Methodology: how RankingLab ranks programs, schools and AI companies

In short: RankingLab is a research and technology lab. Our rankings are produced by our own research engine, a multi-agent AI system that orchestrates agentic search engines and frontier models purpose-trained for research, together with independent research and the judgment of doctoral-level experts. The engine reads the evidence on every candidate, from every lawfully accessible source, at a depth and pace no team could match. People decide.

Principles

• Verification before evaluation. Nothing is assessed until its legitimacy is established.
• Depth over visibility. We assess what an entity actually delivers, above all the depth and quality of its curriculum or its work, not how well known it is.
• Machines research. Experts decide. The engine gathers, verifies and organizes evidence; doctoral-level experts research alongside it and make every judgment.
• Provenance first. Every fact we use is traceable to its source, word for word.
• Integrity screening. Entities of questionable legitimacy or reputation do not appear in our rankings.
• Uncertainty, honestly stated. Every score comes with an uncertainty band.
• Built for the pace of AI. Rankings are refreshed during the year, and the system adapts as the field changes.

RankingLab follows the Berlin Principles on Ranking of Higher Education Institutions (IREG, 2006).

1. Scope: from learning to practice

RankingLab measures the path from academic training, through professional training, to application in the field and its outputs. We rank in three categories:

• Academic rankings, our core: degrees, programs and institutions. Quality in any profession starts with quality education.
• Practice rankings: where learning is applied. Companies, agencies, products, providers and professional training outside academia: everything that shapes how skills become results.
• Additional rankings: the links between learning and practice in the AI era, such as research labs, accelerators, learning platforms and communities.

Our current focus is business and management, software and AI development, engineering and related fields. The system is deliberately flexible: its frameworks, research protocols and agents are built to adapt as reality changes, so new fields and new kinds of rankings can be added without changing how we work.

2. Who can be ranked

Academic entities. Only institutions legally established and recognized, or authorized to award the degree, by the competent authority of their jurisdiction, confirmed in authoritative registers such as national higher-education registers, the IAU World Higher Education Database (WHED), EQAR's DEQAR, and the US Department of Education's DAPIP and the CHEA database. Programs must have graduated at least one cohort.

Practice entities. Companies, providers, credential issuers and other organizations must have a verifiable legal existence, identifiable leadership and a record of activity that can be examined.

The integrity screen. For every candidate we look for regulatory actions, sanctions, loss or suspension of accreditation, fraud findings, and credible, serious public reports of misconduct. Entities of questionable legitimacy or reputation are excluded, whatever their size, visibility or marketing. Exclusion is a decision not to rank, never a public finding about the entity. See Verification and the Eligibility Policy.

3. What we examine

We assess every entity as deeply as external means allow. The central question is always the same: what does this entity actually deliver, and how good is it?

Academia: degrees, programs and institutions

The curriculum is at the center of every academic assessment. It is where quality is built, and where the gap between a strong program and a weak one is widest.

• The curriculum itself. Structure and sequence, required and elective components, learning outcomes and course content, the depth of theory, methods and research training, and how current the content is, including how AI is taught and used across the core, not only in electives.
• Rigor and assessment. How learning is examined: exams, projects, theses and supervised research, and whether assessment matches what the degree claims.
• Duration and intensity. Length of study, credit load, contact hours, residencies and the workload behind the degree.
• Who teaches. Faculty qualifications, research and professional experience.
• Accreditation and its quality. Which accreditations are held, and the standing of each accreditor: recognized by a government or a recognized quality-assurance body, or an accreditation mill. Accreditation is evidence, never a precondition.
• Recognition and reputation. Recognition by competent authorities, standing among peers and employers, and recognition of the degree in practice.
• Reported problems. Violations, regulatory findings, sanctions and complaints, weighed by their seriousness and how reliably they are established.
• Outcomes in the field. Live projects, practitioner involvement and partnerships, and what graduates go on to do.

Academic entries are scored on the pillars of the academic rubric, each weighted for the level of study (see section 7).

Example. Each ranking sets how much each pillar counts for its field and level. A research doctorate puts more weight on academic depth; an MBA, on AI integration, field connection and standing. Every ranking page says in words what it examined and what counted most, as the Top 30 DBA Programs for the AI Era does.

Practice: companies, products, providers and professional training

For practice rankings we look at everything that can establish, and predict, the quality of what an entity delivers in the field. This includes professional training and credentials offered by companies and other organizations outside academic frameworks.

• Product and services. What the entity actually delivers, its depth and quality, and verifiable examples of its work and outputs.
• The people behind it. The skills, professional experience, education and record of its leadership and team, assessed only as they bear on the entity.
• Track record. Time in operation, verifiable clients, outcomes and contributions, and the consistency between what is claimed and what can be shown.
• Standing and conduct. Legal standing, professional recognition, reputation and any reported violations or problems.
• For training and credentials. Curriculum depth, assessment that means what the credential claims, and recognition in practice.

Practice rankings are scored against our professional framework for training providers, credentials and firms. The people behind an entity count only through what they bring to it: we never publish profiles of individuals.

Additional rankings

Additional rankings cover the entities that connect learning and practice. Each is scored against the framework closest to what it delivers, adapted where needed and published with the ranking.

4. The research engine

RankingLab runs on a research engine we designed and built ourselves. It reflects how we believe a research and technology lab working on rankings should operate today: agents do the reading at scale, every claim is anchored to its source, machines check each other, and people hold authority at every decision point.

Human direction

Editorial panel

Defines each research brief, approves candidate lists and directs research. Its input outranks every agent: an editor's value is never overwritten.

Orchestration

• Research plannermaps every question to the framework
• Priority schedulereditor tasks first
• Depth controlstaged, targeted depth
• Audit trailevery step recorded

Specialist agents, in parallel

• Discoverybuilds the candidate universe
• Extractionagentic search, field by field
• Verificationan independent engine re-checks
• Adjudicationreads the source, settles disputes
• Deep researchwhere the evidence is thin

Evidence ledger

• Sourceevery value traceable
• Exact quotechecked against the live page
• Confidencestated per value
• Indicatormapped to the framework

Human judgment

Doctoral-level research, scoring and decision

Experts run their own research in parallel, score against the published framework, and the editorial panel decides the final ranking.

Its design principles:

• Orchestrated specialists. A planner breaks each ranking into research tasks mapped to the framework. Dozens of specialist agents run in parallel, each assigned to what it does best: discovery, extraction, verification, adjudication or deep research.
• Agentic search and research-grade models. Agents search, browse and read like researchers: across official registers, institutional and company publications, open databases and the wider web, using frontier models purpose-trained for multi-step research.
• Disagreement as a signal. Every value is extracted by one engine and independently re-checked by another. Where they disagree, a third model reads the cited pages and decides from the source itself, and the disagreement is recorded.
• Provenance by design. Every value carries its source, an exact quote, a confidence level and a date. Quotes are verified against the live page before a value is accepted.
• Depth where it matters. Research is staged: a plan, a candidate universe, a first pass over every candidate, then targeted deep research where the evidence is thin or the stakes are high.
• Multilingual. Research runs in the languages entities publish in, so those that publish mainly in other languages compete on equal terms.
• Human authority. Editors can steer, override or stop any task. Their instructions shape every prompt and go to the front of the queue, and no automated stage publishes anything.
• Accountable. Every step is logged, and the AI systems in use are disclosed under our Responsible AI Use Policy.

AI systems never assign scores, decide eligibility or publish anything without human review.

5. How a ranking is made

1. 01BriefHuman decision

   The editorial panel defines scope, criteria and framework.

2. 02Candidate universeEngine, human-verified

   Discovery agents map every eligible candidate; editors approve the list.

3. 03Verification gateHuman decision

   Recognition, legal standing and reported misconduct checked. Fail, and an entity is never ranked.

4. 04First passEngine, human-verified

   Every candidate researched field by field by one engine, checked by a second.

5. 05AdjudicationEngine, human-verified

   Disagreements settled by a third model reading the cited source itself.

6. 06Deep researchEngine, human-verified

   Targeted, multi-source research where it matters, directed by editors.

7. 07Expert research and scoringHuman decision

   Doctoral-level researchers cross-check and score against the framework.

8. 08Decision and publicationHuman decision

   The editorial panel decides; scores ship with uncertainty bands.

6. Human research and judgment

Experts are in the loop from the first step to the last. Researchers with a PhD or equivalent doctorate shape each research plan, approve candidate lists and direct further research. Alongside the engine, they carry out their own academic research: reading primary sources, following their own lines of inquiry, and cross-checking what the engine found.

Entries are scored against the published framework by doctoral-level assessors, independently and blind to each other's scores where more than one assessor reviews an entry; significant disagreements are resolved by a senior assessor. The editorial panel then assembles the final ranking: it weighs the evidence, the scores and the cross-checks, and applies its own expert judgment, including where the evidence is incomplete or where a strict application of the scores would mislead.

Independent expert studies

Where independent expert studies of a field exist, the editorial panel puts them to work. Each study ranks its own top ten under its own criteria. The panel gives every study equal weight, converts its positions into points (first place earns 10 points, tenth place 1, a program it does not list 0), averages the points and sets the top of the ranking in that consensus order. Ties go to the program that more studies rank, then to its best single position.

The consensus moves positions, never facts. A placed entry keeps its evidence and pillar profile; its overall score is set between its new neighbours, and its pillars and bands move with it, so the table reads consistently. A program a study names joins the research first, and is listed only once it passes the verification gate: recognition by a competent authority is never waived. Where our own evidence on a placed program is still thin, the panel's judgment, informed by the studies, carries its position. The ranking page sets out each study's approach, where every program stands in each study, and the reasons the studies give, with the program features they cite. The studies are published as expert opinions, without their authors' names.

Everyone who takes part declares conflicts of interest for every assignment and steps aside where conflicted. See the Assessor Code of Conduct.

7. Scoring frameworks

Every indicator is scored on the same five levels, from Absent to Leading, each with a written description. Academic rankings use the pillars below and a capped external reference signal. Reputation & Standing counts only in the rankings that use it. The full indicator set is in the academic rubric. Practice and additional rankings use the professional framework.

AI Integration

Is AI part of how and what students learn?

• AI.1 AI across the core curriculum
• AI.2 Responsible AI and governance
• AI.3 Tools, data and compute access
• AI.4 AI policy and teaching capability

Academic Depth

Is the learning rigorous and grounded in research?

• DEP.1 Research quality and relevance
• DEP.2 Faculty qualifications and research activity
• DEP.3 Rigor of assessment
• DEP.4 Doctoral training and supervision (institutions, DBA, PhD only)

Field Connection

Does learning meet real organizations and real problems?

• FLD.1 Experiential learning with partners
• FLD.2 Practitioner involvement in the curriculum
• FLD.3 Applied research and knowledge transfer
• FLD.4 Entrepreneurship and innovation support

Future Readiness

Is the program built to keep changing?

• FUT.1 Curriculum renewal
• FUT.2 Interdisciplinary pathways
• FUT.3 Sustainability, society and the future of work
• FUT.4 Lifelong learning (master's, institutions, DBA only)

Learner Outcomes

What happens to people who study here?

• OUT.1 Graduate outcomes
• OUT.2 Completion and support
• OUT.3 Access and value
• OUT.4 Learner and alumni voice (from edition 2028.1)

Reputation & Standing

How strong are the institution, its accreditation and its graduates?

• REP.1 Institutional reputation (computed by the data team)
• REP.2 Global leadership in the field (computed by the data team)
• REP.3 Accreditation (computed by the data team)
• REP.4 Alumni profile (computed by the data team)

External Reference

What do established, permissibly usable sources say?

• EXT.1 External reference signal (computed by the data team)

8. Scores, bands and editions

Each entry receives a score out of 100 with a 95% uncertainty band and a rank band, computed by resampling the evidence thousands of times, so readers can see where differences are meaningful. Scores are composite judgments, not exact measurements: the band shows how far the evidence supports them. Editions are numbered by year and release (for example 2027.1), rankings are refreshed as new evidence is reviewed, and every release and correction is logged on the updates page.

9. Sources

We examine every database and source available online that can be reached lawfully: official registers and publications, curricula and program regulations, accreditation and quality-assurance reports, regulatory and court records, company and professional registers, open bibliometric and institutional data, research literature, professional and industry sources, and public reporting. Where sources conflict, the engine records the conflict and the source itself decides.

Established rankings are valuable and we respect them. Where their terms permit, they enter our scores only as a small, capped external reference signal, and we never reproduce third-party ranking tables. See the Data Sources Policy.

10. What a RankingLab ranking is

A RankingLab ranking is an independent expert opinion, formed by applying this published methodology, including weighted criteria and editorial judgment, to the evidence available to us at the time. It lists the entities we judge to be the strongest in its scope; it is not an exhaustive list, and absence from it is not a statement about any entity. The underlying facts are sourced and dated; the positions are judgments. See the Legal Notice.

11. Known limitations

• External assessment can only see what is published or otherwise observable. Entities that publish more, and in English, are easier to assess. Multilingual research reduces this but cannot remove it.
• Curriculum documents describe intentions; outcome data lags by a year or more.
• Fields with few eligible entities produce wider bands. We publish them rather than hide them.