Here's a truth every private-markets operator learns early: the diligence starts before the call. An allocator who hears your name doesn't schedule a meeting — they search you. In those ninety seconds, your website is the due-diligence artifact: the first data room anyone opens, and the cheapest place to lose a deal you never knew you were in.
Ollivierre Holdings — a private-markets holding company styling itself an Economic Growth Engine, built on four verbs: Position. Prepare. Connect. Scale. — was operating a serious institutional mandate without a front door that could pass that screen. Five divisions (Advisory, Capital, Academy, PHIAM, Holdings), each addressing a distinct buyer, flattened into nothing online. Inbound arriving by email and introduction with no typing, no record, no audit trail — untenable for capital-side conversations. And the firm's single strongest trust asset — a fourteen-person advisory bench of senior operators — living entirely in the founder's head and inbox. The conventional agency quote for the fix: six to ten weeks, 240–320 engineering hours. The mandate needed to be live in a fraction of both.
This is, by lines of code, the smallest build on this shelf — and one of the strongest stories on it, because the product is the delivery model.
The Build: Everything to the Standard of the Screen
Rocket Now ran the engagement as a fully agentic, Claude-driven build — and the loop, not the output, is the headline. Every substantive increment began as a written plan the client approved or redirected before code moved, so direction got corrected at the cheapest possible moment. Claude then generated the components, the schema migrations, and the security policies — and verified its own work against the running application: browser-driven exercise of the changed surface, console and network inspection, type and lint gates, before anything shipped. Plans as artifacts, not vibes; done means exercised, not "looks right."
The front door that passes. A 14-route institutional site — home, all five division pages telling five distinct stories to five asymmetric audiences, pillars, portfolio across four asset categories with active/pipeline status, a news hub with five-category filtering, about, hire, contact — under a design system with a hard acceptance criterion: no generic AI aesthetics. The client rejected the interchangeable look explicitly, and the shipped identity is defensibly bespoke — Playfair Display headlines over light DM Sans, pastel blue and warm grey with thin gold rules, editorial whitespace, a navigation bar that begins below the hero and sticks on scroll. Every colour lives as a semantic token; the identity is one file away from a rebrand. When "AI builds all sites the same" is the objection, then this site is the rebuttal.
The bench, made public. Fourteen senior advisors — former CEO-level operators, ex-Big-Four leadership, fund managers, deep-tech IP specialists — published with titles, focus lines, and verified LinkedIn links. The firm's best trust asset converted from private knowledge into a public screening surface.
The intake that leaks nothing. A dual-path enquiry flow asks the one question that matters at the point of capture — investor or founder? — validates, and writes into a status-tracked pipeline (new → reviewed → archived) behind a role-gated admin dashboard with filters, expandable detail, inline status updates, and a live new-enquiries counter. No public signup; access is provisioned deliberately.
And the part that makes this a reference-grade AI build: the model owned the risky surface. Claude authored the schema, enabled row-level security on 100% of public tables, pinned authorization to a security-definer has_role() function backed by a dedicated roles table — never a role column on a profile, the classic escalation footgun — as its default posture, not a post-hoc audit fix. The result: exactly one public write endpoint (the form insert), zero publicly readable tables, and authorization enforced in the database so a tampered client can read nothing at all. In most AI case studies, the AI writes the marketing page and a human writes the security. Here it was the other way around, and it held.
The Results
- From no institutional presence to a differentiated, division-specific front door across 14 routes — a site that passes the screen an allocator runs before the call
- From unstructured email to a typed, timestamped, status-tracked enquiry pipeline, investor/founder segmented automatically at capture, behind authentication and database-enforced RLS
- Fourteen advisors published with verified profiles — the bench visible at last
- Zero publicly readable tables, zero privilege-escalation surface, zero AI-generated defects reaching production — every increment verified against the running app before ship
- An estimated 85–90% effort reduction: roughly 20–30 hours of directed sessions against a conventional estimate of 240–320 engineering hours — with post-launch changes (a roster edit that historically means a ticket, a sprint slot, and an invoice) now shipping from one sentence in one session
"[Quote pending Ollivierre Holdings sign-off.]" — [Name, Title], Ollivierre Holdings
Why It Worked: The Right Step Was Building to the Screen
The Rocket Now principle, at institutional grade: a single step in the right direction is worth exponentially more than many steps in the wrong one. The wrong steps were the obvious ones — a template dressed up, a brochure with five divisions crammed into one story, security assumed rather than enforced. The one right step was accepting the audience's own standard as the build standard: if the website is the due-diligence artifact, then everything — the bespoke identity, the division-per-buyer architecture, the database-enforced authorization, the audit-ready intake — gets built to survive scrutiny. When the front door is built like a data room, then the firm passes the screen it never sees happening. And because the delivery loop made iteration nearly free, the client bought something better than a website: they bought never being blocked on a vendor again.
Operating a serious mandate behind an unserious front door? Build to the screen. → RocketNow.com
Editor's note — before publishing:
- Confirm client approval to name Ollivierre Holdings publicly; replace the placeholder quote
- The 240–320-hour conventional estimate and 85–90% reduction are the deal-registration's own estimates — keep them framed as estimates, as here
- The advisory bench descriptions are generalized in this piece (no individual names) — publish individual advisor references only with their consent
- "Zero defects reaching production" is the source doc's claim tied to the verification workflow — keep the qualifier "verified against the running app" attached to it