3.8x content output. Same team. 34 clients.

The challenge

Running content production for 34 SMB clients simultaneously is a scheduling and consistency problem before it's a creative one. Each client has their own brand voice, tone guide, target audience, and content calendar — and each expects the content they receive to feel like it was written specifically for them.

The agency's three content strategists were managing all 34 accounts between them. They were good at their jobs, but the cognitive overhead of switching between client contexts — different industries, different voices, different briefs — was compressing the time available for actual strategic work. Most of each day was spent on first-draft production: blog posts, monthly newsletters, social captions, Google Ads copy.

The agency director was concerned about quality consistency under the production pressure, and about what would happen if they landed a few more clients. The team was already running at capacity. Taking on new business would either require new hires or a serious rethink of how content was produced.

The solution

Gilligan Tech built an AI Workflow Automation pipeline on Google Vertex AI (Gemini 2.0 Flash) that handles the first-draft production stage for all content types. The pipeline is driven by a brief — a structured form that captures the client, content type, topic, key messages, tone notes, and target audience. Strategists fill in the brief; the pipeline generates the first draft.

The critical element was brand voice calibration. Each of the 34 clients has a brand voice profile — a small corpus of approved content that defines their voice — stored in the system. Gemini 2.0 Flash is prompted with the brief plus the relevant brand voice profile, producing first drafts that are measurably closer to the client's established tone than generic AI output.

Cloudflare Workers AI handles real-time content preview rendering at the edge — strategists can see how a draft will look in the client's CMS or email template format without leaving the workflow interface, at sub-50ms latency. Completed drafts enter a review queue where the strategist approves, edits, or rejects before delivery. The pipeline never sends anything to a client directly.

Implementation

1. Brand voice profiling: For each of the 34 clients, Gilligan Tech assembled a brand voice corpus from existing approved content (blog posts, emails, social content). Corpus size ranged from 8 to 40 pieces per client.
2. Brief template design: A structured brief form was designed to capture the variables that drive content quality: content type, topic, key messages, word count, tone modifiers, audience description, and optional SEO target keyword.
3. Gemini pipeline: The generation pipeline combines brief + brand voice profile into a dynamic prompt for Gemini 2.0 Flash. System prompt engineering includes explicit tone instructions, content structure guidance, and prohibited phrases specific to each client.
4. Edge preview (Cloudflare): A Cloudflare Worker renders draft previews in client-specific templates at the edge. Strategists see publication-ready previews without switching tools or waiting for server-side rendering.
5. Review and delivery queue: All generated drafts enter a strategist review queue. Approved content is formatted for delivery; rejected content triggers a revision cycle with feedback sent back to the pipeline.

Results

Technology