PART 14

AI × Strategy & Leadership

An assistant can collect authorized evidence, draft a brief, and make scenario assumptions explicit. It cannot know a competitor's intent, predict a market with certainty, decide the company's risk appetite, or replace executive and board accountability. Separate observed facts, modelled estimates, interpretations, options, and decisions on every page.

Cited facts · explicit inference · scenario ranges · accountable decisions

Competitive Intelligence & Executive Briefs

Choose a bounded question, authorized source set, decision date, and named analyst. More monitoring is not automatically better intelligence.

BASELINE RISK

Untraceable strategic narrative

Stale reports, unlicensed collection, incomplete internal data, social posts, and model invention blend together → correlation becomes motive → one forecast becomes “the future” → leadership cannot audit why a recommendation was made.

CONTROLLED PILOT

Cited brief with decision boundaries

Start with synthetic briefing fixtures → use first-party, public, or licensed sources under their terms → retain URL or record, timestamp, and excerpt → label fact, estimate, inference, and unknown → present alternatives and disconfirming evidence → analyst verifies before leadership review.

Evidence package

Representative
Decision-relevant briefing set
Cover product, price, policy, financial, supply, customer, and market questions across quiet periods, shocks, missing data, conflicting sources, languages, and time horizons.
Adversarial
Manipulation and uncertainty
Test planted claims, source spoofing, prompt injection, recycled news, paid content, selective evidence, confidential requests, unsupported causality, false precision, and pressure to hide uncertainty.
Regression
Traceability must persist
Re-run a frozen set after model, prompt, retrieval, source allowlist, taxonomy, or connector changes; block release when citations, fact/inference labels, or access controls regress.
Speed, volume of sources, and polished prose do not establish insight quality. The evidence is factual agreement, citation integrity, calibrated uncertainty, useful alternatives, reviewer burden, and decision outcomes in the defined context.

Scenarios, Options & Decisions

Scenarios are conditional models, not predictions. Their value comes from transparent assumptions and decisions that remain useful across plausible futures.

BASELINE RISK

A single point estimate hides demand, price, cost, capacity, regulation, exchange-rate, and competitor uncertainty; a generated recommendation is adopted without ownership or downside analysis.

CONTROLLED PILOT

Define base, upside, downside, and stress assumptions with ranges and owners → reconcile calculations deterministically → expose sensitivities and missing evidence → compare reversible options, constraints, triggers, and exit costs → executives or the board decide within delegated authority.

Pilot decision record

Brief quality
Verified before distribution
Measure factual and citation agreement, unsupported inference, omitted counterevidence, assumption visibility, calculation agreement, timeliness, reviewer correction, and decision usefulness.
Governance
Stop conditions
Stop on confidential disclosure, cross-tenant retrieval, unlicensed source use, fabricated citation, material arithmetic error, concealed uncertainty, external publication, action attempt, or unexplained regression.
Do not attribute a budget or business outcome to the assistant without an appropriate counterfactual and complete record of the human analysis, assumptions, constraints, execution, and external conditions.

Reference architecture: read-only strategy briefing assistant

Use tenant-isolated, least-privilege read identities, allowlisted first-party and licensed sources, source-level authorization, retention limits, and secrets outside model context. Store citations, timestamps, and evidence snapshots; schema-validate facts, estimates, inferences, unknowns, scenarios, and options as distinct fields. Keep email publication, external posting, budget changes, trading, pricing, hiring, contracting, and system writes disabled. Named analysts verify drafts; executives and boards retain decisions. Log access, versions, evidence, edits, approvals, and incidents; enforce stop conditions, a kill switch, preserved prior briefs, and tested rollback.

Approved Market SourcesRead-only Internal BIScenario Assumption RegisterLeadership Review Queue
Readiness is specific to the briefing task and source set; strategic decisions remain human and governed.

Evidence record to maintain

  1. Question, decision date, source rights and freshness, metric definitions, fact/inference taxonomy, scenario assumptions, constraints, and owners.
  2. Representative, adversarial, and regression results with citation failures, unsupported claims, reviewer edits, and corrective actions.
  3. Access and data-flow reviews, distribution policy, decision log, incident runbook, stop conditions, rollback test, and dated owner approval.