From 10 to 150 visitors a day in eighteen months

A twenty-person consultancy was getting ten visitors a day on its website and one inbound enquiry every three days. Eighteen months later the same site gets a hundred and fifty visitors a day and two enquiries a day, and one enquiry in four becomes signed work. Here is how it happened, step by step, without hiring anyone.

Where it started

The client is a twenty-person firm that advises retail chains on where to open their next outlet and what it can be expected to turn over. It is specialist work, sold to network directors and franchisees.

The website already existed. It described the company accurately. But it did no work. Ten visitors a day, one enquiry every three days, and a sales pipeline that therefore rested almost entirely on the network, on referral and on direct prospecting. This is not a rare case. It is the ordinary condition of a small expert firm whose site serves as a business card and nothing more.

The problem was not the volume of content, nor the design. It was the positioning. The site talked about the company instead of talking about the questions its buyers were asking.

The positioning diagnostic

We started with an analysis phase, using generative artificial intelligence as an instrument rather than as the subject. In practice that meant gathering and reading a great deal of material quickly: competitors’ sites, the searches buyers of this kind of study actually type, the pages already ranking for those searches, and the way each firm describes what it sells.

Three findings came out of it.

The first was that a significant volume of search existed around very concrete questions, asked well before the decision to open a site, and that nobody in the sector was answering them seriously. Traffic sitting there unclaimed, because everybody was publishing service brochures and nobody was publishing useful answers.

The second was that the players in the market all looked alike. The same words, the same promises, the same absence of proof. Genuinely differentiated content, showing the method and the numbers rather than promising them, therefore had room to breathe.

The third concerned conversion. Between the moment a development manager reads a page and the moment they feel able to ask for a quote, there is a gulf. Narrowing it meant making the price and the method legible before any contact.

The content strategy and the site

The strategy that follows fits in one sentence: publish the answers buyers are looking for, in the detail that competitors will not give.

The content was produced with the assistant, under a strict rule. Generative AI structures, drafts and keeps the whole thing coherent. It never invents substance. The figures, the cases and the method come from the firm. We apply the same rule at every client, and it is what separates a site that attracts from a site that readers and search engines both ignore.

The site itself was rebuilt from scratch, directly with the assistant, with no agency and no content management template. This way of working, sometimes called AI-assisted development, has two practical consequences. The cost of producing a page collapses, so you publish far more of them. And the site stays editable day to day, so it follows the positioning instead of freezing it for three years.

In parallel, HubSpot was put in as the customer relationship management tool. Without it, the newly generated enquiries would have fallen back into individual mailboxes, which would have cancelled a good part of the benefit. That side of the work is set out on our page on the marketing and acquisition lever.

Wiring the assistant into the data

This is the step that changed the nature of the work, and it is the least well known.

An ordinary chat assistant sees only what you paste into the conversation. So you export your statistics, summarise them, describe them, then copy its conclusions back into your tools by hand. Half your time goes on carrying data about.

There is now a standard way to connect an assistant to software and to data sources. It is called the Model Context Protocol: a normalised socket that lets an assistant read and write directly in your applications, instead of waiting for you to retype their contents. That is all it is, and it is enough to change everything.

On this project the assistant was connected to HubSpot, to the site’s source code, to the host’s traffic statistics, to Google Search Console, to Google Ads and to Google Analytics.

What changes when the assistant sees the data and the code at once

Take a mundane question: why is this page not converting.

Handled the usual way, it occupies three people for two weeks. Somebody exports the page statistics. Somebody else looks at the searches it appears for. A third opens the code or the content manager to change the text. Between each step, files have to be sent and intentions explained.

When the assistant sees the acquisition statistics and the site code at the same time, the question is dealt with in a single sequence. It observes that the page gets impressions on one particular search but very few clicks, reads the title and summary shown in the results, notices that they do not answer that search, proposes a rewrite, and applies the change in the page source. You read it, you approve it. The cycle goes from two weeks to an hour.

The same mechanism works in the other direction. An assistant that can see HubSpot knows which enquiries actually turned into work. It can therefore trace signed business back to the pages that produced it, rather than reasoning about raw traffic. That is the difference between optimising an audience and optimising revenue.

One last effect, quieter, deserves a mention. This arrangement lets one person hold down work that previously needed an agency, a copywriter and a developer. For a twenty-person company with no marketing team, that is not a convenience. It is the condition that makes the whole exercise possible.

The results

The starting point was ten visitors a day and one inbound enquiry every three days.

The finishing point is a hundred and fifty visitors a day and two inbound enquiries a day. Of those enquiries, one in four signs. That last figure is the most important of the three, because it says the traffic gained is not decorative traffic. The people arriving on the site were looking for exactly what the firm does.

Each of the three levers found at the diagnostic played its part: traffic the competition had left alone, differentiated content, and conversion accelerated by openness about method and price. None of the three would have been enough on its own.

This kind of project is not peculiar to retail location analysis. Comparable work is under way at a company in the remote security monitoring sector. The pattern transfers to any business whose customers look for answers before they look for a supplier. Other projects are described on our case studies page.

What we take from it for your own case

Three lessons stand out, and they hold well beyond this client.

The first is that positioning comes before content, and content before technology. Rebuilding a site without having answered the positioning question amounts to repainting a facade.

The second is that generative AI did not replace the firm’s expertise. It made it publishable. Everything that gives those pages their value comes from the client’s real studies.

The third is that you should connect the assistant to the data as early as you can. As long as it sees nothing, you spend your time describing your company to it. As soon as it sees your statistics and your sources, it works alongside you.

If your site gets few visitors and fewer enquiries, the first thing to do is not to rebuild it. It is to find out what your buyers are looking for and what your competitors are failing to give them. That is precisely the job of a ten-day diagnostic.