Process — design case study
Design thinking Read time ~7 minHow do you sell AI operations to people who are rightly tired of AI promises?
This is the thinking behind the AI Operations Launch™ site: the brief, the credibility problem at its centre, the direction chosen, the three directions rejected, and the trade-offs accepted along the way. It is written the way the work was done — decisions first, decoration never.
The client & the brief
Orvantis Framework No. 03Orvantis Intelligence is an AI-adoption consultancy with a deliberately quiet voice: measured, human-governed, allergic to hype. AI Operations Launch™ is its productised offer — a 30-day framework that wires AI voice agents, WhatsApp sales agents, web chat, lead qualification, booking and follow-up into one system, with every conversation captured to CRM, staff handover rules, and a daily executive summary.
The brief asked for a flagship page that could carry a real engagement pipeline. Three requirements were fixed from day one:
- Communicate the outcome, not the technology. Faster response, more booked customers, better follow-up, on record — never model names, never “powered by”.
- Stay honest. No invented client results, no fabricated metrics. Industry figures allowed only as clearly-labelled illustrative ranges.
- Feel engineered. The offer's whole differentiation is discipline — a programme, rules, measurement. The page must embody that discipline, not just claim it.
The parent brand's site is a separate, deliberately cinematic experience — Orvantis Intelligence. This page is its practical, daylight counterpart, and each links to the other.
The problem — a category that burned its own credibility
Survey Constraint mapEvery AI-agency site makes the same three moves: a dark gradient, a wall of logos, and a promise with a percentage on it. The audience has learned to discount all three. That's the real design problem — not “make AI look impressive” but make competence look different from hype, using only what can honestly be shown.
The constraints sharpen it further. There are no client case studies to lean on (the offer is productised, and the collection's rules forbid inventing results). The deliverable is invisible — wiring, rules, and habits, not a shiny app. And the reader's attention is shortest exactly where most AI sites spend longest: the part where they explain how clever the technology is.
So the page had to generate trust structurally: through specificity (a named programme, day by day), through governance shown as a first-class feature (handover rules drawn as a decision tree, not buried in an FAQ), and through a visual language borrowed from a profession that has spent a century earning the reader's default trust — the engineering drawing.
Direction chosen — and three rejected
Fig P-01 Options A–DChosen — D, “the blueprint.” Present the offer as a technical drawing set: numbered sheets, figure captions, a revision block, a form that reads like a document you countersign. The signature moment makes the promise literal: the system is drawn as a schematic that assembles itself, then runs — message packets travelling the wires, re-routed live by scenario. The page doesn't say “we build reliable systems”; it behaves like one for the entire scroll.
- OPTION A — REJECTEDDark neural gradientThe category default: obsidian, glow, particle mesh. Rejected because it is visually indistinguishable from the hype the offer competes against — and the parent Orvantis site already owns the dark cinematic register in this collection.
- OPTION B — REJECTEDDashboard-first SaaS pageLead with product screenshots and metric cards. Rejected: the offer is a service framework, not a SaaS; convincing screenshots would require inventing numbers, which the honesty rule forbids. The two spec renders that survived are clearly labelled as spec renders.
- OPTION C — REJECTEDCinematic scroll narrativeA slow, chaptered story in the Orvantis register. Rejected: this audience wants the terms, not the atmosphere — and the collection's rule of one exclusive technique per site reserves that language for the parent brand.
- OPTION D — CHOSENThe blueprint / sheet setLight, technical, warm. Trust borrowed from engineering convention, then earned by specificity: 30 named days, 6 handover rules, 4 measurable targets, 1 revision block. Every motion behaves like an instrument.
Key decisions & trade-offs
D-01–D-07 With costs attached- D-01Light palette against a dark categoryPaper and ink make the page read as a document, not a demo — at the cost of instant “AI-ness”. Accepted: differentiation was the point.
- D-02Figures only as labelled rangesThe problem section quantifies missed conversations with illustrative industry ranges, each stamped as such, with a plain-language disclaimer. Weaker than a hard claim; trustworthy where a hard claim would not be.
- D-03Targets, not testimonialsWith no client results to show, outcomes are framed as engineering targets (T-01–T-04) the install is measured against in week four — verifiable structure instead of borrowed praise.
- D-04Handover rules as a centrepieceStaff-in-the-loop governance gets its own sheet, drawn as a decision tree whose every branch ends in the CRM. It slows the sales narrative down; it also answers the reader's real fear before they've voiced it.
- D-05SVG over canvas for the signaturePackets on SVG paths keep the diagram accessible, zoomable and CSS-stateable. The trade-off is a hard ceiling on packet count — irrelevant here, and the discipline keeps the scene calm.
- D-06Metaphor carried to the fine printSheet numbers, drawing numbers, Form F-01, a real revision block in the footer. The risk is preciousness; the mitigation is that every artefact carries actual information. Nothing is engraving for engraving's sake.
- D-07One photograph, exactly (Rev D)The first edition was all system, no consequence. The second adds a single AI-generated plate (Higgsfield Soul 2, art-directed as one site-record shot; no real person depicted) at the handover sheet — the staff member the rules protect — duotoned into the drawing's own inks and framed as Fig 06. One human proof-point; a second photograph would have started dissolving the drawing set.
What the motion is for
Every animated element re-states the offer. The diagram assembling is the 30-day install. The packets are conversations that don't get dropped. The scenario tabs are the system coping with reality — the missed call, the midnight enquiry, the complaint that needs a human. The day rail turns reading the page into completing the programme. Motion that couldn't justify itself in those terms was cut; that's why there is no parallax, no cursor confetti, no scroll-jacking.
What was hardest
Tone. The copy had to sound like someone you'd trust with your phone lines: confident without swagger, specific without jargon. The test applied to every sentence was “would this survive being read aloud to a sceptical business owner at 7:30 in the morning?” — the same hour their daily summary would arrive. Sentences that flexed were rewritten to inform.
Intended impact — framed as design goals
G-01–G-04 Capability shownBecause this is a live offer rather than a finished engagement, impact is stated the only honest way — as the design goals the page is built to and the capability it demonstrates:
- G-01 — Comprehension: a first-time reader can state the offer's what / how long / who stays in control after one pass. The success-criteria questions on Sheet P-02 are the acceptance test.
- G-02 — Differentiation: the page is unmistakable at thumbnail size — paper, grid, one orange signal — in a category of interchangeable dark gradients.
- G-03 — Trust mechanics: every number is either a labelled illustrative range or a stated target; governance is drawn, not asserted. Credibility comes from structure, not adjectives.
- G-04 — Craft as evidence: a zero-dependency, accessible, reduced-motion-correct build — code-drawn throughout, save one labelled photographic plate — is itself the portfolio argument: the same care promised for the client's operations, applied to the page selling it.
For how each of those goals was implemented — the packet engine, the symbol-sliver fonts, the four-pass iteration log — the guide is the companion document.