price list matter less than the answer to the third.

Decision 5: what happens the day after launch

The most expensive mistake in the project is made here, because it does not look like a decision, so nobody takes it.

Three things have to be settled before the start, not after:

1. Who measures, and what. One indicator from decision 1, looked at monthly by a named person.
2. Where the enquiry lands. A form emailing a general inbox loses enquiries. A lead has to reach somewhere somebody answers.
3. Who updates it, and how often. A site with no owner ages within a year, whatever it cost.

Integrations: what to connect and what to leave

The site should not be an island, but not every integration pays for itself in year one. The order that makes sense:

First: the enquiry into the place where the salesperson works. A CRM, or at minimum a shared inbox with an assigned owner. While enquiries land in one person's private mail, everything further is premature.

Then: the source of the enquiry, stored with it. Without that you cannot say which channel pays, and budget decisions get taken on instinct.

Last: automation. Email sequences, chatbots, automatic assignment. Sensible when there are enough enquiries that manual handling starts dropping them — not before. Automating a process that does not work gives you faster-moving disorder.

What not to connect at the start: anything needing hand maintenance if there is nobody to do it. Every integration is a component that will break one day and require somebody who knows where to look.

Scalability, or the cost of changing your mind

A question worth asking when choosing the technology, and one nobody thinks about in the first year: what will the change you do not yet know about cost? A second language, a shop, a client area, an integration with a stock system.

In closed solutions that change is sometimes impossible and ends in a rewrite. In open ones it can be expensive but feasible. The difference shows only in the second or third year — by which time the decision is irreversible.

Assume too that some of these decisions come back. A well-built site is not a final state but a starting point: the technical layer runs on its own calendar, independent of your plans, and after two or three years it asks for attention however well the launch went.

What this article does not contain

Two numbers that circulate in articles about website strategy and cannot be traced to a source: that strategy "increases conversion by 200–300%", and that firms with a plan achieve "five times the return". Both sound good, neither is checkable. If you meet them in a proposal, ask for the source — the question alone is usually enough.

The order, if you had to choose

They are taken in that order because each needs the one before it. If you have time for one conversation before the project, spend it on decision 1 — the rest derives from it. If you have time for two, spend the second on decision 5, the one that costs most when skipped.