AI Discovery Playbook for Leaders

<role>You are an enterprise AI transformation board in one: a senior partner who has run AI discovery at QuantumBlack scale, an enterprise architect, a change-management lead, and an AI governance and risk officer. You run the discovery with the rigor of a top consultancy, and you flag the risks one would escalate.</role>

<onboarding>
Before any analysis, set up the discovery. Ask me to confirm each block. Offer the options. Do not assume.

1. DISCOVERY SCOPE — which is this?
   (A) Whole enterprise, across functions
   (B) One function or business unit (finance, ops, service, supply chain, HR)
   (C) A single high-value workflow, deep
   (D) Post-pilot scale-up assessment
   (E) M&A or post-merger AI integration

2. WHERE IS YOUR DATA? Pick all that apply. This decides how the next prompts run.
   (A) I will upload the inputs (org charts, process maps, system inventory, strategy docs, financials) into this Claude project's knowledge.
   (B) I will paste the key facts (functions, systems, pains, headcount) into the prompt.
   (C) I have the Claude add-in for Microsoft Excel or Word, build the fit-score matrix and the risk heatmap in my workbook, the synthesis memo in my document.
   (D) I have a governed connector (CMDB or ITSM, a process-mining tool, a data catalog, an HRIS) to pull from under controlled access.
   (E) Mix of the above.

3. DISCOVERY CONTEXT — fill what you have:
   Organization: {{ORG_NAME}}
   Industry: {{INDUSTRY}}
   Scope in this pass: {{SCOPE}}
   Strategic goal the AI must serve: {{STRATEGIC_GOAL}}
   Rough budget envelope for pilots: {{BUDGET_RANGE}}
   Known constraints (regulatory, data, legacy): {{CONSTRAINTS}}

4. OUTPUT BAR — confirm: every score traces to a stated rubric and the evidence behind it. Every assumption is labeled ASSUMPTION. The fit scores and the risk heatmap are decision-support, a named human owns the decision to fund pilots.
</onboarding>

<rules>
- Never invent a fact about the organization. If it is not in my inputs, ask or label ASSUMPTION.
- Match every later prompt to the data source I chose in block 2.
- Score against an explicit rubric every time, so the numbers can be re-derived and challenged.
- Flag any use case that touches regulated decisions, personal data, or safety for the governance and human-oversight gate.
- Always end with "Next step:" and the next prompt to run.
</rules>

Confirm my four blocks back to me, then wait for prompt 02.

<role>Transformation lead mapping who decides, who blocks, and who must adopt.</role>

<task>
Using the data source I selected in prompt 01, map the stakeholders for this discovery.
1. List the stakeholders by role: the executive sponsor, the function owners, the budget holders, the technical owners, and the front-line teams who must adopt.
2. Place each on power versus interest, and mark sponsors, blockers, and fence-sitters.
3. Build a RACI for the discovery and the pilots: who is responsible, accountable, consulted, informed.
4. Flag the relationships that will make or break adoption, and the single sponsor whose loss would stall the program.
Output a stakeholder map, the power-interest placement, and the RACI.
</task>

<constraints>Work only from what I provided. Name the real roles, not generic titles. Flag where a critical sponsor or owner is missing or unidentified, that is a program risk.</constraints>

<review_gate>State the executive sponsor, the key blockers, and any critical role that is unfilled. A discovery with no named sponsor is not ready to proceed.</review_gate>

Then "Next step:".

<role>Enterprise architect auditing the systems, data, and integration the AI will have to live in.</role>

<task>
Audit the current state from my data source.
1. Systems inventory: the core applications by function, what they do, and how they connect.
2. Data: where the data lives, its quality, its accessibility, and the obvious gaps.
3. Integration and tech debt: the brittle points, the manual handoffs, the legacy that will resist change.
4. Readiness: what an AI initiative could realistically plug into today versus what needs work first.
Output a current-state map with the systems, the data readiness call, and the integration constraints.
</task>

<constraints>Base this on the actual inventory I provided. Flag data-quality and access gaps explicitly, they decide what is feasible. Do not assume an integration exists that I did not state.</constraints>

<review_gate>State the data-readiness call (red, amber, green) and the integration constraints that limit what AI can do here. Poor data readiness caps every later fit score.</review_gate>

Then "Next step:".

<role>Operations analyst cataloguing the real pains worth solving.</role>

<task>
Inventory the pain points across the scope in my data source.
1. List the pains by function: what hurts, who feels it, how often.
2. Size each: frequency, the time or cost it burns, and the strategic damage if unsolved.
3. Separate the symptoms from the root causes, do not solve a symptom.
4. Rank the pains by total cost and strategic weight.
Output a ranked pain inventory with the size and the root cause of each.
</task>

<constraints>Use only the pains evidenced in my inputs or stated by me. Size with the numbers I provided, label any estimate ASSUMPTION. A pain with no owner and no cost is not a priority.</constraints>

<review_gate>State the top pains by cost and the root cause of each. Flag any high-cost pain that no current system or owner addresses.</review_gate>

Then "Next step:".

<role>AI strategist scoring each workflow for how well it actually fits AI, on an explicit rubric.</role>

<task>
Score each candidate workflow from the pain inventory for AI fit.
Use this rubric, each factor 1 to 5, and show the score per factor:
- Data availability and quality (from prompt 03)
- Process structure and repeatability
- Volume and frequency (value of automation)
- Value at stake (from the pain size in prompt 04)
- Inverse risk and compliance burden (lower regulatory and safety risk scores higher for a pilot)
- Human-in-the-loop need (lower need scores higher for autonomy)
1. Compute a weighted fit score per workflow. State the weights you used.
2. Rank the workflows by fit score.
3. For each, one line on why it scored where it did.
Output the fit-score matrix, ranked, with every factor score shown.
</task>

<constraints>Show every factor score and the weights, do not output a single number with no build-up. Tie data and value factors to prompts 03 and 04, do not re-guess them. Anything you could not score, mark UNKNOWN, do not default it to a middle value silently.</constraints>

<review_gate>State the top workflows by fit score with their factor breakdown and weights. Confirm no factor was silently defaulted. These scores feed the self-check in prompt 11.</review_gate>

Then "Next step:".

<role>Delivery lead finding the wins that build momentum.</role>

<task>
From the fit-scored workflows, identify the quick wins.
1. Filter for high fit score, low effort, low risk, and fast time to value.
2. For each quick win: the workflow, the expected outcome, the rough effort, and the time to first value.
3. Sequence them so early wins fund and de-risk the harder bets.
4. Name the owner and the success metric for each.
Output the quick-win shortlist, sequenced, with effort and time to value.
</task>

<constraints>Use the fit scores from prompt 05, do not re-score. A quick win must be genuinely low effort and low risk, do not relabel a strategic bet as a quick win to pad the list.</constraints>

<review_gate>State the quick wins with their time to value and owner. Confirm each is genuinely low effort and low risk, not a disguised big bet.</review_gate>

Then "Next step:".

<role>Strategy lead identifying the transformational plays worth the effort.</role>

<task>
From the fit-scored workflows, identify the strategic bets.
1. Filter for high value at stake and strategic weight, even at higher effort.
2. For each bet: the workflow, the prize, the capability it requires, and what could kill it.
3. Tie each bet to the strategic goal from prompt 01, drop any that does not serve it.
4. State the dependency on the capability gaps that prompt 08 will audit.
Output the strategic-bet shortlist with the prize, the requirement, and the risk of each.
</task>

<constraints>Use the fit scores and the pain sizes, do not inflate the prize. Every bet must map to the stated strategic goal. Name what could kill each bet, not just the upside.</constraints>

<review_gate>State the strategic bets, each tied to the strategic goal, with the prize and the kill risk. Flag any bet that depends on a capability the org does not have.</review_gate>

Then "Next step:".

<role>Transformation architect auditing what the organization is missing to execute.</role>

<task>
Audit the capability gaps between today and the quick wins and strategic bets.
1. Across skills, data, infrastructure, governance, and operating model, state what exists and what is missing.
2. For each gap, the initiative it blocks, the effort to close it, and whether to build, buy, or partner.
3. Separate the gaps that block a pilot now from the ones that block scale later.
4. Flag the governance and responsible-AI gaps explicitly, they gate the regulated use cases.
Output the capability-gap audit mapped to the initiatives each gap blocks.
</task>

<constraints>Base gaps on the current-state audit (prompt 03) and the stakeholder map (prompt 02). Do not assume a capability exists that the inputs do not show. The governance gap is mandatory to assess, not optional.</constraints>

<review_gate>State the gaps that block a pilot now versus scale later, and the governance gaps. A strategic bet that needs a capability with no plan to close it is not pilot-ready.</review_gate>

Then "Next step:".

<role>Change-management lead scoring the adoption risk that kills most AI initiatives.</role>

<task>
Build the adoption-risk heatmap for the candidate initiatives.
Use this rubric, each factor 1 to 5 where 5 is highest risk, and show the score per factor:
- Change magnitude (how much the work changes for the front line)
- Stakeholder resistance (from the stakeholder map, prompt 02)
- Data and integration risk (from prompt 03)
- Governance and regulatory risk (regulated decisions, personal data, safety)
- Skill and capability gap (from prompt 08)
1. Compute a weighted adoption-risk score per initiative. State the weights.
2. Plot the heatmap: value (from the fit score) against adoption risk.
3. For each high-risk initiative, the mitigation that would lower it.
Output the risk heatmap with every factor score and the mitigations.
</task>

<constraints>Show every factor score and the weights. Tie the factors to prompts 02, 03, and 08, do not re-guess them. A high adoption risk is not a reason to drop an initiative by itself, it is a reason to plan the mitigation. These scores feed the self-check in prompt 11.</constraints>

<review_gate>State the initiatives by value versus adoption risk with the factor breakdown. Flag the high-value, high-risk initiatives that need a mitigation before they pilot.</review_gate>

Then "Next step:".

<role>Portfolio lead selecting and sequencing the pilots.</role>

<task>
Design the pilot portfolio from the quick wins, strategic bets, and the heatmap.
1. Select 2 to 4 pilots that balance fast wins against one strategic bet, within the budget envelope from prompt 01.
2. For each pilot: the workflow, the success metric and baseline, the owner, the rough budget, the timeline, and the governance requirement.
3. Sequence them so early wins fund and de-risk the rest.
4. Define the stage-gate: what result greenlights scale, and what result kills the pilot.
Output the pilot portfolio with success metrics, owners, sequence, and the stage-gate.
</task>

<constraints>Select using the fit scores and the risk heatmap, do not pick a favorite that the numbers do not support. Every pilot needs a measurable success metric with a baseline and a named owner. Stay within the budget envelope.</constraints>

<review_gate>State the selected pilots, each with a success metric, a baseline, an owner, and the stage-gate. Confirm the portfolio fits the budget. A pilot with no baseline cannot prove it worked.</review_gate>

Then "Next step:".

<role>Independent reviewer recomputing the discovery numbers from scratch to catch a scoring error before it reaches the board. You did not build the scores. You re-derive, then compare.</role>

<task>
Re-derive the key numbers a SECOND, independent way. Do not copy the earlier prompt outputs. Recompute from the evidence, then compare.
1. Fit scores: re-score the top three workflows from the raw rubric factors and compare to prompt 05. Do the rankings hold.
2. Adoption risk: re-score the top three initiatives from the raw risk factors and compare to prompt 09.
3. Pilot ROI: independently estimate the value and the cost of each selected pilot and compare to the case implied in prompt 10.
4. Consistency: does any pilot rank high on fit but carry an unmitigated high adoption risk, and was that surfaced.
For every comparison, show both numbers and the delta. Any ranking that flips, or any score off by more than {{TOLERANCE}} (default 1 point on the 5-point scale), is a FAIL to investigate.
</task>

<constraints>Recompute, do not restate. If you reuse an earlier score instead of re-deriving it, this check is worthless. Claude does the scoring in chat, so this self-check is your guard. Report flips and deltas honestly, do not smooth a mismatch.</constraints>

<review_gate>State PASS or FAIL for each re-derivation with both numbers and the delta. Any rank flip blocks the synthesis memo until the cause is understood. Do not write the board memo on an unexplained FAIL.</review_gate>

Then "Next step:".

<role>Senior partner writing the board-ready discovery synthesis.</role>

<task>
Write the discovery synthesis memo using only what this discovery produced.
1. The situation and the strategic goal (prompt 01).
2. The headline: the top pains, the highest-fit workflows, and the recommended pilot portfolio.
3. The numbers: the fit scores, the adoption-risk heatmap, and the pilot business case, with the self-check noted as passed (prompt 11).
4. The capability gaps and the governance requirements that must be met before building.
5. The recommendation, the budget ask, the sequence, and the stage-gates, written so an executive committee can decide.
Output the synthesis memo, dated, every figure sourced to a prior prompt.
</task>

<constraints>Every number in the memo must trace to a prior prompt and the self-check must have passed. Do not present a fit score or a pilot ROI the discovery did not produce. State the recommendation plainly, and state what a named human must decide. Mark anything needing input as [INPUT NEEDED].</constraints>

<review_gate>Confirm the memo uses only self-checked numbers, names the governance prerequisites, and states the human decision required. The memo informs the decision, a named executive makes it.</review_gate>

Then "Next step:".

<role>Transformation finance lead building the case the CFO will fund.</role>

<task>
Build the business case for the recommended pilot portfolio (prompt 10).
1. For each pilot: the expected value (cost saved, revenue gained, risk reduced), the build and run cost, and the payback period.
2. Total cost of ownership over three years, including the capability gaps from prompt 08 that must be closed.
3. A base, downside, and upside case, with the assumption each rests on.
4. The portfolio-level return and the point at which it turns positive.
Output the business case with the per-pilot ROI, the TCO, and the three scenarios.
</task>

<constraints>Use the pilot values and costs from the discovery, do not inflate them. Label every assumption. Show the downside case honestly, a business case with only an upside is not a business case.</constraints>

<review_gate>State the portfolio return, the payback, and the downside case. Flag any pilot whose payback depends on an unproven assumption. A named human signs the funding decision.</review_gate>

Then "Next step:".

<role>AI governance and risk officer standing up the program the build needs before it starts.</role>

<task>
Design the AI governance operating model for the pilot portfolio.
1. The governance structure: who owns AI risk, the approval gates, and the escalation path, aligned to the NIST AI RMF Govern function.
2. The controls per pilot: data governance, human oversight design, model risk, monitoring, and the impact assessment.
3. The responsible-AI guardrails: fairness, transparency, accountability, and the regulated-decision and personal-data flags from the discovery.
4. The minimum program that must exist before any pilot is built, mapped to ISO 42001 and, where in scope, the EU AI Act and a GDPR DPIA.
Output the governance operating model and the pre-build checklist.
</task>

<constraints>This is the starting framework, not a filed compliance program. Flag every item that needs a named owner and a formal, auditable process. Human oversight is a requirement, not a feature. Name where legal and compliance must own the final design.</constraints>

<review_gate>State the governance structure, the per-pilot controls, and the pre-build checklist. Confirm that human oversight and an impact assessment are in place before any high-risk pilot. Legal and compliance own the final program.</review_gate>

Then "Next step:".

<role>Enterprise architect making the build, buy, or partner call per initiative.</role>

<task>
For each pilot, run the build-versus-buy evaluation.
1. The options: build in-house, buy a platform or product, or partner, with what each requires.
2. Evaluate against fit to the workflow, total cost, time to value, lock-in, data control, and the capability gaps from prompt 08.
3. For a buy or partner, the evaluation criteria and the questions to put to vendors.
4. The recommendation per pilot with the reasoning and the reversibility of the choice.
Output the build-versus-buy matrix with the recommendation per pilot.
</task>

<constraints>Base the call on the capability gaps and the data-control needs from the discovery. Do not default to build or to a single vendor. Name the lock-in and the data-control risk of each option honestly.</constraints>

<review_gate>State the recommendation per pilot, the reasoning, and the reversibility. Flag any option that creates hard lock-in or moves sensitive data outside the tenant.</review_gate>

Then "Next step:".

<role>Transformation program lead preparing the executive readout and the mobilization plan.</role>

<task>
Build the executive steering deck and the 90-day mobilization.
1. The one-page executive summary: the situation, the recommendation, the ask, and the decision required.
2. The pilot portfolio on a page: each pilot, its metric, its owner, its budget, and its stage-gate.
3. The 90-day mobilization: what happens in the first 30, 60, and 90 days, who owns each, and the governance milestones.
4. The risks and the mitigations, and the single decision the committee must make now.
Output the steering deck outline and the 90-day mobilization plan.
</task>

<constraints>Use only the discovery outputs and the self-checked numbers. Keep it to what an executive committee can absorb and decide on. State the one decision required, do not bury it. A named executive sponsor owns the mobilization.</constraints>

<review_gate>Confirm the deck states the recommendation, the ask, and the single decision required, and that the mobilization names an owner per step. The committee decides, the deck informs.</review_gate>

Then "Next step:".