The ICPC framework: Identify, Frame, Produce, Capitalise

The ICPC framework is the GenieFactory method for structuring the agentic transformation of a company, project by project. Four steps, four deliverables, and a capitalisation loop that ensures each project enriches the next.

Why is a framework necessary?

Because 80% of Data and AI projects fail, and the cause is never technical. It's a methodology problem: wrong use case (38% of failures), tool chosen before the problem (19%), no sponsor (24%), no change management (12%).

The ICPC framework addresses each failure cause with a dedicated step. It's not a waterfall process, it's an iterative cycle where the last step (Capitalise) feeds the first (Identify) of the next project.

Step 1: Identify, choose the right use case

Map potential use cases and score each opportunity on two axes: business value and technical feasibility. Don't start from the technology, start from the problem.

The one-sentence ROI test: if you can't write the expected ROI in one sentence ("this project will reduce working capital by €2M by recalibrating MRP parameters"), the use case isn't scoped tightly enough. Rephrase it or move to the next one.

Deliverable: a prioritised portfolio with an ROI formulated for each use case. At GenieFactory, this identification goes through structured business interviews, AI-guided, validated by experts.

Step 2: Frame, align business, tech, and leadership

This is the phase everyone skips, and the #1 cause of failure. 71% of IT projects fail through deficient framing. The problem isn't that people don't frame, it's that they frame in silos. The business writes a spec that tech doesn't understand. Tech builds what they understood. Leadership discovers the result 6 months later.

ICPC framing aligns all three parties on the same scope, the same objectives, the same constraints, before any line of code.

Deliverable: specifications validated by all three parties, ready to develop. Not a 200-page document, an operational framing that fits in a few pages. At GenieFactory, L'Établi generates these specs from the business interviews, with automatic cross-validation.

Step 3: Produce, deploy in real conditions

Build and deploy AI agents in real conditions. Not in a lab. Not in a demo. In the field, with real users, real data, real edge cases.

Two non-negotiable principles:

Human-in-the-loop is in the architecture, not optional. Systems with human validation have 4.3× fewer critical incidents and +372% median ROI vs +268% without. AI proposes, the human validates.

Deployment is fast. Under 6 weeks: 5% failure rate. Over 24 weeks: 31%. If the framing is good, production is fast.

Deliverable: an application in production, adopted by the teams. Not a prototype sleeping on a staging server.

Step 4: Capitalise, encode what was learned

This is the step nobody does, and the one that changes everything. Each project produces knowledge: documented processes, explicit business rules, reusable patterns, domain ontologies.

This knowledge is encoded in a Knowledge Graph, an organisational memory that belongs to the company. The next project doesn't start from zero. It builds on everything learned before.

That's how you go from "one AI project" to "a transformation." And it's the fundamental difference between digital transformation and agentic transformation.

Deliverable: documented AI capital, owned by the client. Enriched Knowledge Graph, identified patterns, reusable sector templates.

How does the ICPC cycle accelerate over time?

The Capitalise → Identify loop is the key mechanism. The first project starts from zero. The second builds on the first project's Knowledge Graph. The third benefits from the patterns of the previous two.

Concrete example: a supply chain diagnostic on an industrial mid-market company produces business ontologies (CTO/ETO assembly), scoring rules, diagnostic patterns. The next diagnostic, same sector or adjacent sector, starts with this base. Framing time decreases, diagnostic quality improves, ROI accelerates.

It's capitalisation that creates the defensive moat: the more the client uses the platform, the richer their Knowledge Graph becomes, the higher the switching cost. The asset is proprietary, but it's all the more valuable for being built within the GenieFactory ecosystem.

Apply the ICPC framework to your project →

---

Related article: What is agentic transformation?