CDF — an AI deployment methodology

Is the organisation ready to deploy AI?

Before anything enters the schedule or any price is quoted

A systematic readiness assessment of the entire organisation — from operational staff through managers to the board. Over a hundred parallel AI-led interviews over one or two days at the customer's premises. No data leaves the organisation

We do not diagnose to sell more services. We diagnose to avoid deploying the wrong solution to the right problem — or the right solution to the wrong problem. That is why an initial assessment sometimes recommends not deploying AI

Where you are and where you want to go

Two to four weeks of conversations — with the CEO and sales team, IT architect and lawyer, HR and compliance. We combine these into a maturity assessment across four areas: data, governance, skills and psychological readiness. This produces an ACE Configuration Profile defining the stages, controls and documents appropriate for the organisation

The knowledge audit (F0.K) is often the most revealing part of the project. Key procedures exist only in the minds of two experts approaching retirement; commercial terms sit on one salesperson's drive. Without organising this knowledge, AI will confidently answer using incomplete data

Where there are many legacy systems, we conduct an F0.L assessment: which are ready for AI now, which require process or data redesign, and which should be left aside for the time being

Structure before tools

Where autonomous AI agents operate, governance is an architecture of control: who granted an agent permission, how independently it acts, what happens if it makes a mistake, and who finds out — how quickly and through which channel

The AI agent register is the central inventory of all agents in the organisation. Each entry records a digital identity, permissions, autonomy level (0–4), business owner, usage limit and emergency shutdown procedure

Uncontrolled agents appear in every organisation. Shadow Agent Governance is not about punishing initiative, but providing a path to approval and continuous monitoring

People genuinely supervise instead of just clicking “Approve”

Before an employee approves a critical AI-recommended decision — a loan application, supplier selection or contract change — the system asks 5–15 questions to check their understanding of key facts and consequences. Too few correct answers block approval

The hardest part of transformation

AI transformations fail because of people who are afraid, do not understand or feel excluded — not because of technology. Productivity drops by 10–30% in the first 3–9 months. A programme that does not plan for this is judged a failure precisely when it is on the right track

AI Champions in every department are ambassadors for change, the first line of support and a feedback channel to the AI board

CDF Academy: 40 hours across three tracks — foundational, role-specific and supervisory — each ending with internal certification. Employees leave with concrete skills: how to formulate instructions, detect a fabricated answer and report an incident

First real value in 2–4 weeks

A Cognitive Sprint is not a research project. It delivers value step by step, with built-in measurement of AI answer quality. If a pilot does not deliver value in 2–4 weeks, something is wrong with its scope, data or process

The Scale-or-Kill Gate: at the end of the pilot there is one question and no extension option. Production deployment costs on average 3–5 times as much as a pilot — a good decision to stop saves more than an unsuccessful deployment

From project to programme

This stage starts with a quality review of F0–F4 documents. For high-risk systems (AI Act Annex III), conformity assessment is required before use, and public bodies register the system in the EU database (Article 49). We guide the customer through this process

Redesign the process first, then add AI. Automating a poor procedure produces poor results faster — redesign holds 80% of the potential value

Ongoing operations for systems that think

AI systems break, models drift, knowledge bases age and regulations change. Without an operating model, an organisation loses most of the value it created within 6–12 months. CogOps is an ITIL equivalent designed for thinking systems — with a dedicated allclouds consultant

The Knowledge Freshness Index checks whether regulations used by agents are from this week or two years ago. The agent lifecycle runs from registration to retirement with memory archiving. Quarterly reviews update the AI policy

Questions about CDF methodology

12 answers