What projects do
- Ship a deliverable against a stage-gate.
- Measure success by launch, not operation.
- End when the charter is closed.
- Leave a hand-over deck and degrade.
- Run on deadlines and heroics.
AI Factory / CoE Buildout
Stop running AI projects. Start running an AI Factory.
A Factory is the durable capability that produces agents continuously under governance. Not a programme. Not a vendor engagement. Not a slide deck. Built on four years running one that shipped 55+ production agents to 300% ROI.
- Production agents shipped
- 55+
- Processes mapped
- 250+
- SOPs documented
- 165
- ROI delivered
- 300%
The framing
Factory, not project.
The single biggest predictor of AI success is how you frame the work. MIT found 95% of GenAI pilots fail to scale. Projects degrade the moment the programme ends. Factories compound.
What factories do
- Ship production agents continuously.
- Measure success by throughput, SLOs, and ROI.
- Never "finish", the backlog is the roadmap.
- Standardise so marginal cost per agent drops.
- Run on operating discipline and compound interest.
Further reading: Why AI Factories Beat AI Projects, the full argument, with data.
The four pillars
What a Factory actually needs.
Skip any of these and the Factory collapses back into a project. In this order.
-
01
Process inventory
You cannot automate what you cannot see. Map 150–300 processes across 8–12 functions, score each for automation potential and governance risk, prioritise by marginal ROI.
The moat: 250+ processes mapped, 165 SOPs documented in a 4-year Group CIO engagement. Most "AI strategies" skip this phase and fail because of it.
-
02
Platform
Not a vendor relationship, a platform. Orchestration (n8n, LangGraph), LLM plumbing (LangChain and custom), enterprise RAG with permission-aware retrieval (Pinecone, Weaviate, pgvector, FAISS), MCP as the integration layer, observability, cost controls. Self-hostable, model-agnostic, evolving.
The tell: the model layer changes every six weeks. Your platform should not.
-
03
Governance
Human-in-the-loop by default. Responsible-AI policy. ISO 9001 and ISO 42001 alignment. EU AI Act audit-trail mapping. Auditable logs, escalation paths, vendor risk reviews, bounded autonomy per agent. Not overhead, the reason the business unblocks the rollout.
The tell: governance done well accelerates velocity. Governance done badly explains why projects stall.
-
04
Deployment standard
Every agent ships the same way: defined inputs, monitored outputs, fallback behaviour, named owner, SLOs, cost envelope. This is how you scale from one agent to fifty-five without the Factory turning into bespoke lineage nobody can maintain.
The tell: if every agent looks different, you don't have a Factory. You have a portfolio.
12-month roadmap
How the build sequences.
The order matters. Get it wrong and the Factory never reaches escape velocity.
-
Month 01
Readiness & design
Exec alignment, maturity assessment, governance charter, platform decision, team structure, 12-month roadmap.
-
Month 02–03
Process inventory
Map processes across 6–8 priority functions. Score for automation. Prioritise the first cohort of 5–10 agents. Document the SOPs.
-
Month 03–04
Platform stand-up
Deploy orchestration, retrieval, observability, MCP integration layer, deployment standard, governance toolchain. First CI/CD for agents.
-
Month 04–07
First cohort
Ship the first 5–10 agents to production under governance. Prove the throughput model. Capture unit economics per agent.
-
Month 07–10
Scale-out
Second cohort of 10–20 agents. Expand governance coverage. Operationalise the SLO dashboard. First ROI board report.
-
Month 10–12
Handover & compounding
Factory operating independently. In-house lead running cadence. Our role reduces to advisory. The backlog becomes the roadmap.
Proof
This isn't theory.
Between 2021 and 2025, while running Group IT at a diversified Oman conglomerate, I built this Factory for real, 55+ production agents, 250+ processes mapped, 165 SOPs documented, 300% ROI on transformation investments, 35% operational cost reduction. The case study walks through the engagement, the obstacles, and the outcomes.
- 55+
- Autonomous agents deployed
- 12
- Business functions covered
- 40%
- Response-time reduction via conversational AI
- 50+
- AI-powered workflows shipped
Engagement shapes
Two ways to start.
Start here
Factory Readiness
from A$20,000
4-week diagnostic + design. Process inventory sample, platform decision, governance charter, team model, 12-month roadmap with budget.
- Executive & function interviews
- Process inventory across 2–3 priority functions
- Platform architecture brief
- Governance & compliance gap analysis
- 12-month roadmap with cost & ROI projection
Full build
Factory Build
Custom
6–12 month embedded programme. Stand up the Factory, ship the first cohort, instantiate governance, train the in-house team. Co-delivered with your platform and function leads.
- Embedded factory lead (fractional or full-time)
- Process inventory across 6–8 functions
- Platform stand-up + deployment standard
- First cohort of 5–10 agents to production
- Governance instantiated and operating
- Handover to an in-house lead at exit
Usually paired with a fractional CAIO engagement for the first 6 months.
Common questions
Before we talk.
What is an AI Factory?
A standing organisational capability, a centralised unit that owns the platform, governance, patterns, and talent to ship production AI across the enterprise. Unlike an AI project, a Factory runs on throughput, SLOs, and continuous improvement.
Why does a Factory succeed where projects fail?
MIT found 95% of GenAI pilots fail to reach production. Projects ship a deliverable and degrade when the programme ends. Factories ship continuously, measure on throughput, and standardise so marginal cost drops over time. The framing is the force multiplier.
How long does it take to stand up?
4 weeks readiness. 6–12 months embedded build. Factories compound from there.
Is the Factory framework proprietary?
No, Nvidia, Microsoft, IBM, Publicis, and most major banks use variants. What we bring is four years of operator experience running one: 55+ agents, 250+ processes, 165 SOPs, 300% ROI, 35% cost reduction. We ship factories that actually produce.
What's the team model?
6–12 people at steady state: factory lead, architect, 2–4 AI engineers, 1–2 process analysts, governance lead, platform engineer, part-time change-management. Many roles start co-delivered and transition to in-house.
How does a Factory handle EU AI Act and other compliance?
Governance is a Factory pillar, not an afterthought. EU AI Act (2 Aug 2026), ISO 9001 / 42001, NIST AI RMF, responsible-AI policy all embedded as deployment-standard gates. Compliance becomes an accelerator, not a bottleneck.
Do you build AI agent factories for Sydney and Melbourne enterprises?
Yes. The practice is based in Melbourne and we run AI Factory engagements across Sydney, Melbourne, Brisbane, Perth, Adelaide, and remote. Discovery workshops happen where your executive team sits; the embedded build is typically hybrid (on-site kick-off + remote delivery, quarterly on-site leadership reviews).
How to build an AI agent factory, in order?
1) Framing (factory vs project). 2) Process inventory (map 150+ processes, score value x feasibility). 3) Platform and governance (orchestrator, model routing, observability, risk tiering). 4) First cohort of 3–5 production agents. 5) Factory cadence (one agent every 4–6 weeks, reusable patterns). Readiness covers 1–3; Build delivers 4–5.
Further reading
The Factory, in practice.
How the model was built, why the framing matters, and the delivery discipline that keeps agents shipping.
The 7-Phase AI Transformation Roadmap (From a Real Operator)
The 7-phase roadmap I used to ship 55+ agents at 300% ROI. In-order phases, hard gates, the pitfalls, no consulting deck, just the sequence that actually works.
Read analysisProcess Inventory: The Moat Nobody Maps
The single best predictor of whether an AI programme reaches production is whether anyone has mapped the processes first. Here's the methodology, the scoring model, and why it's the moat.
Read analysis55 AI Agents in Production: The Full Playbook (300% ROI, 35% Cost Cut)
How we shipped 55+ autonomous agents at 300% ROI and 35% cost reduction. Phase-by-phase playbook from an operator, the unsexy moves the 5% that ship actually do.
Read analysis
Book a discovery call
Let's start with the readiness.
30-minute call. We'll cover the shape of your organisation, where AI has stalled, and whether a Factory framing is the right move, or whether something lighter (a fractional CAIO, an MCP build, a process-discovery sprint) is the better first step.