Case study · Flagship
Six months running a whole marketing function on AI systems.
From April to September 2026 the entire marketing function of an enterprise AI company preparing launch ran on one person and the AI systems she built for the job. This is that machine, phase by phase.
Sanitised case: the employer's numbers stay theirs. Every system described here runs in production, several are inspectable below, and I can walk your team through the full case. By Amber Lan, AI marketing engineer. Last updated 28 September 2026.
Explore all selected workThe brief.
A pre-launch company needed a full marketing department, built from one hire.
The surface area
Content, brand, design, events, outbound, search visibility and go-to-market strategy, all owned by a single operator. Whatever a team or agency would carry, she carried.
The maths
Working harder tops out at one person's hours. The only way the remit fits inside a week is to engineer the function, so output scales while headcount stays flat.
The story, in five phases.
Each phase turned a category of repeated work into a system, then moved on. By the end, the operator's week was judgement.
- CodifyThe brand voice and the visual identity became versioned, machine-checkable skills. A glossary and a claim register now gate what the company can say in public, so quality stops depending on who is holding the pen.
- EnginesA monthly content engine drafts a full artefact set across LinkedIn, X, email and web through codified executive voices, with every line passing a voice gate before sign-off. Graphics became HTML artboards rendered to on-brand PNG and PDF by a Playwright pipeline, and the same approach renders animation into MP4 video.
- DemandOutbound became an agent: enrichment through Apollo feeding sequenced touches through HeyReach and HubSpot across three senders, writing in the company voice with soft, invitational asks. Inbound became answer-engine optimisation: llms.txt, pillar pages and tracked citations, so AI engines cite the company on the category problems it named.
- OperationsA morning kickoff agent triages the workspace and inbox into a daily priority board. Ten strategic initiatives track roughly 190 tasks, and approvals moved from live pings to written gates, so the function keeps moving when the operator is in a room.
- In publicThe machine carried a conference season: a platinum summit sponsorship negotiated directly, booth collateral produced through the render pipeline, and a live interactive benchmark kiosk designed, coded and deployed solo for the events.
Key decisions.
Four calls that made one operator behave like a department.
Skills over prompts
Style guidance in a prompt decays. Versioned skills with explicit rules hold the standard for months, and they survive handover to a team.
The gate is the product
Generation is cheap. The audits, the claim register and the brand checks are what make output shippable, so most of the engineering lives there.
Judgement is the job
The systems draft. The operator judges. Her week collapsed from writing to deciding, the only part that ever needed a human.
Deliverables leave systems behind
Every campaign, deck and card was built so the next one costs less. Six months of work reads as infrastructure, and it compounds.
The result.
A marketing department's surface area, held by one person, with the standard written down.
The inventory
A codified voice and design system, a monthly content engine, a graphics and video render pipeline, an outbound agent, an answer-engine layer, a sixteen-document marketing foundation and a daily operating board. All in production, buying back about 10 hours a week on content and about 20 hours a week across three team members on outbound.
Evidence
The content engine case study and the AI-search toolkit are public. The benchmark kiosk is live. The skill library runs on this site.
What I learned
A one-person function is an engineering problem. The real constraint was how much judgement one person can apply in a day. Systems that draft and gate raise that ceiling.
Let's build this machine with your team.
Flow AI installs this same function inside your business: the engines, the gates, and the person who built them working as an extension of your team.