AI agency Australia, production AI deployed daily, not in pitch decks.
Australian AI-native marketing agency. 18 months of production AI deployed daily across the client book, not slide-deck demos. Gemini 2.5, Imagen 4, Veo 3, Vertex AI, BigQuery, Claude, GPT-class models, Cursor for engineering. Documented AI Governance policy published openly. Senior-strategist review on every AI-augmented output. Brand-trained Imagen models per retainer client. NL2SQL self-service for portfolio reporting. AI is in the toolchain, the strategist is human.
Gorilla Glue
Rose-Hip Vital
Six layers, all running in client work today.
"AI agency" without proof is a slide deck. Here's what's actually deployed across the 53-case-study book.
Gemini 2.5 + Claude for first-draft content
Long-form first-draft content (case studies, news posts, capability statements, pillar pages) drafted with Gemini 2.5 + Claude, then re-written by a senior content lead. Source: client transcripts, Xero / API data, brand-voice corpus. The draft compresses 6 hours into 90 minutes, the senior re-write is non-negotiable.
Imagen 4 with per-brand chassis
Per-retainer-client Imagen 4 prompt chassis (locked colour palette, lighting, camera lens, composition rules) generated through Vertex AI. Used for editorial / atmospheric imagery, not for depicting clients or factual scenes. Senior creative direction on every shoot. Photography direction →
Veo 3 for short-form motion
Veo 3 for short-form motion content where it fits the brief, atmospheric brand spots, B-roll variants, founder-content motion-frames. Live-action production when the brief calls for real people, real places, real product. Veo 3 augments the production schedule; it doesn't replace the production crew.
NL2SQL on BigQuery for self-service
Portfolio CMOs and PE-backed-group operators query the BigQuery portfolio dataset in natural language ("show me CAC by clinic across the last 12 months") via a custom NL2SQL layer. Output is reconciled SQL + chart. No analyst-bottleneck on routine questions. BigQuery portfolio pattern →
Cursor + Claude for code production
All web-build, integration, and analytics-pipeline work runs through Cursor with Claude / Sonnet in the loop. Pair-coding model, engineer drives, AI reviews + suggests, engineer commits. Deployment velocity on this 298-page website (built in-house) is the proof point. The CF Pages stack →
Gemini for client / category classification
Classification of 124 client websites across our 132-client book into vertical / sub-vertical / platform / engagement-shape taxonomy run through Gemini 2.5 with a custom rubric. Reconciled to Xero invoice records and Google Ads MCC + Meta Marketing API data. The classification underpins every "X clients in vertical Y" claim on this site.
What we will and won't do with AI on client work.
What AI does in our toolchain
- First-draft long-form content (case studies, pillar pages, news posts)
- Editorial / atmospheric imagery within a per-brand chassis
- Code production (Cursor + Claude pair-coding loop)
- Data classification and reconciliation
- NL2SQL self-service portfolio reporting
- Transcription, summarisation, internal-research synthesis
- A/B subject-line and ad-copy variant generation (creative-director-reviewed)
What AI does not do
- Strategy. The strategist is human, every time.
- Senior creative direction. Briefs are written by humans.
- Final-edit copy. Senior content lead re-writes every AI draft.
- Depict clients, real people, real places, factual scenes.
- Generate AHPRA-regulated medical-claim content without senior review.
- Fabricate stats, client quotes, or case-study numbers.
- Replace a junior practitioner, it accelerates them.
Read the full policy, including model-provider list, data-handling rules, client-consent posture, and senior-review-gate definitions. AI Governance policy →
Three case studies, three different AI applications.
For Health
9-sub-brand portfolio reporting with NL2SQL on BigQuery. Portfolio operators query CAC, booking-conversion, and clinic-level performance in natural language. Reconciled SQL + chart output, no analyst-bottleneck.
Charleston's
Per-brand Imagen 4 chassis for editorial / atmospheric campaign imagery. Locked colour palette, lighting and composition rules. Senior creative direction on every output. Inside the monthly capped retainer.
PYBAR Mining Services
Capability content + project case studies first-drafted via Gemini 2.5 from operator-interview transcripts, then re-written by a senior content lead. 52 unbroken months of cadence. Tier-1 panel-pursuit positioning.
Three ways Australian businesses engage us on AI.
$15K · 4-week assessment
Marketing-stack AI audit. We map your current state, surface where AI is genuinely useful (vs hype), draft a governance policy template, and recommend a 90-day deployment plan. Outcome: a documented roadmap and a costed proposal.
$15K Audit →3-6mo deployment retainer
Deployment retainer to operationalise the audit roadmap. Brand-trained Imagen chassis, NL2SQL portfolio reporting, content-pipeline AI integration, AI Governance policy rollout, internal training. Typical: capped monthly retainer for 3-6 months.
AI Automation →AI inside an existing engagement
For existing 121 Group clients on a marketing retainer, AI is in the toolchain at no incremental cost. Brand-trained imagery chassis, content-production AI, NL2SQL reporting all sit inside the standard scope. Already included.
Talk to your account lead →Questions Australian businesses ask first about AI.
How is "AI-native" different from "we use AI"?
"We use AI" is most agencies, ChatGPT for first-draft email subject lines and a Midjourney login. "AI-native" is what we mean: AI in the toolchain across content, imagery, video, code production, classification, and reporting; documented governance; senior-review gates; per-client brand chassis; portfolio-level NL2SQL self-service. The difference is operational depth, not vibe. The AI stack →
Which AI providers are in your stack?
Google (Gemini 2.5, Imagen 4, Veo 3, Vertex AI, BigQuery), Anthropic (Claude / Sonnet), OpenAI (GPT-class), Cursor for engineering. Plus a small set of category-specific tools (transcription, deliverability, ad-platform native AI). All listed in the AI Governance policy.
Will my brand data train someone else's model?
No. We use enterprise / API tiers across all providers, no training on customer data, by contract. Brand-trained Imagen chassis are tenant-isolated to your project on Vertex AI. Client transcripts and Xero / analytics data sit in our BigQuery, not in shared model-training pools. Detail in the AI Governance policy.
Do you use AI for AHPRA-regulated content?
For first-draft only, never for final-edit, never published without senior healthcare-content review. AHPRA-regulated medical-claim content is one of the explicit senior-review gates in the AI Governance policy. The AI compresses the production schedule; the regulator-aware human still owns the output.
Will AI replace my marketing team?
No, and we've designed our own operating model around that bet. AI accelerates senior practitioners and compresses junior production work. It doesn't write strategy, doesn't direct senior creative, doesn't replace the relationship between a senior marketer and a category. We staff the strategist; AI staffs the toolchain.
Want to deploy AI properly, without the slide deck?
Thirty-minute call. We'll talk through your current marketing-stack AI usage, where the genuine opportunities are, what your governance posture should look like, and what the deployment roadmap costs. No commitment.