After dozens of enterprise AI engagements, we've converged on a three-phase delivery framework that consistently produces working systems instead of impressive-but-useless demos. Here's how it works.
What You Need to Know
- Enterprise AI delivery has three phases: Discover (map the opportunity), Build (ship the first capability + foundation), Scale (compound with subsequent capabilities). The order is sacred; the timelines are flexible.
- Discovery takes 4-6 weeks. Build takes 8-14 weeks. Scale capabilities take 3-6 weeks each. A realistic timeline from "we want AI" to "we have production AI" is 16-20 weeks.
- Each phase has a clear deliverable and a go/no-go decision point. You can stop after any phase with tangible value: a roadmap, a working system, or a scaled platform.
- The foundation is built during Build, not as a separate phase. Capability #1 and the shared infrastructure ship together.
Phase 1: Discover (4-6 Weeks)
Input: "We want to explore AI for our business."
Output: Prioritised roadmap with business cases and data readiness assessments.
The discovery sprint maps your opportunities, scores them, and produces a clear recommendation for where to start. The key output isn't a technology plan. It's a business plan that happens to use AI.
Go/no-go: Does the roadmap identify at least one capability with strong business impact, acceptable data readiness, and clear foundation potential? If yes, proceed to Build. If no, invest in data readiness or process redesign first.
Phase 2: Build (8-14 Weeks)
Input: Top-ranked capability from Discovery + data access.
Output: Production AI capability + shared foundation infrastructure.
This is where most of the engineering happens. The team builds the first capability and the foundation simultaneously:
Weeks 1-3: Data pipeline + knowledge base. Ingest and process the documents, connect to source systems, build the RAG infrastructure.
Weeks 4-8: AI capability. Build the specific application: the claims intelligence tool, the contract analyser, the knowledge assistant. Integrate with existing systems.
Weeks 9-12: Production readiness. Security, governance, monitoring, testing. User training and phased rollout.
Weeks 12-14: Stabilisation and optimisation. Production data flowing, feedback loops active, performance tuning.
Go/no-go: Is the first capability delivering measurable business value? Is the foundation ready to support additional capabilities? If yes, proceed to Scale.
Phase 3: Scale (3-6 Weeks Per Capability)
Input: Working foundation + next capability from roadmap.
Output: Additional production AI capabilities, each faster and cheaper than the last.
This is where the compound advantage becomes visible. Each new capability reuses the foundation infrastructure (data pipeline, knowledge base, integration framework, governance patterns) and adds only the unique components.
The compound test: Capability #2 should take 50-60% of the time capability #1 took. Capability #3, 40-50%. If each capability takes the same effort, the foundation isn't working.
Capability transfer: Throughout Scale, your internal team takes increasing ownership. By capability #3-4, they should be leading delivery with the AI partner in an advisory role. Building a sustainable AI team is part of the delivery, not an afterthought.
- Can we skip Discovery and go straight to Build?
- If you already know exactly what to build, why, and what data you need, yes. In practice, even organisations with clear ideas benefit from a compressed 2-week discovery to validate assumptions and identify the foundation architecture. The cost of building the wrong thing is much higher than the cost of a discovery.
- What if we only need one AI capability?
- Build it. But invest 20-30% more to make the infrastructure reusable, because organisations that "only need one capability" almost always identify a second once they see what the first delivers. The marginal cost of a reusable foundation is small compared to rebuilding from scratch for capability #2.