A product claim, a reported result and a verified outcome answer different questions. Our method keeps those distinctions visible, so leaders can judge what the evidence supports.
The Center of Eminence is built on a fundamental principle: decisions about AI should be informed by evidence, not simply by what technology providers promise.
One standard across three centers
These standards apply to all research published by the Center of Eminence: CoE | AI Cost & Governance, for finance leaders; CoE | AI Talent & People, for HR and people leaders; and CoE | Vendor Intelligence, for buyers of AI. They also apply to the research drawn from the two shared tools, Eminence Scan and EmTrace.
Each center organizes its coverage around the decisions its readers make: what AI costs and how to control it in finance work; whom to hire and how to assess judgment in AI-assisted work; and which providers to consider and what to ask them. Coverage expands as new research areas complete our quality review process.
Formats include practice notes, workflow playbooks, research and analysis, product and provider profiles, market and regulatory watch pages, and diligence questions.
Our objective is not to declare a universal best product or assign a single numerical score to solutions that may perform differently across organizations. Instead, we aim to give readers the information, context, and evaluation criteria needed to determine suitability for their own circumstances.
Five principles
- Evidence before conclusions. Material factual claims must be supported by identifiable sources. Conclusions reflect the strength of the available evidence, not the prominence of a provider or the persuasiveness of its marketing.
- Transparency about uncertainty. Where evidence is incomplete, conflicting, or unavailable, we identify those limitations rather than presenting assumptions as established facts.
- Clear separation of information types. Provider claims, third-party research, documented outcomes, practitioner experiences, and our own analysis are distinguished so readers can see what supports a conclusion.
- Continuous evaluation. AI technologies, regulations, commercial offerings, and market conditions evolve. Published research is reviewed and updated as new evidence emerges.
- Editorial accountability. AI-supported systems assist with research, drafting, analysis, and verification, while Eminence Growth Solutions retains responsibility for editorial decisions and published content.
What supports a statement?
Not all information carries the same evidentiary weight. We classify information by its source, nature, and level of substantiation.
| Evidence category | How to read it |
|---|---|
| Vendor-stated claim | What a provider says about capabilities, requirements, performance, or benefits. It establishes what the provider represents; it does not independently demonstrate that a product achieved a business outcome. |
| Independent market evidence | Research, regulatory publications, third-party assessments, and other material produced independently of the provider being evaluated. Its credibility, methodology, and limitations determine its evidentiary value. |
| Substantiated outcome | A documented result with identifiable support. We distinguish vendor-reported customer outcomes, independently corroborated results, and outcomes established through direct testing or other verifiable evidence. |
| Pricing and commercial observation | A dated observation about price, packaging, licensing, implementation, or contractual terms. |
| Known limitation | A documented constraint, dependency, functional gap, or risk. Demonstrated limitations are distinguished from potential risks and unresolved questions. |
| CoE analysis | Our interpretation of the evidence and its implications for decisions, kept separate from the evidence so readers can assess both. |
| Community and practitioner observation | An experience or pattern reported by practitioners, customers, or industry participants. It can identify an issue worth examining; it is not automatically a verified fact or a representative market finding. |
| Member input under review | Feedback, a correction, additional evidence, or a firsthand experience submitted by a member that has not yet completed editorial evaluation. It remains private during review and does not automatically become part of published research. |
Our guiding distinction is straightforward: a claim, an observation, and a substantiated outcome are not interchangeable.
How we examine sources and claims
Research begins with relevant primary and secondary sources: legislation, standards, regulator publications and guidance; provider documentation, commercial disclosures and release notes; published research and independent assessments; corporate filings, court records and other public disclosures; and documented practitioner experience. Where available and appropriate, primary sources are prioritized over secondary reporting.
A citation alone is not sufficient evidence that a claim has been verified. We examine the source for the specific information being relied upon, including its context, publication date, applicability, and limitations. Automatically generated summaries can help locate a source; they are not a substitute for reviewing it.
Material factual claims have evidence records showing what each claim relies on, where its supporting evidence came from, and when the material was reviewed. These records support later verification, editorial review, and reassessment when a source changes.
Verification challenges the draft
A separate AI-assisted verification process is designed to challenge, rather than simply confirm, the initial analysis. It examines whether the cited evidence supports each claim, whether contradictory information exists, and whether material qualifications have been overlooked. Where appropriate, verification is performed independently of drafting to reduce the risk of reinforcing unsupported assumptions.
Claims that cannot be adequately substantiated are revised, removed, or explicitly identified as unconfirmed where their inclusion remains relevant.
Numbers need a method as well as a source
Financial calculations, estimates, and other quantitative assessments receive additional scrutiny. Where applicable, inputs, assumptions, calculations, and outputs are examined separately to identify mathematical errors, inconsistent definitions, or unsupported interpretations.
We distinguish documented historical outcomes, provider-reported estimates, independently sourced market data, and our own planning models and illustrative scenarios. Illustrative assumptions are not presented as established industry benchmarks. Where we provide cost models or potential savings calculations, the methods and assumptions are published so readers can test the results with their own inputs.
Accuracy includes proportion
A technically accurate claim can still mislead when important limitations are omitted, contradictory evidence is minimized, or conclusions exceed what the evidence demonstrates. Our review considers whether material uncertainties, risks, dependencies, and alternative interpretations are given appropriate weight.
What can change between reviews?
Prices and contracts. Published prices are observations at a point in time, not permanent commercial commitments. Where pricing is included, we aim to identify the source, the date of the published pricing, relevant currency, and material qualifications. Public list prices are distinguished from negotiated enterprise terms, and implementation or ongoing operating costs are not assumed to be included unless the evidence supports that interpretation.
Rules and legal developments. We distinguish proposed requirements, enacted legislation, final regulations, guidance, and enforcement actions, in the context of their jurisdiction, effective dates, and current status. Historical or superseded requirements are not presented as currently applicable.
Products. Releases, revised documentation, commercial restructuring, and changes to underlying technology can change an assessment. Product assessments identify the timing of their information and are reviewed when material developments occur.
When we review content
The review cycle depends on the nature of the information and how quickly it is likely to change.
| Review tier | Standard cadence | Typical coverage |
|---|---|---|
| Product Watch | Monthly and following relevant product changes | Named product profiles, functionality, documentation, and capabilities |
| Active Watch | Quarterly and following material events | Pricing, regulatory developments, consultations, enforcement, incidents, and site policies |
| Periodic Confirmation | Semiannually | Survey findings, industry perspectives, reference materials, and glossary content |
| Established Reference | Annually | Enacted statutes, final standards, and concluded matters |
Pages identify their review classification and last-reviewed date. The date shown on a piece is the date of its analysis; later reviews and changes are recorded in its last-reviewed date and version history. A scheduled review may confirm that existing information remains accurate without a substantive revision.
Material changes may trigger review ahead of schedule. Where a source becomes outdated, we assess whether the affected claim can be updated on its own or whether the broader assessment needs reconsideration. We distinguish routine updates, material corrections, and changes to analytical conclusions. Content past its scheduled review date should be identifiable as overdue until it is reassessed.
These review cycles establish a maintenance standard, not a guarantee that every external development will be identified immediately.
Specialist subjects and consequential claims
Tax, accounting, legal, employment, and insurance matters carry professional, regulatory, or legal implications, particularly where interpretation depends on jurisdiction, professional standards, or an organization’s circumstances. We provide research, orientation, and decision-support information in these areas, but do not present general research as a substitute for advice from appropriately qualified professionals.
Content on these subjects is identified as orientation rather than professional advice, so readers know to apply it with their own qualified advisers.
Potentially consequential claims about named organizations receive additional attention to the underlying evidence, source reliability, context, and any material disputes. Unverified allegations and vendor-generated assertions are not treated as established findings.
What practitioner experience can contribute
Publicly available research does not always capture what it is like to evaluate, implement, or operate AI. Practitioners can surface unexpected costs, integration challenges, service limitations, governance concerns, and differences between advertised capabilities and actual use.
When we use practitioner observations, we consider source credibility, date, relevance, claimed firsthand knowledge, corroboration, and whether similar themes appear across independent accounts. Published observations may be classified as isolated reports, recurring themes, disputed accounts, or corroborated findings. A recurring observation can justify investigation without establishing how common it is across the market.
Public practitioner content is curated with reference to its original source and applicable usage rights. Social media discussions and review-site ratings are not treated as standalone proof of product outcomes. Contributions sent to us follow a separate private review process before any resulting information is considered for publication.
Who decides what is published?
AI assistance does not eliminate editorial accountability.
AI-supported systems assist with source discovery, evidence extraction, claim evaluation, drafting, independent verification, and monitoring for relevant changes. Eminence Growth Solutions retains responsibility for what is published and for ensuring that editorial conclusions reflect the available evidence.
Published material is subject to approval controls, and automated research processes do not independently authorize publication. Materially revised content passes through the applicable review and approval process.
Our Editorial Independence & Right of Reply policy covers ownership, commercial relationships, editorial independence, and how organizations can respond.
How the record can be challenged
Research quality depends in part on the willingness to revisit conclusions when better information becomes available. We welcome factual corrections, additional evidence, relevant developments, and constructive challenges from readers, practitioners, and organizations referenced in our coverage, through Help Improve the Center of Eminence. Contributions are reviewed privately and are not published automatically.
The source of a correction request does not determine whether it is accepted. The quality and relevance of the evidence do. Where a material error is identified, the affected content is corrected and the change is documented under our Corrections & Updates policy, which distinguishes routine updates, factual corrections, and material revisions to an assessment.
What the method cannot do
No research process can eliminate uncertainty, particularly in a market evolving as rapidly as AI. Information may change between scheduled reviews. Relevant evidence may not be publicly available. Commercial terms may vary across buyers. Reported outcomes may depend on conditions that are not fully disclosed. AI-supported research and verification can misinterpret a source, overlook context, or miss contradictory evidence.
Our methodology is designed to reduce these risks through structured research, source traceability, independent verification processes, editorial controls, and continuing review. It does not constitute external certification, independent audit assurance, or a guarantee that every published assessment is complete or error-free.
Where consequential uncertainties remain, we make them visible so readers can incorporate them into their own evaluation and decision-making.
The Center of Eminence is designed to help leaders make decisions based on what the evidence supports, with a clear understanding of what it does not. The objective remains the same: credible, practical intelligence that supports more informed investment decisions, stronger governance, and responsible AI adoption.
Version history
| Version | Date | Change |
|---|---|---|
| 2 | 2026-10-08 | First published. |
| 3 | 2026-10-19 | Clarified that prices carry the date of the published pricing, that the date on a piece is the date of its analysis, and that specialist subjects are identified as orientation rather than professional advice. |
| 4 | 2026-10-20 | Extended these standards to the whole Center of Eminence: its three centers (AI Cost & Governance, AI Talent & People and Vendor Intelligence) and the research drawn from Eminence Scan and EmTrace. Reorganized the page around readers’ questions, with the evidence categories in one table. |