22 AI Skills That Rebuild a $1.5M Transformation Methodology
What the big consulting firms refined over 15 years and bill at seven figures, a frontier AI model can now reorganize into 22 executable skills across 5 layers: diagnostic, discovery, design, deployment and operating.
Companion guide to the AI Strategy Playbook. The playbook covers discovery in depth, with the detailed prompts. This guide covers the full lifecycle, all the way to ongoing operations.
Why this guide exists
The AI transformation methodologies of the big consulting firms represent years of refinement and engagements of $1.5M and up. Yet their analytical skeleton (the frameworks, the matrices, the checklists, the math) can now be rebuilt by a frontier model in minutes.
This guide documents exactly that skeleton: 22 skills, each with its inputs, its deliverables and the precise dividing line between what AI produces and what only an operator can carry.
Because the fundamental distinction stands: analytical scaffolding has become a commodity. Judgment, political navigation and production execution have not, and never will. The recipe is free. The restaurant still matters.
The 5-layer stack
The 22 skills form an operating stack. Each layer answers a different question, and they run in order.
- Diagnostic (6 skills). Where the organization really stands. Typical duration: 2 to 3 weeks.
- Discovery (4 skills). What is really happening on the ground, operationally and politically. Typical duration: 1 month.
- Design (5 skills). How the future state will work: roles, governance, decisions, escalations, vendors.
- Deployment (4 skills). How the solution reaches production without dying along the way. Design and deployment: about one quarter.
- Operating (3 skills). How the system lives after go-live. Duration: ongoing, this is the phase that never ends.
How to use this guide
Every skill follows the same format: its purpose, what to feed the model, what you get back, then the dividing line between AI and the operator.
Eight skills from the diagnostic and design layers exist in a detailed version, with a full prompt, in the AI Strategy Playbook: we link there rather than duplicate. The other 14 skills include their copy-paste prompt here.
Use a frontier model with a large context window, and treat every output as a draft to validate in the field, never as truth.
Layer 1: Diagnostic
6 skills · 2 to 3 weeks
This layer's question: where the organization really stands. Not where leadership believes it stands.
Skill 1: AI Readiness Assessment
Determine whether the organization has the capacity to carry the weight of an enterprise AI deployment: data, security, structure, foundations.
What to feed the model
- Org chart
- Technical architecture documentation
- IT and security audits
- Data strategy overview
What you get back
- Gap analysis
- Foundational blockers
- Executive summary ranked by severity
The operator owns: breaking the political and governance silos the analysis reveals.
Detailed prompt: module 1 of the AI Strategy Playbook
Skill 2: Workflow Bottleneck Analysis
Identify the inefficiencies of manual work and the real automation candidates, by comparing official procedures with reality on the ground.
What to feed the model
- Official procedures (SOPs)
- Frontline interview transcripts
- Cycle-time notes
What you get back
- Gaps between procedure and reality
- Shadow IT map
- Automation candidates, scored for feasibility
The operator owns: understanding the political and historical context that created each bottleneck.
Detailed prompt: module 2 of the AI Strategy Playbook
Skill 3: Use-Case Prioritization
Sort every AI idea into four actionable categories: sandbox, distractions, incremental gains and flagship pilots.
What to feed the model
- Raw use-case list
- Bottleneck analysis (skill 2)
- Strategic objectives
What you get back
- Four-quadrant framework
- Flagship pilots backed by metrics
The operator owns: gauging political bandwidth and validating leadership air cover.
Detailed prompt: module 3 of the AI Strategy Playbook
Skill 4: Build vs Buy
Decide whether to buy a platform or build the workflow logic in-house, without falling into the commodity trap.
What to feed the model
- Selected flagship pilot
- In-house engineering capacity
- Vendor pricing and features
What you get back
- Commodity-versus-proprietary-advantage analysis
- 3-year total cost of ownership
- Verdict: build, buy or hybrid
The operator owns: honestly calibrating the engineering team's real capacity.
Detailed prompt: module 4 of the AI Strategy Playbook
Skill 5: ROI Business Case
Build a business case that holds up in front of the CFO: hard savings, full costs, the change-management tax and the payback period.
What to feed the model
- Baseline costs of the manual workflow
- Estimated project costs
- Operating parameters
What you get back
- Conservative ROI model with a 30% tax
- Payback period
- Answers to the CFO's objections
The operator owns: validating the baseline numbers and defending the case in the boardroom.
Detailed prompt: module 5 of the AI Strategy Playbook
Skill 6: Competitive Landscape Scan
Assess what competitors have actually shipped to production, beyond their press releases.
Act as a senior competitive intelligence analyst.
From the attached competitor list, earnings call transcripts,
press releases, job postings and analyst reports, produce:
1. Shipped versus announced: what each competitor has actually
put into production, as opposed to what it has announced
2. The gap analysis against our current position
3. Two recommended competitive moves, with rationale What to feed the model
- Competitor list
- Earnings transcripts and press releases
- Job postings and analyst reports
What you get back
- Shipped-versus-announced comparison
- Gap analysis
- Two recommended moves
The operator owns: telling real product from PR narrative.
Layer 2: Discovery
4 skills · about 1 month
This layer's question: what is actually happening, operationally and politically. This is where the official org chart stops being useful.
Skill 7: Pilot Scoping
Define a tight, bounded pilot scope before the VPs' "little additions" turn it into a two-year program.
Act as a disciplined program director.
From the flagship use case, the team's capacity, the timeline
and the success criteria leadership expects, produce a one-page
scoping document:
1. In scope / out of scope: explicit lines, no grey zone
2. The time box: a firm pilot duration
3. The stop criteria: the exact conditions that end the pilot,
no matter the enthusiasm What to feed the model
- Flagship use case
- Team capacity and timeline
- Leadership's success criteria
What you get back
- One-page scoping document
- Firm time box
- Explicit stop criteria
The operator owns: defending the scope against every "little addition" requested in meetings.
Skill 8: Stakeholder Mapping
Map the political reality of the project, not the fiction of the org chart.
Act as an organizational strategist.
From the attached org chart, pilot scope, meeting notes and
budget context, produce:
1. A political RACI that reflects the reality of power, not
the official org chart
2. Identification, with rationale: the real sponsor, the
declared blockers, the silent killers, the swing votes
and the operational allies What to feed the model
- Org chart and pilot scope
- Meeting notes
- Budget and reporting-line context
What you get back
- Political RACI
- Sponsor, blockers, swing votes and allies identified
The operator owns: spotting the silent killers, the leaders who never block openly but quietly defund.
Skill 9: Current-State Architecture Audit
Identify integration risks and data gaps before the production deployment, not after.
What to feed the model
- Architecture documents and schemas
- API inventory and sample payloads
- Latency profiles and data requirements
What you get back
- Missing data vectors
- Structural deficits and latency risks
- Prioritized remediation plan
The operator owns: negotiating access, unblocking firewalls and fixing the plumbing.
Detailed prompt: module 7 of the AI Strategy Playbook
Skill 10: Change-Readiness Audit
Assess whether the organization is ready to absorb the transformation, or whether the pilot will die from resistance nobody saw coming.
Act as a change management specialist.
From the pilot scope, the stakeholder map, the engagement
surveys and the post-mortems of past change initiatives,
produce:
1. The leadership air-cover assessment: real support, not
stated support
2. The middle-management fear analysis: where and why
3. The frontline adoption signal: who will get on board,
who will resist
4. The three priority interventions to raise the odds
of success What to feed the model
- Scope and stakeholder map
- Engagement surveys
- Post-mortems of past initiatives
What you get back
- Leadership air-cover assessment
- Resistance analysis by level
- Three priority interventions
The operator owns: reading resistance in the room and holding the hallway conversations with middle managers.
Layer 3: Design
5 skills · within the design-and-deployment quarter
This layer's question: how the future state will work. It is the step rushed organizations skip, and it is exactly why their pilots die.
Skill 11: Target Operating Model
Define the human-AI collaboration structure: who does what, where the handoffs sit, where the human stays in the loop.
What to feed the model
- Current workflow
- Existing roles
- Planned AI agents
What you get back
- Role evolution map
- Recommended new roles
- Handoff structure and escalation protocol
The operator owns: navigating ego, turf and managerial fear, then getting leadership sign-off.
Detailed prompt: module 6 of the AI Strategy Playbook
Skill 12: AI Governance Framework
Establish the governance that makes the deployment compliant, auditable and defensible, including Law 25 in Quebec.
What to feed the model
- Proposed workflow
- System architecture
- Data classifications
What you get back
- Risk-tiered framework
- Logging requirements per tier
- Model approval workflow and ethics review triggers
The operator owns: defending the policy in front of the board and carrying the regulatory accountability.
Detailed prompt: module 8 of the AI Strategy Playbook
Skill 13: Decision Rights Map
Eliminate committee paralysis by assigning explicit decision authority, decision by decision.
Act as an operational governance designer.
From the target operating model, the org chart and the
proposed workflow, produce:
1. The list of every decision point in the workflow
2. For each decision: a single named owner, not a committee
3. An SLA per decision: the maximum response time
4. The fallback authority if the owner does not decide
within the deadline What to feed the model
- Target operating model
- Org chart
- Proposed workflow
What you get back
- Decision-by-decision accountability map
- SLA per decision
- Fallback authority
The operator owns: confirming one-on-one that each executive will actually decide, and preventing the quiet dilution of authority.
Skill 14: Escalation Paths
Define precisely when and how AI hands off to a human, in three tiers with triggers and deadlines.
Act as an operations process designer.
From the target operating model and the human-in-the-loop
protocol, produce a three-tier escalation structure:
1. For each tier: the exact triggers that escalate a case
2. The receiver of each tier: role and person
3. The pickup SLA per tier
4. The documentation requirements at every escalation What to feed the model
- Target operating model
- Human-in-the-loop protocol
What you get back
- Three-tier escalation
- Triggers, receivers and SLAs
- Documentation requirements
The operator owns: verifying that the tier 2 and tier 3 receivers can actually absorb the volume.
Skill 15: Vendor Selection Matrix
Choose the vendor or platform that minimizes lock-in and integration burden, not the one with the best salesperson.
Act as a neutral technology procurement advisor.
From the build-or-buy decision, the target operating model
and the vendor capability and pricing data, produce:
1. A comparison matrix: capabilities, lock-in risk,
integration burden, 3-year total cost
2. Flags on risky contract clauses
3. A firm recommendation, with rationale What to feed the model
- Build-or-buy decision
- Target operating model
- Vendor capabilities and pricing
What you get back
- Complete vendor matrix
- Risky clauses flagged
- Justified recommendation
The operator owns: the reference calls, where you tell nuanced enthusiasm from polite frustration.
Layer 4: Deployment
4 skills · within the design-and-deployment quarter
This layer's question: how the solution reaches production. This is where 95% of pilots die, almost always for reasons that have nothing to do with technology.
Skill 16: Sequencing Plan
Determine the optimal order of pilots to maximize the odds of political and technical success.
Act as a transformation portfolio director.
From the flagship pilots, the stakeholder map, the
change-readiness audit and the architecture audit, produce:
1. A three-pilot sequence, with the rationale for the order
2. The political dependencies: who must succeed before whom
3. The technical dependencies: which foundations each pilot
requires
4. The roadmap dominoes: what each success unlocks next What to feed the model
- Prioritized flagship pilots
- Stakeholder map
- Change and architecture audits
What you get back
- Justified three-pilot sequence
- Political and technical dependencies
- Roadmap domino effect
The operator owns: judging whether the first pilot will survive the budget battles ahead.
Skill 17: Pilot Charter
Draft the pilot's constitution: a signed document that bounds the scope, the success criteria and the budget.
Act as a program director.
From the pilot scoping and the operating model, draft a
formal charter in six sections:
1. Scope
2. Measurable success criteria
3. Stop criteria
4. Budget guardrail
5. Team and capacity commitment
6. Reporting cadence What to feed the model
- Pilot scoping
- Operating model
What you get back
- Formal six-section charter
- Measurable success and stop criteria
The operator owns: getting the sponsor's signature. An unsigned charter is not a commitment, it is a wish list.
Skill 18: Production Gating
Define the explicit checklist that keeps a pilot from moving to production prematurely.
Act as a production reliability lead.
From the pilot charter, the governance framework, the
architecture audit and the operating model, produce a
production gating checklist across five dimensions:
1. Performance
2. Security and governance
3. Operational readiness
4. Adoption readiness
5. Financial controls
For each dimension: the exact pass criteria and the artifact
that proves them. What to feed the model
- Pilot charter
- Governance framework
- Architecture audit and operating model
What you get back
- Five-dimension checklist
- Pass criteria per dimension
- Required proof artifacts
The operator owns: enforcing the gates despite the sponsor's impatience.
Skill 19: Rollback Planning
Establish the recovery procedures before the production incident, not during.
Act as a site reliability engineer.
From the architecture, the workflow and the governance
framework, produce a rollback plan:
1. The catalogue of plausible failure modes
2. The exact trigger thresholds that activate the rollback
3. The withdrawal sequence: step by step, who does what
4. The communication plan: who to inform, in what order,
with what message What to feed the model
- Architecture and workflow
- Governance framework
What you get back
- Failure-mode catalogue
- Thresholds and withdrawal sequence
- Communication plan
The operator owns: rehearsing the scenario until the team has the reflex, before the real outage.
Layer 5: Operating
3 skills · ongoing
This layer's question: is the system still alive. It is the phase that never ends, and the one where programs die quietly during year two.
Skill 20: Adoption Tracking
Measure whether the deployed AI is actually used in daily work, or whether it is dying a slow death.
Act as a product adoption analyst.
From the target-user definitions, the usage telemetry and
the engagement survey data, produce an adoption dashboard:
1. The usage signal: who uses what, how often
2. The drop-off cohorts: who stopped using it, and when
3. The behaviour-change index: is the work being done
differently, or is the AI one more step
4. The intervention candidates: where to act first What to feed the model
- Target-user definitions
- Usage telemetry
- Engagement surveys
What you get back
- Adoption dashboard
- Drop-off cohorts
- Intervention candidates
The operator owns: understanding why a cohort dropped off, like that private VP directive to deprioritize the tool.
Skill 21: ROI Attribution
Credibly answer the CFO's question: "what exactly changed because of AI?"
Act as a rigorous financial controller.
From the baseline metrics, the current metrics, the original
business case and the context on confounding factors, produce:
1. The hard savings achieved, with the math
2. The soft gains achieved, separated and labelled as such
3. The attribution confidence: for each metric, the degree of
certainty that AI caused it, with the reasoning
4. The gap analysis: projection versus outcome, no sugar-coating What to feed the model
- Baseline and current metrics
- Original business case
- Known confounding factors
What you get back
- Savings achieved, hard and soft
- Attribution confidence per metric
- Projection-versus-outcome gap
The operator owns: defending the attribution in front of the CFO and securing year-two funding.
Skill 22: Model Retraining Cadence
Prevent model drift by establishing a continuous retraining discipline, before quality degrades in silence.
Act as a senior MLOps engineer.
From the model specifications, the input data profiles over
time, the performance metrics and the workflow change log,
produce a retraining cadence plan:
1. The drift detection signals and their thresholds
2. The retraining budget: frequency and cost
3. The model version control and rollback plan
4. The accountability: who is responsible for making sure
retraining actually happens What to feed the model
- Model specifications
- Data profiles and metrics over time
- Workflow change log
What you get back
- Drift signals and thresholds
- Retraining budget and cadence
- Versioning and accountability plan
The operator owns: making sure retraining actually happens. Silent year-two drift is the number one killer of programs.
What AI rebuilt, what stays human
A frontier model reorganized in minutes what took years of refinement. But be precise about what it rebuilt, and what it will never touch.
AI rebuilt
- The analytical scaffolding
- The structured frameworks and diagrams
- The synthesis of large document volumes
- The math and scenario modelling
- The checklists and artifact mapping
AI does not touch
- Spotting and neutralizing the silent killers
- The CFO and leadership relationship, and the persuasion
- The middle-management fear loops
- The board narrative through CEO transitions
- The risk judgment at production deployment
- The capacity-versus-volume calibration in escalations
- Reading reference calls and institutional truths
The real cost of going solo
Running the 22 skills yourself takes 200 to 400 hours of a focused operator over 4 to 6 months, per business unit. That is not prompt time: it is time spent collecting inputs, validating in the field, aligning stakeholders and defending deliverables.
And roughly 95% of enterprise AI deployments do not die from a lack of methodology. They die because execution was never the link that got covered. The skills give you the complete map. Someone still has to drive.
FAQ: The 22 skills of AI transformation
How is this different from the AI Strategy Playbook?
The playbook covers the discovery phase in depth: 10 modules with detailed prompts. This guide covers the full lifecycle in 5 layers, from diagnostic through ongoing operations. Use both together: this guide for the big picture and the sequence, the playbook to dig into the discovery modules.
Which AI model should I use to run these skills?
A frontier model with a large context window (Claude, GPT). The model is not what matters: input quality and field validation of the outputs are.
How long does the full cycle take?
Going solo: 200 to 400 hours over 4 to 6 months per business unit. Diagnostic in 2 to 3 weeks, discovery in a month, design and deployment in a quarter, then ongoing operations.
My internal documents are sensitive. Can I submit them to an AI model?
Use an enterprise offering (no training on your data), anonymize personal information and meet your obligations: in Quebec, Law 25 governs the disclosure of personal information. When in doubt, redact: the prompts work very well with anonymized documents.
Does this really replace a consulting firm?
It replaces the analytical scaffolding: the frameworks, the matrices, the math and the documents. It does not replace judgment, political navigation or getting to production. That is exactly the dividing line every skill in this guide makes explicit.
Two ways to move forward
Reply within one business day. No commitment.