AI Adoption Strategy for PRMA Consulting

Session zk6mw0 · Duration: 15.2 min · Board: Idea Generator, Reality Checker, Market Scanner, First Principles, Wild Card

Top 5 Ranked Ideas

#1

The AI Error & Correction Log (Compound Knowledge Engine)

A simple shared document capturing every human correction to AI output — "AI wrote X, I changed it to Y, because Z" — that simultaneously builds a prompt library, quality checklist, EU AI Act governance trail, and competitive moat.

#2

The Task Decomposition Workshop (90-Minute Pre-Sprint Foundation)

A structured 90-minute workshop where partners decompose each task into "judgment verbs" vs. "execution verbs," surfacing tacit quality criteria that become both the first AI prompts and the seed of the compound knowledge engine.

#3

The Two-Job Failure Pattern (Universal Workflow Design Principle)

Every AI failure in PRMA consulting traces to the same root cause — asking AI to perform two cognitively distinct jobs in one prompt. The fix: decompose every task into an information-processing layer and a judgment layer, with a human gate at the seam.

#4

The 4-Week Adoption Sprint (Sequenced by Feedback Speed)

Four experiments in four weeks, sequenced not by strategic importance but by feedback speed — building AI intuition on zero-stakes material before touching anything consequential.

#5

The CDP Wall as a 1-Week Legal Question

The confidentiality barrier blocking AI on CDP tasks may not legally exist — most client NDAs predate AI and are silent on it. One partner, three contracts, and a highlighter could unlock the highest-value tasks in weeks.

Prioritized AI Use Case Map

PRMA AI Adoption — Use Case Map & Exploration Path Ranked by: Time Savings × Implementation Ease × Confidentiality Risk (inverse) Phase 1: This Week AI-Ready — Start immediately Phase 2: Weeks 2-4 AI-Assisted — Sprint experiments Phase 3: Month 2+ Infrastructure-Dependent — NDA review unlocks AI-READY Meeting Summaries MacWhisper → Claude extraction → human review Pre-condition: validate transcription pipeline Highest priority — zero blockers AI-READY Landscape Research Claude structured overview → source-verify 3 claims Public data only — zero confidentiality risk Highest priority — zero blockers AI-ASSISTED GVD Outline Human brief → AI scaffold → human rebuilds strategic sequencing for target payer AI-ASSISTED GVD Section Writing AI drafts prose → human selects & verifies EVERY evidence claim explicitly (not passive review) AI-ASSISTED HTA Review & Analysis AI extracts/summarizes → human interprets payer implications & strategic meaning AI-ASSISTED Presentation Slides 2-STAGE: AI argument skeleton → human gate → template/designer for visuals (never combine) INFRASTRUCTURE-DEP CDP Gap Analysis NOW: de-identified description → AI flags gaps AFTER NDA REVIEW: full CDP via enterprise tool Blocked by: NDA review (1 week) + tool decision INFRASTRUCTURE-DEP Evidence Generation Strategy AI generates option menu → human selects AI stress-tests; human leads strategic decisions Blocked by: infrastructure deployment builds confidence NDA unlocks Foundation Layer (Runs in Parallel) AI Error & Correction Log Task Decomposition Workshop NDA / Contract Review Billing Model Discovery AI-Ready AI-Assisted Infrastructure-Dependent Foundation (parallel)

Discovery Questions (15 Questions, 6 Themes)

Workflow & Pain Points

Q1. "Walk me through the last GVD section you wrote from scratch — what were you actually doing for the first two hours?"

Reveals: where time actually goes — searching, structuring, finding analogues, or writing.

Workflow & Pain Points

Q2. "On a typical GVD project — how many hours does the team spend on work that feels like assembly versus work that feels like judgment?"

Reveals: the actual AI-addressable percentage of project hours. If assembly >40%, the business case is self-funding.

Workflow & Pain Points

Q3. "When was the last time a client asked you to move faster than you could — and what did you lose because of it?"

Reveals: cost of inaction. One lost project makes this a survival conversation.

Confidentiality & Data

Q4. "In your standard client agreements — does anything specifically address AI processing of client data, or are the confidentiality clauses silent on that?"

Reveals: whether the CDP wall legally exists. 40% chance it doesn't — NDAs predate AI.

Confidentiality & Data

Q5. "When you open a CDP today, what systems does that data live in? Do you use Microsoft 365?"

Reveals: the fastest infrastructure path. If M365 → Copilot in 2-4 weeks.

Quality & Failure Modes

Q6. "When AI output is wrong in your domain — what does 'wrong' look like specifically?"

Reveals: the specific failure mode to design review gates around.

Quality & Failure Modes

Q7. "When you review a junior's GVD draft — what percentage of your comments are about form versus content?"

Reveals: if mostly form → AI owns 70-80% of drafting. Single biggest ROI indicator.

Quality & Failure Modes

Q8. "When you tried AI for slides and it was terrible — what specifically was bad?"

Reveals: which sub-task failed. Argument failure = capability problem. Visual failure = tool mismatch (fixable).

Team Readiness

Q9. "Is there anyone whose reaction to AI adoption you're privately worried about?"

Reveals: the phantom veto. In small firms, the real block is often social, not technical.

Team Readiness

Q10. "Is there a version of AI-augmented work that you'd be uncomfortable with — not because of confidentiality, but because it changes what the job feels like?"

Reveals: the professional identity ceiling — if reviewing AI output feels like a demotion from authorship.

Team Readiness

Q11. "If I told you 2 of 4 experiments would produce embarrassing results — would that feel like failure or learning?"

Reveals: experimental tolerance. "Failure" = reframe as research. "Learning" = green light.

Strategic Impact

Q12. "Last project you finished faster than expected — did you keep the full fee, or adjust?"

Reveals: billing psychology. Kept fee = AI is pure upside. Discounted = time-billing trap to address first.

Strategic Impact

Q13. "Three years from now, if this firm is known for something — what? And does AI help or hurt that story?"

Reveals: whether AI is central to positioning or backstage infrastructure.

Strategic Impact

Q14. "What would have to be true in 3 months for you to feel this worked — and what would feel like it damaged something you care about?"

Reveals: the success definition AND the real protection boundary.

Growth & Junior Dev

Q15. "How do you train someone to know what AI got wrong — if they don't yet know what right looks like?"

Reveals: whether AI adoption is compatible with the firm's growth model.

Idea Clusters & Themes

Cluster 1: The Compound Knowledge Engine

Three ideas converge into a single artifact:

A Google Sheet started on Monday that captures "AI wrote X, I changed it to Y, because Z" simultaneously produces operational learning, regulatory compliance, and competitive moat.

Cluster 2: The Two-Job Decomposition Framework

Every task is actually two tasks masquerading as one:

Universal workflow: AI (processing) → Human gate → AI (formatting) → Human review

Cluster 3: The Adoption Infrastructure

Cluster 4: Adoption Anti-Patterns (Reverse Brainstorm)

Confidentiality Architecture Options

TierSolutionTimelineCostData Guarantee
0De-identification / anonymizationThis week€0CDP data never enters AI
1aClaude Teams / ChatGPT Enterprise1-2 weeks€25-40/user/moZero data retention, DPA
1bMicrosoft 365 Copilot (if on M365)2-4 weeks€30/user/moEU tenant, existing DPA
2Azure OpenAI private deployment4-8 weeks€200-800/moPrivate tenant, EU region
3Local model (Ollama + Llama 3.3 70B)2-3 months€3,500-5,000 one-timeAir-gapped, absolute

The 4-Week Sprint

WeekTaskToolMeasureDecision Gate
1Meeting summaryMacWhisper + ClaudeSummary matches expert notes? Edit time vs. write-from-scratch timeIf editing >80% of writing time → prompt redesign
2Landscape research (known area)ClaudeHallucination rate, coverage, what was missedIf fact-check every claim → first-draft accelerator only
3GVD section skeletonClaude + templateStructure quality, time to publishableIf structure consistently sound → strong scaffolding tool
4HTA review summaryClaudeQuality vs. expert blind comparisonGap reveals where human expertise is irreplaceable

Workshop Structure (90 Minutes)

PhaseDurationActivity
120 minVerb extraction on a neutral example — judgment verbs vs. execution verbs
235 minIndependent decomposition of 2-3 tasks each, then compare
320 minWrite one prompt per partner using quality criteria as the specification
415 minLive experiment in the room — test the prompt, react, iterate

Board Member Highlights

MemberKey BreakthroughRound
Idea GeneratorThe compound knowledge engine concept — every AI interaction produces training data for a firm-specific intelligence layerR1, R2
Reality CheckerThe Week 1 kill shot analysis (transcription failure masquerading as AI failure, "competent but subtly wrong" summary destroying trust)R3
Market Scanner"Immediate vs. delayed quality signal" as the real classification system. EU AI Act governance trail as competitive moatR1, R3
First PrinciplesThe slide failure contains the general theory of why AI adoption fails in knowledge work — undifferentiated task blobs judged on the hardest sub-taskR2, R4
Wild Card"What do clients pay for: time, output, or judgment?" — the billing model question that restructures the entire ROIR4, R5

Recommended Next Steps — This Week

  1. Monday: Create the AI Error & Correction Log (Google Sheet, 5 minutes). Columns: Date | Task | Prompt | AI Output | Human Correction | Reason | Score 1-5.
  2. Tuesday-Wednesday: Krzysztof reviews 3 client contracts for AI/data processing language (2 hours). This unlocks or confirms the CDP infrastructure path.
  3. Thursday: Schedule the 90-minute Task Decomposition Workshop for next week. Both partners block the time.
The idea to explore first: The AI Error & Correction Log. Not because it's the flashiest — because it's the one that makes everything else compound. Every experiment feeds it. Every correction improves the next prompt. Every entry is simultaneously learning, governance, and moat. Start it before anything else. The clock only starts when you start capturing.