AI features are cheap when the groundwork exists, and expensive when it does not
The difference between a two-week AI feature and a three-month one is almost never the AI. It is whether the groundwork existed: content in structured form, data flowing through known endpoints, events already instrumented, and the business's knowledge written down somewhere a retrieval system can read. This page is the boring groundwork list, the things to do NOW, while AI is still optional, so that later it is a plug-in instead of an archaeology project. The AI-readiness checklist is the decision-layer version; this page is the site-layer version.
The five groundwork pieces
- Structured content. AI features read structure, not visual layout: products with clean fields, articles with real headings and metadata, FAQs as actual question/answer pairs (schema markup doubles as AI food). Sites whose content lives in tidy models get chatbots and summaries nearly free; sites whose content is welded into page builders get a content-migration project first.
- Clean data flows. Know where every form submission, order and enquiry lands, in what format, with what access path. AI features (lead qualification, summarization, routing) attach to those flows, undocumented flows mean the AI project starts with plumbing.
- Event instrumentation. The analytics events (leads, WhatsApp clicks, checkout steps) that measure the business today are the same events an AI feature will use as context and success signals tomorrow. The analytics checklist is the groundwork.
- Documented processes. Every AI-support feature answers from written knowledge: policies, procedures, product rules. If they live in the owner's head, the AI project includes writing them down, which is good for the business anyway, and is exactly what the documentation habit produces as a byproduct.
- An API surface. A site whose data and actions are reachable through clean endpoints (even simple ones) lets AI features attach without surgery. This is the quiet argument for the headless pattern and for avoiding tools that hold content hostage.
What this groundwork is worth later
When the AI feature conversation happens, support bot, product assistant, content operations, lead scoring, the quote is scoped against the groundwork. Sites with it: the AI plugs into structure that exists. Sites without it: the project begins with content modeling, event design and data cleanup at consulting rates. The groundwork costs the same either way, it is just done under a cheaper name today: "good website hygiene." The working method builds it in by default, the analytics and content-readiness checklists cover two of the five pieces for free, and the brief starts the rest when you are ready, today, while it is still called groundwork.