In a decade of enterprise AI work the failure pattern is consistent: scope unclear at kickoff, architecture chosen before requirements stabilize, governance retrofitted after the first incident. The methodology below sequences those three concerns deliberately — discovery before architecture, architecture before governance, governance before implementation. Every phase has an exit criterion; you do not advance because the calendar says so.
A precise scope is the most expensive deliverable. Most engagements that fail did so because the scope was hand-waved at week zero.
The architecture decision freezes more variables than any other choice in the engagement. It deserves more attention than it usually gets.
The cost of designing governance into a system is roughly 10% of the cost of bolting it on after an incident. Sequence accordingly.
Typical duration: 2-3 weeks. Engagement-ending exit criterion: scope and risk-classification document accepted by the named accountable owner.
The scope statement passes the "describe in one sentence" test: a stakeholder outside the project can read it and accurately predict what the AI will and will not do. If not, Discovery is not done.
Roughly one in five engagements stops at the end of Discovery because the answer the data tells us is "this AI deployment will not produce the value the sponsor expected." That outcome saves the engagement budget by an order of magnitude.
Typical duration: 3-4 weeks. Exit criterion: architecture document approved by engineering leadership, proof-of-concept demonstrably meets baseline metric.
The proof-of-concept matches the baseline metric (or beats it) on data the engineering team has not previously seen. The cost model is defensible at executive review. The architecture document is detailed enough that an external engineering team could build the system from it without ambiguity.
Typical duration: 2-3 weeks, often parallel to late Phase 2. Exit criterion: governance package signed off by accountable owner and (where relevant) risk/legal.
The governance package answers every question a regulator or auditor would ask without anyone needing to remember context. The rollback mechanism has been demonstrated in a non-production environment. The accountable owner accepts the system in writing with the controls documented.
Typical duration: 8-16 weeks (matches your engineering team's build timeline). Exit criterion: production-readiness sign-off across architecture, governance and operational dimensions.
Implementation is done by your engineering team or a delivery partner — not by us. Slavin AI's role in Phase 4 is to ensure the system that gets built is the system that was designed. We review code-level deliverables against the architecture, run governance checkpoints at predefined gates, and certify production readiness when all controls are demonstrably in place.
Same methodology, different role for us in the program.
We run Phase 1 and Phase 3 (Discovery and Governance Design). Phase 2 is reviewed but not led by us. Phase 4 oversight is light-touch.
Best for: organizations with strong internal architecture capability who need an independent governance and risk perspective.
Typical duration: 6-10 weeks of active engagement, then quarterly review cadence.
We run Phases 1-3 in full and provide Phase 4 oversight throughout the build. Engineering execution stays with your team or partner.
Best for: organizations new to enterprise AI or facing a regulated deployment where architectural and governance discipline are non-negotiable.
Typical duration: 4-8 months from kickoff to production readiness.
Acts in an interim role inside your organization — head of AI engineering, chief AI architect, or equivalent — for 6-12 months while you build the permanent function.
Best for: organizations standing up an AI capability who need senior leadership in seat immediately while permanent hiring runs.
Typical duration: 6-12 months, with explicit handover plan.
The maturity model, the twelve-control baseline and the LLM risk taxonomy the Phase 3 governance design builds on.
Read the AI Governance page →
The decision frameworks referenced in Phase 2 architecture choices: build vs buy, RAG vs fine-tuning, deployment topology, ROI measurement.
Read the AI FAQ →
Engagements where this methodology was applied — healthcare RAG, financial-services compliance assistant, manufacturing document AI.
Browse Case Studies →
AI closes the first 80% of any project well. Production is decided by the last 20% — failure modes, concurrency, data integrity, behavior under load, security, cost at scale — which AI does not model. This four-phase methodology exists to apply the judgment that catches the architecturally plausible but operationally lethal decision before it ships.
Read the full position →
An Architecture Review is a compressed pass through Phase 1 — two hours to identify scope, baseline metric, regulatory exposure and the highest-risk architecture decision.