Turning client onboarding into software.
A marketing platform serving financial advisors needed to turn client onboarding into software. Custom AI Studio built a modular AI content engine that reads one discovery call transcript and generates a firm's brand identity, website copy, blog articles, video scripts, and social calendar. Five content types, seeded from a single conversation.
The onboarding tax every agency pays.
Every marketing agency pays the same tax on every new client. Before a single deliverable ships, someone sits through discovery, works out what the client actually sounds like, builds a brand messaging framework from scratch, and produces the first real tranche of work: website copy, blog articles, video scripts, a social calendar. It is skilled work, it is slow, and it happens before the client has seen anything worth paying for.
The client, an AI content-marketing platform serving financial advisors, had proven that AI could compress that week of work into a conversation. What they had not proven was that it could run in production.
Their founder validated the concept the way founders usually do, by stitching one together. Figma for the interface, Make.com for the automation, custom GPTs for the generation. It demonstrated the idea. It could not be sold. There was no real database, so nothing persisted or accumulated. The Make.com workflows were fragile and expensive to maintain, and every new content type made them worse. Figma is a mockup, so there was no interface a customer could log into. They were demoing a product that did not exist yet.
Why off-the-shelf onboarding software didn't solve it.
Off-the-shelf client onboarding software organizes onboarding. It does not do the onboarding. Portals, task trackers, document collectors, and kickoff checklists make up a crowded, mature category, and none of it touches the part that actually costs money.
The expensive work is not tracking that brand positioning is due on Thursday. The expensive work is producing the brand positioning. Generic AI tools for marketing agencies have the inverse problem: they will write a blog post, but they have no idea who the client is, so every output gets rewritten into the client's voice by the person who was supposed to be saved the trouble.
The gap this platform occupies sits between the two. It is onboarding automation that produces the deliverables, seeded from the one artifact that already holds the client's voice: the discovery call.
One transcript, five content engines.
Custom AI Studio rebuilt the platform as a production-grade, modular AI content engine: five generation modules orchestrated in N8N, running on a real relational database behind a deployable frontend. Each module is triggered from the dashboard, and every one is seeded from the same discovery call transcript.
Brand identity generator. A three-stage pipeline that produces the firm's brand foundation. Stage 0 analyzes communication style from the transcript: tone, formality, signature phrases, and what the advisor thinks makes them different. Stage 1 generates vision, mission, core values, core offers, and brand behaviors, with word-count validation. Stage 2 produces the positioning statement, the core belief, and what the platform calls the "Monster," the primary pain a firm's clients are actually fighting.
Website content generator. Full copy for five pages (Homepage, About, Approach, Who We Serve, Services), generated from scratch or through page templates such as "Values First" and "Philosophy First," depending on how the firm wants to lead.
Blog article generator. Long-form SEO content with keyword optimization, meta descriptions, and compliance scanning, which is a requirement rather than a nicety in financial services.
Video script generator. Audience-tailored scripts in a clean narrative format, inferring context from the title and the selected client life-stage.
Social media calendar. A branded calendar across personas (Building Wealth, Nearing Retirement, Business Owners) with holiday-aware scheduling driven by a cleaned and expanded 83-entry calendar database tuned for financial-advisor use cases.
Contact and FAQ page generators were deliberately deferred from v1 pending design references. Everything else shipped.
Three bands, one brand foundation.
The system runs in three bands: onboarding, a brand pipeline, and content generation. Every band reads from the same brand foundation, which is why the outputs stay coherent across channels.
Onboarding. The advisor completes a discovery call. The transcript is uploaded to the dashboard, which fires a webhook to N8N. N8N pulls the full transcript from Supabase.
Brand pipeline. The three brand stages run in sequence against GPT-4o. A custom JavaScript node merges them into a single brand-foundation object and writes it to the database. This object seeds everything downstream.
Content generation. The user selects a content type. The frontend calls the relevant N8N workflow, which fetches the brand foundation, routes through a Switch node to the correct sub-workflow, calls GPT-4o, parses and validates the output, and writes it to the appropriate table. The dashboard renders it for review and approval.
Because every module reads from the same brand foundation, the blog sounds like the website, which sounds like the video, which sounds like the advisor. All of them trace back to the same conversation.
The hard parts weren't the prompts.
The interesting problems were not in the prompting. They were in making N8N behave reliably at volume. Four decisions were worth naming.
N8N's AI Agent nodes drop input context. Once execution enters an agent node, the original item data is gone, which quietly breaks any workflow that needs to know which record it was operating on. We wrapped the agent step in a looping architecture that processes items individually and preserves the link between payload and output.
The built-in structured output parser is unreliable. It fails often enough to be unusable at volume. We replaced it with custom JavaScript parsers that validate, clean markdown, and structure output into JSON.
Different page templates need different schemas. Each page template produces different section keys, so the output schema is dynamic and the frontend adapts to it rather than forcing every template into one fixed shape.
Blog generation is one-shot, not multi-stage. We chose this for cost efficiency and narrative coherence. A multi-stage blog pipeline costs more and reads worse.
What it runs on.
- Orchestration — N8N. All AI routing and content pipelines as webhook-triggered workflows.
- Data layer — Supabase / PostgreSQL. Transcripts, brand identities, generated content, and templates.
- AI model — OpenAI GPT-4o. Called via N8N AI Agent nodes.
- Frontend — Lovable (React). The advisor-facing dashboard.
- Version control — GitHub.
What shipped.
The build put five generation modules into production, each seeded from a single discovery call and reading from one shared brand foundation. Three things are worth stating plainly, with the cost figures given as internal estimates rather than measured client outcomes.
A three-stage brand identity pipeline reads a discovery call transcript and produces vision, mission, values, positioning, core belief, core offers, and brand behaviors, at an estimated $0.14 per firm.
Five generation modules run live: brand identity, website copy across five pages, SEO blog articles with compliance scanning, audience-tailored video scripts, and a persona-driven social calendar built on an 83-entry holiday database.
A full website content set is generated for an estimated $0.70 to $1.00 per firm, seeded entirely from one call, with every module reading from the same brand foundation so outputs stay coherent across channels.
Frequently asked questions.
What is onboarding automation for a marketing agency?
Onboarding automation for a marketing agency is software that produces a new client's first deliverables, not just the checklist that tracks them. In this build, one discovery call transcript seeds a firm's brand identity, website copy, blog articles, video scripts, and social calendar.
Can AI generate a full brand and content set from a single discovery call?
Yes. This platform reads one discovery call transcript, builds a shared brand foundation from it, and generates five content types from that foundation. Because every module reads from the same foundation, the outputs stay consistent across channels.
Why not use off-the-shelf client onboarding software?
Off-the-shelf client onboarding software organizes the process but does not produce the deliverables. The costly work is producing the brand positioning and first content set, which is the part this system automates.
What does it cost to generate a content set this way?
Internal estimates put the brand pipeline at roughly $0.14 per firm and a full website content set at roughly $0.70 to $1.00 per firm. These are Custom AI Studio cost calculations, not client-reported figures.