Chapter 5a: The Client Build — Discovery to Direction with Lucy's Bakery

The intake choreography: the inbound email gets NO requirements answered — the reply thanks, redirects to the questionnaire ('this is about them. Make it about them — by linking to the client questionnaire you are basically saying, hey, I am a professional'), and pre-books the call. The call-request email quotes the client's own stated goal back verbatim so it reads personal, not templated. On the call itself: 'you need to take control from the beginning because what many people do, they take on this employee role — like, oh, client, just tell me more.' Control means announcing the four phases, the agreement after each, and that backtracking costs money — 'trust me, it's a trap. I have fallen into it many times.'

The recap email closes discovery as a documentation move: restate everything agreed ('to make sure we don't lose any important information, I've written it down'), ask for confirmation, request brand assets, name the next phase. And a human touch that prevents a real misread: tell the client you're taking notes during the call — 'sometimes clients think you're on WhatsApp.'

The discovery lessons are one extended demonstration of inference over transcription. Lucy's answers, decoded: 'answer recurring bread questions + reflect who we are' = content goal, not conversion goal. 'Neighborhood regulars and new-to-town people' = LOCAL SEO only. 'Not sure how many pages' = his rule of thumb: 'when clients say this, it's a good idea to offer a 1 page design.' 'Update viewers' = a newsletter that is really weekly-bread notifications — 'think outside of the box when you're dealing with real clients.' No contact form, deliberately: 'conversations in the store is where it's at' — a form is 'extra management' she doesn't want. Already at full capacity = the site must NOT be growth-optimized: the overbooked dentist from Chapter 2, in the wild.

The call surfaces what forms can't: customer questions (ingredients, allergies, fermentation) become an allergy-tag system on the bread list, not an FAQ graveyard; 'what makes people come back' (craftsmanship, non-factory feel) becomes a photography requirement — real humans, prominent placement.

AI enters fourth, after competitors, client re-reading, and inspiration — as a CROSS-CHECK with context, never a cold start. The ChatGPT prompt is role-primed and constrained ('act like a web design professional... don't propose a multi page website if a 1 pager is enough'), fed the full questionnaire and call notes, and asked for both sections AND a business summary for the next tool. Its output validates the 1-pager and earns its keep with two genuinely new adoptions: an FAQ section and allergy-safe tags.

Relume gets the summary — and the critique: 'ReLoom has the tendency to create too many pages for simple websites... it clearly said 1 pager and then it still generated 4 pages or even 5,' e-commerce sections included despite explicit no-webshop instructions. Editing is drag/delete; the discipline is the unclicked button: never generate wireframes for the client-facing share, 'because your client is probably gonna start commenting there' — on blocky placeholder content instead of the structural question actually being asked. Second AI role later: drafting the sitemap's section copy in bulk, then re-prompted when it performs — 'don't act like a copywriter. This is a sitemap.'

The default, stated as a heuristic: modest content volume plus client page-count uncertainty = offer a single-scroll page. Depth without navigation comes from POP-UPS (extended bread info) so the structure stays flat; CMS editability questions (the client updating specials) are flagged as development concerns, not Figma concerns. Section ORDER carries psychology: About/Vision before product — 'you first want to know why something is good before you become interested in reading' — the settled sequence: hero (photo slider) → About/Vision → bread list → seasonal highlight + newsletter pairing → hours/location → contact-by-store → Instagram near footer. The uncertain item (a press section, seen on competitors) is neither included nor dropped: 'we're just gonna include it and see what the client says' becomes a client QUESTION — deferral as a first-class move.

Honesty note carried from the brief: the scroll-vs-click cognition claim arrives without citation — settled-sounding UX wisdom, flagged here as assertion rather than evidence.

Direction narrows in two passes of increasing resolution. COARSE, on the discovery call: two deliberately OPPOSITE Dribbble bakery screenshots (classic/minimal vs fun/bold) — Lucy picks playful/modern/human, and the entire style space halves before any design tool opens. FINE, in visual direction: two moodboards ('Option 1'/'Option 2') built from Dribbble AND Behance screenshots (Behance surfacing fuller brand cases), trimmed to 2-3 examples each — a third 'Leftovers' frame parks the excess, hidden from the client.

Delivery craft: the frames wire into a Figma PROTOTYPE (connectors, instant/fade), element offsets normalized so nothing jumps between screens, QA'd in an incognito tab — and the client receives the PROTOTYPE link, never the raw file. The email states a mild preference while leaving the door open, ties the recommendation to Lucy's own brand words, and offers the none-of-these exit. Her reply — 'leaning more towards version 2... just make sure to not make it too playful' — closes the gate: 'we don't really need an ultimate agreement. An email like this is good enough.' A cautionary Behance case rides along: a gorgeous logo with a weak site — 'branding designers are not web designers and web designers are not brand designers.'