A Fulcrum & Co. product
Most companies have no idea what their AI is actually doing to them. The AI Architecture Audit measures it across seven dimensions, scores the integrity of every deployment against the load it carries, and produces the decision document and the 12-month plan.
The AI Architecture Audit is a 30–45 day diagnostic engagement. It scores AI Architecture Maturity across the seven dimensions Fulcrum already measures—each examined through what AI is doing to it—identifies the company’s AI Architecture Archetype, and produces six deliverables and three working sessions with a Fulcrum partner. It measures one thing precisely: the integrity of AI deployment relative to the load it carries, not how much AI the company uses.
The Execution domain reads against Bain’s decision-effectiveness research (Blenko, Mankins, Rogers). The Financial domain loads AI ROI net of supervision and rework against McKinsey’s State of AI findings on the gap between adoption and value capture. The Talent domain applies Smart’s Topgrading scorecard discipline. The compliance read carries the EU AI Act Article 4 AI-literacy obligation that did not move when the May 2026 Digital Omnibus deferred the high-risk timeline to December 2027. Every methodology reference is named and verifiable.
AI Architecture Audits are commissioned by the CEO facing board pressure to articulate an AI strategy, the CFO testing whether claimed AI return is real once supervision, integration, and rework are loaded against it, the board preparing for AI questions in a financing or M&A process, or the founder who cannot tell whether AI is creating leverage or quietly creating fragility.
Schedule a discovery conversation →What the AI Architecture Audit is built for
Four situations most leadership teams recognize.
A $30M company’s board asks the CEO for an AI strategy at the next meeting. The CEO has six tools in use across four departments and no single view of what any of them is doing to the business. The audit produces the scored diagnostic, the per-deployment decision document, and the board-ready language—so the CEO walks into the meeting with an architecture, not a list of subscriptions.
A $12M services firm reports AI-driven savings to its lenders. The CFO has never loaded the supervision time, the rework, or the per-seat price trajectory against the claimed gain. The audit runs that accounting deployment by deployment and separates the initiatives that are net-positive from the ones that are net-negative once senior review time is counted—and from the ones whose renewal curve no one has modeled.
A $60M company expects to run a sale process within eighteen months. Its most valuable workflows depend on AI, but the vendor contracts have never been reviewed for assignability, the prompts and tuned models have unclear IP ownership, and the data rights may not convey to a buyer. The audit surfaces the transfer gaps a Quality-of-Earnings team will probe—before the buyer’s diligence does.
A founder-led $8M company runs faster than ever on AI, and the founder is the prompt engineer, the quality-control gate, and the only person who knows which outputs to trust. The audit names the bus-factor-of-one workflows and the verification load that has migrated onto the founder, and sequences the work to convert trapped value into owned, transferable capability.
These are not edge cases. The pattern beneath them is the same: AI gets adopted tool by tool and department by department, and no one holds the cross-cutting view of what it is doing to decision velocity, margin, founder dependency, senior-team load, data exposure, talent concentration, or transferability. The firm that sold the tool measures adoption. The firm that implemented it measures uptime. Neither measures the questions a CFO, a board, a lender, or an acquirer asks. The AI Architecture Audit exists to answer them.
How the AI Architecture Audit is structurally different
Four structural commitments that shape what the audit is and isn’t.
Fulcrum holds no AI vendor relationships, takes no implementation revenue, and receives no platform referral compensation. The findings are not steered toward any tool. Independent in the structural sense, not the marketing-language sense.
Fulcrum measures; it does not deploy. The audit tells you what your AI architecture is doing and what to do about it. Building, integrating, and operating the tools is the company’s work or its chosen vendors’.
The audit does not tell you which AI products to buy. It evaluates the deployments you already run against leverage, fragility, dependency, ROI reality, compliance posture, and transferability—and recommends continue, restructure, sunset, or expand per deployment.
The audit produces the diagnostic, the decision document, and the roadmap. Execution is the company’s work. The audit is also not a substitute for legal, accounting, or regulatory counsel; it informs decisions that specific filings and transactions still require the appropriate professional advisors to execute.
The discipline
In the Execution domain, the rubric reads against Bain’s decision-effectiveness research (Blenko, Mankins, Rogers)—the finding that how well and how fast an organization makes and executes decisions tracks closely with its financial performance. In the Financial domain, it reads AI ROI net of supervision, integration, and rework against McKinsey’s State of AI findings on the distance between AI adoption and value capture. In the Founder-Constraint domain, it tests Warrillow’s hub-and-spoke dependency and Goldsmith’s letting-go capacity against the case most companies don’t expect—AI deepening founder dependency rather than reducing it. In the Talent domain, it applies Smart’s Topgrading scorecard discipline to AI competency. In the compliance read, it carries the EU AI Act Article 4 AI-literacy obligation—in application since 2 February 2025, enforceable by national authorities 2 August 2026, and not deferred when the May 2026 Digital Omnibus pushed the high-risk obligations to December 2027.
Specificity is the standard. Vagueness is the failure mode.
The score measures one thing precisely: the integrity of AI deployment relative to the load it carries—not how much AI the company uses. Three deployments running cleanly, with documented prompts, defined ownership, and bounded vendor exposure, score higher than twelve deployments accumulating bus-factor risk, governance gaps, and unsupervised output drift. The adoption-percentage benchmarks AI vendors and consultancies publish measure surface usage. This measures whether the architecture holds.
The standard exists because the alternative is the failure mode most AI commentary falls into: confident-sounding content that doesn’t survive a sophisticated reader’s scrutiny. The audit is built to be defensible to the senior people in the company’s life—the CFO, the board, the attorney, the lender, the acquirer’s Quality-of-Earnings team—when they read the work product.
Talk to a Fulcrum partner →What the AI Architecture Audit measures
The audit is not a new measurement domain bolted onto AI, and it is not a ninth Fulcrum instrument. It is a transverse lens: it takes the seven dimensions Fulcrum already measures in its leadership diagnostic and asks what AI is doing to each one. Each domain is scored on a five-level behavioral rubric. The scale is native to the audit and runs 1–5; it is not the platform’s EQI scale or CAM’s capital-readiness scale, and it is never averaged across them.
Anchored in Bain’s decision-effectiveness research (Blenko, Mankins, Rogers). The pattern it catches: AI accelerates the production of drafts, analyses, and summaries while quietly slowing the decisions those artifacts feed, because each AI output now carries a human verification loop that adds latency no one is tracking—and that verification burden migrates up to senior people. The score predicts whether decision throughput keeps pace with growth over the next 12–24 months or hits a verification ceiling.
Damodaran on revenue quality, read alongside McKinsey’s State of AI findings on the gap between adoption and value capture. The audit loads the real cost of each deployment—supervision overhead, integration cost, rework, and vendor price trajectory—against its claimed return. The pattern it exposes: an initiative reported to the board as a saving that is net-negative once senior review time and rework are counted, with per-seat pricing that has no modeled renewal trajectory. The score predicts margin exposure from unmodeled AI cost escalation over the next 12–24 months.
Warrillow’s hub-and-spoke dependency and Goldsmith’s letting-go capacity. AI is widely assumed to reduce founder dependency; the case this domain tests is the opposite—AI deepening it, because the founder becomes the prompt engineer, the quality-control gate, and the only person who knows which outputs to trust. The observable signal is a critical AI workflow with a bus factor of one. The score predicts a transferability gap that surfaces in diligence over the next 12–24 months.
Applies Fulcrum’s EVRI™ lens, informed by Carucci’s research on executive performance and derailment, to AI’s effect on senior-team load. The pattern it measures: the time AI was expected to give back gets absorbed into more work, with verification load concentrated on the few people trusted to check AI output. This is an organizational-continuity measure—executive capacity as an asset—not an assessment of any individual’s wellbeing or mental state. The score predicts vitality-risk concentration in the senior team over the next 12–24 months.
Edmondson on psychological safety and the Denison and Lencioni traditions on cultural alignment. The audit measures the gap between sanctioned AI policy and actual AI usage—shadow AI—and quantifies the share of the organization running company and customer data through tools the company has no contract, data-processing agreement, or audit trail with. The score predicts the likelihood of a governance or data-exposure incident, and a customer-disclosure gap, over the next 12–24 months.
Smart’s Topgrading scorecard discipline. The audit distinguishes AI-native judgment from AI-dependent habit and examines whether hiring scorecards assess AI competency at all. The concentration it locates: high-value AI workflows living in one or two people’s undocumented prompt habits, with no scorecard discipline behind the next hire. The score predicts a capability cliff when a key person leaves over the next 12–24 months.
The Exit Planning Institute / CEPA framework (Snider) and Quality-of-Earnings diligence standards. The audit examines whether the AI architecture can actually transfer in a transaction—vendor-contract assignability, IP ownership of prompts and tuned models, data rights, and the governance documentation a buyer’s Q-of-E team will probe. The transfer gap it surfaces: AI that makes the business run but cannot legally or operationally move to a buyer. The score predicts diligence friction and valuation exposure at a transaction over the next 12–24 months.
Within those domains, the audit inventories every AI deployment in the company across ten categories—the same taxonomy the Capability Map and the Decision Matrix use end to end:
Every deployment—sanctioned or shadow—is inventoried, scored, and given a recommended action.
The composite resolves to an AI Architecture Archetype—a structural classification, not a personality label or a grade. It is determined by two inputs, not by the score alone: the maturity composite across the seven domains, and a Low / Moderate / High read of how load-bearing AI is to current operations. Both inputs matter because a low maturity score means different things at different exposure levels.
High exposure with matching maturity: workflows documented and owned by more than one person, AI’s financial contribution measured net of overhead, dependency reduced rather than deepened, an architecture that would read as enterprise-value-accretive in diligence.
AI produces the appearance of capability while concealing a weakness underneath: the apparent output is high, but it doesn’t survive honest accounting, because supervision and rework consume the claimed gains or the AI is substituting for a capability the company should own. Same surface; opposite structure.
The archetype anchors the recommendation logic in the Decision Matrix and the sequencing in the Integrity Plan—a high-exposure, low-maturity company closes fragility before it adds capability.
Deliverables
Every engagement produces the same six deliverables. Depth is consistent at the single-tier price; companies that move into a Fulcrum tier engagement see this architecture scale.
The engagement
The audit runs 30–45 days from signed contract to Report delivery.
Three 90-minute working sessions with a Fulcrum partner anchor the engagement. Session 1 (Kickoff) frames the engagement, confirms the AI-decision owner and participants, and reviews the intake. Session 2 (Mid-Engagement Findings Review) presents first-pass findings across the seven domains so language and weight are calibrated against the company’s reality before the Report is finalized. Session 3 (Executive Readout) walks the bound Report and the archetype call, works through the Decision Matrix deployment by deployment so the leadership team leaves with decisions rather than findings, hands off the Integrity Plan with owners and milestones, and includes a 15-minute demonstration of the AI Strategic Mirror.
The engagement runs on a structured intake of roughly 85 fields—company financials and AI economics, the deployment inventory across the ten categories, team and decision-owner structure, governance posture, vendor concentration, and compliance context—plus a defined document review (organization chart, AI vendor contracts, any existing AI usage policy, recent board materials on AI strategy). The intake is completed before the Kickoff, so the first session reviews what was captured and surfaces what is missing. A small set of fields is answered by both the CEO and the executive who owns AI decisions; where their two answers diverge, the gap itself is a Cultural Coherence × AI signal the partner reviews in Session 1.
Every audit includes 60 days of complimentary access to the AI Strategic Mirror—a private, Claude-powered interface loaded with the company’s specific findings, the 12-month roadmap, and sector context, sourced directly from the Capability Map. The 60-day window begins at Report delivery. Companies that convert to a paid subscription within it do so at $2,000 per month (12-month minimum) with the $5,000 setup fee waived, because the Mirror’s setup is already complete—it was built from the audit’s Capability Map. The Mirror is a standalone product, not an audit accessory; the inclusion exists because the audit produces exactly the artifact the Mirror needs to run.
The audit is a single-tier standalone product. Companies whose findings warrant deeper, integrated work move into a Fulcrum & Co. tier engagement (Growth Diagnostic and above), where the AI transverse lens is integrated alongside the seven leadership-diagnostic domains rather than examined in isolation. The audit is the entry point; the tier engagements are where the work scales.
How to engage
Every engagement begins with a discovery conversation—a structured 45-minute call that surfaces the company’s AI footprint, the decision-maker’s current question, and whether the audit is the right starting point. A proposal follows, scoped to what the conversation surfaces.
On acceptance, the AI Architecture Audit Engagement Agreement and a Mutual Non-Disclosure Agreement are sent for signature. Full payment is due within five business days; on receipt, the engagement is scheduled, targeting the Kickoff within ten business days of signature. The intake and document request are issued at or before Kickoff and completed before Session 1. The 30–45 day sequence then runs to Report delivery and the Executive Readout.
FAQ
The firm that sold you an AI tool measures adoption. The firm that implemented it measures uptime. Neither measures what the deployment is doing to your decision velocity, your margin, your founder dependency, your senior-team load, your data exposure, your talent concentration, or your transferability in a transaction. The audit is independent—no vendor relationships, no implementation revenue—so its recommendations on which deployments to continue, restructure, sunset, or expand are not steered toward any product. It evaluates the architecture you have; it does not sell you the next tool.
Structural conflict of interest. An implementer’s revenue depends on the deployment continuing and expanding; a recommendation to sunset or restructure works against its own engagement. The “sunset” and “restructure” calls are precisely the ones an implementer is structurally least able to make. Independence here is not a marketing claim—it is the reason the recommendations can run in whichever direction the evidence points.
The leadership diagnostic measures the company’s operational and leadership architecture across seven domains and eight public instruments—execution infrastructure, founder or CEO dependency, cultural coherence, talent-system fragility, and the rest. The audit is a transverse lens over those same seven domains, measuring what AI specifically is doing to each. They are complementary products from the same firm. Companies whose audit findings warrant deeper, integrated work move into a tier engagement, where the AI lens runs alongside the full leadership diagnostic rather than in isolation.
No—by design. The audit applies Fulcrum’s existing methodology to AI context rather than introducing a new instrument. AI is too multidimensional to summarize in a single public composite, and the firm adds an instrument only after its methodology proves stable across delivered work. The maturity score is native to the audit and scored 1–5; it is not the EQI scale or CAM’s capital-readiness scale, and it is never averaged across them.
Different product, different measurement, different deliverable suite. CAM measures capital readiness—financial hygiene, cap table clarity, diligence readiness, and path-by-path eligibility—and produces a capital architecture and execution kit. The AI Architecture Audit measures what AI is doing to the company across seven operating domains and produces the AI diagnostic, the decision document, and the roadmap. Both are Fulcrum Architecture products; they answer different questions, and a company can engage both.
Low deployment volume is not the same as low exposure. A company with two AI tools in critical workflows—vendor contracts never reviewed, owned by one person, no fallback if the tool changes its API—carries higher structural risk than a company with eight tools that are documented, assigned, and bounded. The audit is built to produce a meaningful diagnostic at any deployment level; the per-deployment block in the intake scales with the company’s actual AI footprint, so a small footprint produces a precise read, not a truncated one.
The Mirror is provisioned from the Capability Map the audit produces—not from manual configuration. It is a private Claude-powered conversational interface, loaded with the company’s specific findings, the 12-month roadmap, and sector context, accessible to the CEO and leadership team during the 60-day complimentary window. After the window, conversion is available at $2,000 per month (12-month minimum); clients who convert within the 60-day window do so with the $5,000 setup fee waived because the Mirror’s setup is already complete. The Mirror is a standalone product; the 60-day inclusion exists because the audit produces exactly the artifact the Mirror needs to run.
A structured 45-minute call that surfaces the company’s AI footprint, the decision-maker’s current question, and whether the audit is the right starting point.