From Website to CRM to Campaign in Three Days

A client asked us on a Monday for a site, a working enquiry pipeline and a campaign to point at it. We had the first version in front of them on Wednesday. Not a mockup - a live site, forms landing in a CRM, follow-up mail going out on its own, and a set of images nobody had to brief a designer for.

Two years ago that is three weeks of work and four people. The interesting part is not that it got faster. It is which parts disappeared.

The old shape of the work

A small launch used to break into roughly five jobs: design, build, integrate, produce assets, then set up the marketing. Each waited on the one before it. Design waited on copy. Build waited on design. The CRM integration waited on the build. Assets waited on somebody with a licence for the design tool and a free afternoon.

None of those dependencies were technically necessary. They existed because each step needed a different person, and people are not available in the order your project needs them.

That is the constraint that has actually changed.

What each tool is genuinely good for

There is a lot of noise about AI tooling. Most of it treats these as interchangeable. They are not - each of these earns its place for a specific reason.

Monday: the site

Start with content, not layout. Every hour spent arranging boxes before you know what goes in them is an hour spent twice.

We use a CMS-driven front end, so the pages are a consequence of the content model rather than a set of hand-built files. A content type gains a field, every page that reads it updates. A new section is a type in the CMS, not a route in the code.

The AI contribution here is narrower than the marketing suggests. It writes components well, it reads an unfamiliar codebase quickly, and it is genuinely good at the mechanical parts - a form with fourteen fields, an accessible dropdown, a responsive grid that behaves at 320px. It is not good at deciding what the page should do. That is still the part that needs a person with an opinion - and it is why our AI consulting and strategy work starts with questions rather than code.

Tuesday: the CRM connection

This is the step that used to be quietly expensive. A contact form is trivial. A contact form that reaches the right pipeline, with the right owner, tagged with where the lead came from, that fires a notification and does not lose anything when the CRM is briefly down - that is a day's careful work.

We usually land these in TinTorch CRM, but the shape is the same whatever the destination. n8n turns it into a flow you can look at. Form submission arrives, enrich it, branch on the answers, create or update the contact, notify the owner, log the result. When something breaks you see which node went red, which is a considerable improvement on reading application logs at eleven at night.

Two things worth doing properly here, because they are the ones that hurt later:

• Make the form the source of truth, not the CRM. Store the raw submission first, then push it onward. If the CRM call fails you still have the lead.
• Put an idempotency key on the create step. Retries are a feature of every queue, and duplicate contacts are the fastest way to lose a sales team's trust in the system.

Wednesday: the campaign

Assets used to gate launches. A landing page needs a hero image, a share card, three ad variants and a thumbnail, and each of those was a request into somebody's queue.

Generating them directly changes the economics of testing. When a variant costs a minute instead of a day, you stop arguing about which headline is better and just run four.

A caveat worth stating plainly: this only works because the brand is doing the heavy lifting. Feeding a generator your actual logo and palette, and asking for very little text in the image, gets you something usable. Asking a model to render a paragraph of copy inside an image gets you the uncanny half-words everyone has learned to recognise. Keep the words in the HTML where they belong - they are better for accessibility and for search there anyway.

What did not get faster

Three things, and they are the three that decide whether the launch works.

Deciding what to say. No tool knows why a customer should choose this business over the one down the road. That comes out of talking to the client, and it takes as long as it ever did.

Verification. Generated code compiles and looks right and is sometimes wrong in ways that only appear on the fourth page. Generated copy states things confidently that nobody has checked. The time you save building gets partly spent checking, and a team that skips this ships faster into a wall.

The boring correctness. Redirects from the old URLs. Canonical tags that point at the right page. A sitemap that lists URLs that actually resolve. None of it is interesting and all of it is the difference between a site that ranks and a site that quietly does not.

The honest summary

The work has not become trivial. It has become unblocked. Five sequential jobs waiting on five different people has become one person with a stack of tools and a much shorter list of things they cannot do themselves.

For a small business, the practical consequence is this: the gap between deciding to fix your online presence and having it fixed is now measured in days. If your current site is three years old and your enquiries arrive as unstructured email, that is no longer a project you need to budget a quarter for. More of how we approach this is on our blog.