How to implement AI
Two service lines for organisations that want to implement AI in compliance with regulations and maintain it in everyday work: CDF methodological consulting and WDR operational documents
In many companies and public offices, AI is implemented chaotically
Not a strategy presentation or a prototype, but a working way of organising work with AIMost consulting companies offer an AI strategy as a presentation or a prototype that never reaches everyday use. allclouds delivers a complete, working way of organising work with AI: from rules and oversight, through working methods, to organising company knowledge and ongoing maintenance
Our offering consists of two service lines
They can be ordered together or separately — the CDF line guides the implementation methodology, while the WDR line delivers the documents that stay in the organisation after the project
Eval · assessment · F0 · discovery · F1 · strategy · F1.5 · compliance · F2 · oversight · F3 · people · F4 · pilot · F5 · scaling · F6 · CogOps · WDR-01 · AI policy · WDR-02 · Working methods · WDR-03 · Working environment · WDR-04 · Naming · WDR-05 · Implementation plan · WDR-06 · Knowledge · architecture · An organisation working with AI · stays after the project ends · rules and oversight · working methods · organised company knowledge · ongoing maintenance · CDF LINE · METHODOLOGICAL CONSULTING · WDR LINE · AI OPERATIONAL DOCUMENTS
CDF line — methodological consulting
CDF is our methodology for implementing AI in large organisations (Cognitive Deployment Framework). Every engagement starts with an Initial Assessment (CDF-Eval) — a one- or two-day diagnosis on the client's premises, involving more than 100 employees and using our own computing environment. The result is a clear recommendation: implement, implement under conditions or do not implement — before you spend any money on the project
If the result is positive, we launch the full methodology: seven main stages (F0–F6), two diagnostic stages (F0.K, F1.5) and an independent audit (CDF-AUD). Each stage has defined entry and exit criteria, mandatory documents and measurable quality indicators — including AI reasoning quality indicators (Cognitive SLA), which are absent from any standard Agile or ITIL methodology
All CDF line servicesWDR line — AI operational documents
Six documents that stay in the organisation after the project ends and work regardless of which AI tools the company uses
- AI usage policy (AI-Operating & Working Agreement)
- A handbook of methods for working with AI for every position (AI Work Methods & Effectiveness Guide)
- An operating manual for the AI working environment (AI Workplace Stack & Configuration)
- A document naming standard designed for working with AI (Corporate Document Naming System)
- An implementation plan: automations, management dashboards, schedule (AI Implementation Plan)
- The company's knowledge architecture: a map of relationships, a glossary of terms, guidelines for knowledge retrieval systems (Corporate Knowledge Architecture)
Three things no one else offers
lower costs for organisations that build compliance in from the start — industry research, depending on the sector
the average cost of a production deployment compared with a pilot — a good decision to stop saves more than a failed deployment
For organisations in regulated sectors
- Finance — banks, insurers, leasing companies
- Central and local government administration
- Defence
- Healthcare
- Energy
- Critical infrastructure
How the services can be ordered
CDF line — methodological consulting
Scope and results of each stage. Each stage can be ordered as part of the full programme or separately; the full description of the methodology is on the CDF methodology page
Discovery and configuration
An assessment of how ready the organisation is to implement AI: data, processes, people's skills, readiness for change and the regulatory situation. On this basis we create the organisation's implementation profile (ACE Configuration Profile), which sets the scope of the following stages
- An AI readiness report with a score and a list of things to improve
- The organisation's implementation profile, which drives the following stages
- A map of processes, data and decisions where AI can add value, with an estimate of implementation costs and a baseline for measuring return
- A list of acceptance criteria for the whole project — an annex to the contract
Company knowledge audit and map
An inventory and assessment of the organisation's knowledge resources — written down (documents, procedures, systems) and the knowledge that exists only in experts' heads, on local drives and in informal notes. Interviews, documentation review, workshops
- A register of knowledge sources with a quality assessment and assigned owners
- A map of knowledge loss risks (e.g. key experts leaving) and priorities for securing it
- A preliminary map of relationships between the company's concepts and data (Knowledge Graph)
- Recommendations for the strategy (F1) and data oversight (F2) stages
Assessment of legacy systems for AI
A review of existing systems, applications and processes: where AI will add value, where the process or system must be rebuilt first, and where implementation would be too risky, too expensive or merely cosmetic (Legacy Systems & AI Opportunity Assessment)
- A system assessment report with a system register: criticality, limitations, AI potential
- A map of AI use cases assigned to specific systems and processes
- An assessment of the systems' integration readiness and data quality
- A priority list: implement / rebuild / postpone / do not recommend
AI strategy and architecture
Turning the F0 diagnosis into approved decisions: the overarching goal, the sovereignty model, costs over 3–5 years, “build or buy” decisions and a contingency plan in case of a change of supplier. Everything is approved by the board before moving on
- An AI strategy document: goal, indicators, budget — approved by the board
- A decision on the deployment model (cloud / own infrastructure / hybrid) with technical and legal justification
- A risk assessment of AI suppliers and models, including geopolitical risk
- A 3–5 year cost model (spreadsheet) and an action plan in case of losing a supplier
Compliance first
We check for gaps in regulatory compliance and build a complete set of documents before the technical implementation begins (Compliance-First Delivery). Mandatory for companies in the financial sector, public administration and other regulated industries
- A compliance readiness report with an AI Act assessment and a remediation plan
- A Statement of Applicability and an impact assessment report for each AI system (plus a data protection impact assessment where required)
- A security annex ready to include in contracts with IT and AI suppliers
- A 24-month regulatory schedule with assigned owners
AI oversight and security
Designing and launching the AI management system: who oversees it, what the permissions are, what the register of running AI agents looks like, how data protection and security work. The foundation for all the following stages
- An AI agent register and a responsibility matrix: who is responsible for what at each level of AI autonomy
- Rules for human oversight of AI and a playbook for dealing with uncontrolled AI tools used by employees
- A data management package (record of processing, quality, data life cycle)
- A specification for security and an immutable event log, ready for IT to implement
Change management and capability building
Preparing people to work with AI — in terms of skills and psychologically. We build internal change leaders who will develop AI after our project ends
- An AI leaders programme: recruitment, training (16 h), materials
- The CDF Academy: three training paths (40 h in total) with internal certification
- A change management plan: communication, schedule, indicators of AI adoption in the organisation
- Measurement tools: the team's psychological safety and an AI fatigue index with thresholds at which action is needed
Cognitive sprint — pilot implementation
2–4-week sprints focused on measurable AI value, fully auditable and with continuous measurement of answer quality (Cognitive Sprint). Each sprint ends with a report and a decision to continue, and the whole stage ends with the Scale-or-Kill Gate
- A working AI quality monitoring dashboard
- Reports after each sprint: results, incidents, recommendations, comparison with the previous sprint
- A preliminary evidence package for ISO/IEC 42001 certification (the AI management standard)
- A documented scale-or-kill decision, with verification of the F0 acceptance criteria
Verification, scaling and acceptance
Formal verification of the implementation and fulfilment of legal requirements before the system is handed over for everyday use, followed by a rollout to the whole organisation. It ends with an acceptance report and the launch of ongoing maintenance (F6)
- A conformity assessment report ready for external audit, with the required technical documentation and registration in the EU database (where applicable)
- Quality management system documentation compliant with the AI Act
- A catalogue of the company's AI systems with documented use cases
- An acceptance report and a staged plan for rolling out to the whole organisation, with readiness criteria
Cognitive operations — ongoing maintenance
Maintenance of AI systems in everyday use (CogOps): answer quality, up-to-date company knowledge, model ageing, the AI agent life cycle and continuous compliance with changing regulations. A monthly ongoing service with a dedicated allclouds consultant
- A monthly AI quality report: indicators, trends, incidents, recommendations
- A quarterly compliance review with an update of the oversight documents
- Ongoing updates of the AI agent register and the regulatory schedule whenever regulations change
- Support during external audits — on request, as part of the service
10 stagesfrom discovery to ongoing maintenance
Book an Initial AssessmentWDR line — AI operational documents
Six documents that stay in the organisation after the project ends and work regardless of which AI tools the company uses
AI policy and oversight rules
Developing or adapting the AI Usage Policy — the core document that sets out the rules, roles, limitations and procedures for all employees who use AI. Tailored to the sector, the organisation's structure and regulations; it is developed in workshops, over several review rounds, with final approval by the board
- An AI usage policy (30–50 pages) tailored to the organisation
- A matrix of AI roles and responsibilities
- A register of systems with built-in AI (Excel template + instructions) and rules on what is allowed, restricted and prohibited in dealings with clients and suppliers
- An onboarding path introducing new employees to the rules for working with AI
Methods of working with AI
A practical handbook for working with AI for every position on three levels: technical (how AI works, how to phrase instructions), methodological (how to process documents, how to defend against fabricated answers) and practical (ready templates for roles). Tool-independent, with a separate section for the client's tools; we build the templates on real cases from the company
- A handbook of methods for working with AI (150–200 pages) tailored to roles and processes
- A library of at least 30 ready prompt templates for individual positions
- A checklist of 10 layers of defence against fabricated answers (printed card + digital version)
- An overview of the client's AI tools with recommendations
Configuring the AI working environment
A complete operating manual for the AI working environment in three layers: how the company's AI tools are built, how employees use them every day and which safeguards protect data and processes
- An operating manual for the AI working environment (40–60 pages) tailored to the company
- Quick-start cards for every position (A4, for print or the intranet)
- A completed register of systems with built-in AI and a map of data flows in the company
- Security requirements for IT to implement
Document naming and order
A document naming standard designed for working with AI: machine-readable, consistent with data classification and ready for automation. Independent of the system in which the company keeps its documents (SharePoint, Google Drive, Confluence, Box and others)
- A full specification of the document naming standard
- A division into document areas and types tailored to the company's structure
- A staged plan for migrating existing documents to the new standard, with logic for automatic name checks
- Rules for maintaining the standard: owners and a change procedure
AI implementation plan
Turning documents WDR-01 to WDR-04 into an actionable plan: a list of automations, putting data in order (one reliable source for each data type), management dashboards and a schedule. The plan defines WHAT to automate and HOW, not on which tool — it goes to the client's IT or an external contractor as a ready specification
- An implementation plan (document + Excel sheet): 36 automations with priorities and a schedule
- A flow map of 14 data types with assigned owners
- A specification of 6 management dashboards: indicators, data sources, screen sketches
- A list of quick wins to implement in the first 4 weeks and a handover document for IT or the contractor
Company knowledge architecture
How the organisation's knowledge should be organised so that AI can use it safely and effectively: a map of relationships between concepts (Knowledge Graph), a glossary of industry terms, knowledge oversight rules and procedures for keeping it up to date. The starting point is the CDF-F0.K audit or the client's own analysis; tool-independent
- A knowledge architecture document with the rationale for decisions
- A map of relationships between concepts and data with owners, and the organisation's glossary of terms
- Knowledge oversight rules: who is responsible, how changes are made, how knowledge is kept up to date
- A technical specification for the team that will build the system
6 documentsindependent of AI tools
Book an Initial AssessmentWhere to start
Every engagement starts with an Initial Assessment (CDF-Eval): a one- or two-day diagnosis on the client's premises, involving more than 100 employees and using our own computing environment. The result is one of three recommendations — before you spend money on the project
AI awareness · and digital skills · Process maturity · and readiness for automation · Organisational culture · and readiness for change · Knowledge management · and data availability · Oversight · and regulatory readiness · IT infrastructure · and the state of data · Initial · Assessment · CDF-Eval · interviews conducted by AI · 1–2 days at the client's site · Readiness index · 0–100 · report for the board · max. 2 A4 pages · Implement · ready for the full CDF programme · Under conditions · a preparation plan with a schedule · Do not implement · we come back in 6–12 months · SIX READINESS AREAS · A CLEAR RECOMMENDATION