The Last Mile. Where The Money Is Now That Anyone Can Generate A Website Before Lunch.

The single most expensive mistake an agency owner can make in 2026 is to keep quoting for the part of the job that the client has already done for free. They generated the site on Tuesday afternoon in between meetings. They typed eleven sentences into a prompt box, watched a preview render, and then sent you a Loom asking what happens next. The build was never the bottleneck. The launch was. And the launch is now the entire business.

Stop selling websites. Sell the last mile.

The client already shipped. They just don't know what shipping means.

Every week another founder forwards you a preview URL ending in some platform's subdomain and asks, in good faith, how to 'put it on the internet'. They are not stupid. They are not cheap. They have, in fact, done eighty percent of the work you used to charge fifteen thousand pounds for. What they have not done — what they cannot do, because no model has ever been trained to care — is the unglamorous tail of work that decides whether the site earns a single pound back.

That tail is now the offer. It is six questions long. Every one of them has a price tag attached. Every one of them is the kind of work AI builders are structurally bad at, because none of it is visible in the preview pane.

Question one. Is it accessible?

Generated sites are an accessibility lottery. Sometimes the model remembers alt text. Usually it does not. Contrast ratios are whatever the design system happened to roll. Focus states are decorative. Keyboard navigation works until the first custom dropdown, at which point it doesn't. Forms ship without labels. Icon-only buttons ship without aria-labels. Modals trap focus or fail to trap it, with equal carelessness.

In the EU the Accessibility Act has been in force since June 2025. In the UK the Equality Act has been quietly enforceable for fifteen years. In the US, ADA web suits have crossed four thousand filings a year. None of this is theoretical. A generated site that fails WCAG 2.2 AA is a legal liability the founder didn't know they signed. You run the audit, you produce the report, you fix the violations, you sign the accessibility statement. That is a fixed-price engagement, billed in advance, repeated annually. It does not require a designer. It requires someone who has read the spec.

Question two. Is it fast?

The preview pane lies. It runs on a fibre line in a data centre next door to the model, on a freshly warmed cache, on a laptop the founder paid three thousand pounds for. The actual customer is on a four-year-old Android on a train somewhere outside Reading with two bars of signal and seventeen tabs open.

Generated sites ship with the entire icon library imported. With hero images served as 4MB PNGs. With render-blocking fonts loaded from three different origins. With no image lazy-loading, no responsive sources, no preconnect hints, no cache headers. The Core Web Vitals are a horror show and the founder has never opened Lighthouse in their life. You open Lighthouse. You produce the before-and-after. You charge for the after. This is a tidy fixed-price package — performance audit, image pipeline, font strategy, CDN setup, monitoring — and it sells itself the moment the founder sees their LCP score on a real phone.

“The preview pane is a showroom. The customer drives the car home in the rain.”

Question three. Is it secure?

The founder generated a site, wired up a database with one click, and now has row-level security policies they cannot read, an anon key in their client bundle, a service role key somewhere they hope is safe, and a contact form that posts directly to an unauthenticated endpoint. They have not configured rate limiting. They have not enabled bot protection. They have not set a Content-Security-Policy. The 'security tab' in their dashboard is a list of red exclamation marks they have learned to scroll past.

Every one of those red marks is billable. RLS policy review. Secret rotation. CSP and security header rollout. CAPTCHA on public forms. Webhook signature verification. Login throttling. Backup verification. None of it is glamorous. All of it is the difference between a small business and a small business with a 2am incident, a leaked customer table, and a GDPR notification clock running. You charge for the audit. You charge for the remediation. You charge a retainer to keep watching.

Question four. Is it easy to use?

Generated sites are visually competent and behaviourally feral. The nav has seven items because the prompt mentioned seven things. The hero asks the user to do four things at once. The pricing page has three CTAs in the same colour. The mobile menu animates beautifully and hides the only link that matters. The form has eleven fields because the model assumed every business needs eleven fields.

This is information architecture, not design. Five users on a Loom. Two rounds of cuts. A single primary action per screen. A nav restructured around the actual customer journey rather than the founder's org chart. It is a week of work, it is the single highest ROI intervention on the site, and the founder will pay for it the moment you show them the recordings of real humans failing to find the buy button.

Question five. Will it rank in search engines?

No. Not the way it was generated. The title tag says the name of the template. The meta description is the first sentence of the hero. There is no sitemap. There is no robots.txt that wasn't copied from a tutorial. There is no schema markup. There are no canonical tags on the paginated archive. The H1 says 'Welcome'. There is no content because the founder has not written any content, because the model produced enough lorem-adjacent prose to fill the wireframe and everyone agreed not to look at it too closely.

Technical SEO is now a one-week fixed-price engagement: titles, descriptions, OG tags, sitemap, robots, schema, internal linking, redirects from whatever the previous site was. Content SEO is a retainer: keyword research, briefs, edits, publishing cadence, refreshes. Generative engine optimisation — getting cited by ChatGPT, Perplexity, Claude, Gemini, the AI Overviews — is a brand-new line item that did not exist eighteen months ago and that no in-house team has the bandwidth to learn. You learn it. You sell it. You charge a premium because nobody else in your client's town has heard the acronym yet.

Question six. Will it convert?

The site exists. It is accessible, fast, secure, usable, and indexed. None of that pays the founder's rent. The only metric the founder actually wakes up caring about is whether a visitor turned into a customer, a lead, a booking, a sale, a signup, a paying anything.

Conversion work is the highest-margin service in the entire stack and the one with the cleanest narrative. You install analytics that actually attribute. You set up the events that matter. You read the session replays. You write the hypotheses. You ship the A/B tests. You report the lift. You charge a base retainer plus a percentage of incremental revenue, or a flat fee per qualified lead, or a quarterly performance bonus — anything other than hours. Hours are how the old agency died. Outcomes are how this one eats.

Package it. Price it. Stop apologising.

The mistake is to keep selling these as line items on a proposal the founder reads as 'extras'. They are not extras. They are the entire product. Package them:

• →Launch Audit. Fixed fee, two weeks. The six questions, answered with evidence. Deliverable is a prioritised, costed remediation plan the founder can either approve or take elsewhere. Most approve.
• →Launch Sprint. Fixed fee, four to six weeks. Execute the plan. Accessibility, performance, security, UX, technical SEO, analytics. Site goes from 'generated' to 'shippable'. Signed off, documented, handed over.
• →Growth Retainer. Monthly. Content, conversion, monitoring, incident response, quarterly re-audits. The site is alive and someone is responsible for keeping it that way. That someone is you, for a fee that is not negotiable per hour.

Three offers. One ladder. Every client who lands at the top walks down it. None of them are buying a website, because they already built one. They are buying the rest of the job — the part the model could not see, the part the preview pane never showed, the part their nephew with the Lovable subscription is structurally incapable of finishing for them.

“The build collapsed to zero. The launch became the business. Charge accordingly.”

What this means for you on Monday.

• →Strip 'website design and development' off the homepage. Replace it with 'we launch the site you already built'. Watch the inbound shift inside a quarter.
• →Build the Launch Audit as a productised offer with a fixed price, a fixed scope, and a fixed turnaround. Stop scoping every engagement from scratch.
• →Hire and train for auditors, not designers. The next ten hires are people who can read WCAG, read Lighthouse, read RLS policies, read GA4. Taste is cheap now. Rigor is not.
• →Kill the hourly line item. Price every package as an outcome with a number on it. If you cannot put a number on it, it is not a service, it is a hobby.
• →Set a quarterly target for retainers, not projects. Projects are how you starve between Christmases. Retainers are how you sleep.

The work is not gone. The work moved. The agencies and soloprenuers who notice the move this quarter will spend 2027 turning clients away. The ones still quoting for 'a new five-page website, two rounds of revisions, four to six weeks' will spend it explaining to their accountant why the pipeline dried up. Pick which conversation you would rather be having.