Knowledge | M.A.I. Consulting

The EU AI Act, as it now standsWhat Article 4 actually requires today, what the Digital Omnibus did and did not move, and who counts as a deployer.3 parts

Article 4 of the EU AI Act was amended. The wording you were told to comply with is goneBriefing · 3 min "Deployer" is the word that puts you in scopeBriefing · 3 min The Omnibus delayed high-risk rules and rewrote Article 4. Neither is a reason to waitAnalysis · 3 min

Start by seeing your own AI useThe evidence on delegation, then the three artefacts that make real use visible: a tool inventory, a data classification and an honest list of what models cannot do.4 parts

The 19-point problem: AI made consultants worse at one task in the study everyone quotesAnalysis · 4 min Data classification from zero: what can go into an AI toolGuide · 4 min The AI tool inventory: the least glamorous artefact that unlocks everything elseGuide · 4 min What a language model cannot do, stated plainlyExplainer · 4 min

Measuring readiness: the ten dimensionsWhat a readiness score measures, the ten dimensions behind it, and what to do with a mid-range result.6 parts

AI readiness dimensions 1-3: fluency, delegation practice, description disciplineGuide · 4 min The ten AI readiness dimensions, named and explainedGuide · 3 min What a readiness score is actually measuringExplainer · 4 min AI readiness dimensions 4-6: verification, logging, data protectionGuide · 4 min AI readiness dimensions 7-10: workforce impact, politics, safety, future-readinessGuide · 4 min Scoring 62 out of 100: what a mid-range score means and what to do firstFramework · 4 min

Choosing where to startWhich department goes first, how to pilot for evidence rather than anecdote, and when the honest answer is not to use AI.5 parts

Before/after measurement without surveillanceFramework · 4 min The pilot that produces evidence instead of anecdoteGuide · 3 min Which department should pilot AI first? Four questions to askFramework · 4 min Reading a vendor benchmark without being taken inGuide · 4 min When the honest answer is "you do not need AI for this"Analysis · 4 min

Making training stickCapability levels, attendance records, breadth versus depth, and why the playbook belongs to the team.5 parts

The seven capability levels, and why Level 3 is the floorFramework · 3 min The attendance record: trivial to keep, impossible to reconstructGuide · 4 min Training breadth vs behavioural depth: the trade-off nobody pricesAnalysis · 4 min Why the playbook belongs to the team, not the trainerExplainer · 3 min When leadership is the least fluent group in the buildingAnalysis · 3 min

Department playbooksWhat to delegate, what never to delegate and what to check, department by department.7 parts

AI playbook for Finance and Administration: what to delegate, what to keepGuide · 3 min AI playbook for Programmes and Monitoring: five to delegate, three neverGuide · 3 min AI playbook for Communications and Advocacy: what to delegate, what to keepGuide · 4 min AI playbook for HR: the department where most red lines liveGuide · 4 min AI playbook for Leadership and Governance: what to delegate, what to keepGuide · 3 min Fundraising AI playbook revisited: what changed after twelve monthsAnalysis · 3 min AI playbook for Legal, Protection, IT and Procurement: what to delegate, what to keepGuide · 4 min

Redesigning work at task levelThe task is the unit of AI value. Decomposing it, prompting on real documents, and counting the cost of checking.7 parts

The unit of AI value is a task, never a jobExplainer · 4 min Prompt patterns that survive contact with real documentsGuide · 3 min Task decomposition for AI: finding the handover pointGuide · 4 min The verification tax: what checking costs, and when it exceeds the gainAnalysis · 4 min Before and after: one grant report with AI, honestly measuredGuide · 3 min Context engineering vs prompt engineeringExplainer · 4 min The four checks before anything leaves the buildingGuide · 3 min

Role-by-role patternsResearch, funder intelligence, monitoring and evaluation, board papers, plain language and translation review.6 parts

AI-assisted research a programme officer can defendGuide · 3 min Donor and funder intelligence without a compliance incidentGuide · 4 min Monitoring and evaluation: where AI helps and where it corrupts the dataAnalysis · 4 min AI for board papers: encode the structure, write the judgement freshGuide · 4 min Plain-language rewriting: the highest-value low-risk task in the sectorGuide · 4 min AI translation: delegate the first draft, keep the review with a bilingual humanGuide · 4 min

Writing an AI use policyPermissions first, red lines, role tables, annexes, escalation and the review cycle.6 parts

AI permitted use by role: the four-column table that does the workGuide · 3 min AI policy red lines: the decisions that never get delegatedGuide · 4 min Write your AI policy permissions first, not the prohibitionsExplainer · 4 min Annexes: why casework, fundraising, HR and procurement need their own rulesExplainer · 4 min AI policy escalation path and the 24-hour ruleGuide · 4 min AI policy review: the six-month cycle and what triggers an early oneGuide · 4 min

AI Consulting: leadership briefingsBoard-level pieces on ownership, accountability, value and vendor risk, from the M.A.I. Consulting newsletter.11 parts

Who owns AI here? The operating model a 300-person organisation actually needsAnalysis · 5 min When AI shapes the decision, who signs it? Accountability that survives an auditAnalysis · 5 min Five questions every board should ask about AI this quarterAnalysis · 6 min AI theatre: why leadership announcements change nothing, and what doesAnalysis · 5 min Time saved is not value: turning hours into capacity, outcomes and budget linesAnalysis · 6 min Cost per completed task: the unit economics of AI nobody reportsAnalysis · 7 min Pilot theatre: why AI demos succeed and production failsAnalysis · 6 min Vendor concentration and the exit plan: diversify AI suppliers without doubling costAnalysis · 6 min Advice or sales pitch? Five tests for any AI recommendation, including oursAnalysis · 6 min Before you trust an experimental AI feature in production: the client's checklistAnalysis · 6 min The audit trail for an AI-influenced decision: what to record and for how longAnalysis · 6 min

Data and disclosureBeneficiary data, personal data in prompts, impact assessments, funder disclosure and safeguarding checks.5 parts

Beneficiary data and AI: the question with no comfortable answerAnalysis · 4 min Disclosing AI use to funders: when you must, should, and when it backfiresFramework · 4 min AI safeguarding language checks: build the rulebook once, apply it every timeGuide · 4 min DPIA and AI: when an impact assessment is triggeredFramework · 4 min What "personal data" means once a prompt leaves the buildingExplainer · 5 min

Building the evidence fileThe five artefacts that demonstrate compliance, and what an auditor, a board and a funder each ask to see.4 parts

The five artefacts of an AI governance evidence fileGuide · 4 min Documentation is not certification, and saying so is an advantageAnalysis · 4 min Re-scoring AI readiness: theatre if nobody acted on the first score?Analysis · 4 min What an auditor, a board and a funder each actually ask forAnalysis · 4 min

Standards you can use without certifyingISO/IEC 42001, the NIST AI RMF, and OECD and UNESCO principles, applied at the scale of a small organisation.3 parts

ISO/IEC 42001 for organisations that will never certifyBriefing · 4 min NIST AI RMF applied to a 60-person NGOBriefing · 4 min OECD and UNESCO principles: what they actually oblige you to doBriefing · 4 min

Deciding what to automateAgent, skill or tool; the tasks we will not configure; and what makes an agent survive its author.4 parts

Agent, skill or tool: what to build and what each really costsAnalysis · 4 min The tasks we will never automate with an agent, and why the list is short but absoluteBriefing · 3 min Write the method before you buy the agentAnalysis · 4 min Named owner, operating guide, test set: what makes an agent survive its authorGuide · 4 min

Agent securityPrompt injection, document-store attack surfaces, blast radius, browser agents and human oversight that adds value.5 parts

Indirect injection: your own document store is an attack surfaceExplainer · 4 min Prompt injection explained for executives and DPOs who will never write a filterExplainer · 4 min Blast radius: designing what an automation can reachFramework · 4 min Browser agents in a logged-in session: should you allow them at all?Briefing · 4 min Human-in-the-loop, and where the human actually adds valueFramework · 4 min

AI fluency foundationsPrompting, the 4Ds, delegation and discernment: the habits that make everything else work.3 parts

Prompt in short, checked steps: one ask, a plain start, specific feedbackGuide · 5 min Calibrating trust in AI: diagnose the limit, match scrutiny to the stakesFramework · 5 min Checking AI output before it ships: five habits that cost minutes, not afternoonsGuide · 5 min

Working with AI day to dayWorkflows, Microsoft 365 Copilot and choosing the right tool for the task.9 parts

Choosing an AI tool: start with where the work lives, not a rankingFramework · 5 min Why one paid AI assistant used daily beats hopping between free tiersGuide · 5 min Copilot is a habit problem: the specific-ask routine to teach your team firstBriefing · 5 min Copilot in Outlook, Teams, Word, Excel and PowerPoint, and when to reach for ClaudeGuide · 4 min Why most AI work tasks go slowly, and a four-step method that fixes itFramework · 5 min Decision-ready documents from raw material: briefs, memos, decks and dashboardsGuide · 5 min Recurring reports and updates for many audiences: write once, adapt with AIGuide · 4 min Before and after the meeting: AI workflows for prep, follow-up and action plansGuide · 4 min Three repeatable AI workflows for small teams: grants, donor updates, meeting recapsGuide · 4 min

Leading AI adoptionLeading a team through adoption, and AI for mission-driven organisations.4 parts

Mission first, then one real task: how small nonprofit teams make AI stickAnalysis · 5 min Making AI adoption stick: real work, short sessions, two conversationsGuide · 5 min Leading an AI-enabled team: teach checking first, judge decisions, measure moreGuide · 5 min Ground rules for team AI use: what not to paste, what to say, what to call thingsFramework · 4 min

If you are considering a package, read these.

Package	Start here	Then	If you have time
Foundation	The AI Readiness Check: the free self-check, distinct from the assessment the package runs	The sample playbook: the deliverable you receive	Compare the market: whether a free programme covers you first
Adoption	The sample playbook: four of these, one per department	The extended assessment: the instrument behind the baseline and the re-score	The value case: the per-seat arithmetic against a catalogue licence
Institution	EU AI Act Article 4: the obligation the evidence file answers	The extended assessment: its standards mapping	Claude in Chrome: how we assess a specific tool's risk

Assess the thinking before you commission the work.

Find the piece that answers your question.

AI Readiness Assessment

Extended AI Readiness Assessment

Compare the market

EU AI Act Article 4

Claude in Chrome: impact, trade-offs and warnings

A team playbook

AI use cases in practice

Claude for business

If you are considering a package, read these.

Ready to discuss it?