How We Work

Defined scope. Evidence discipline. Independent analysis.

SWANK AI conducts structured, defined-scope reviews of the institutional systems responsible for governing artificial intelligence.

Every engagement begins with a clear question.

The purpose is to establish:

  • what is being reviewed;
  • why it is being reviewed;
  • what evidence is available;
  • which systems and processes fall within scope;
  • what limitations exist;
  • and what analytical deliverable is required.

Where applicable, reviews are conducted using the SWANK Institutional AI Assurance Framework — SIAAF.

No substantive review begins without an agreed scope.

The SWANK Review Process

A substantial SIAAF engagement follows eight stages.

01 Scope Definition
↓
02 Evidence Intake
↓
03 Evidence Mapping
↓
04 Institutional Systems Analysis
↓
05 Challenge Analysis
↓
06 Findings
↓
07 Factual Accuracy Review
↓
08 Final Report

The scope of individual engagements may differ.

The analytical pathway remains consistent enough for findings to be reviewable, traceable and repeatable.

01 — Initial Enquiry

An engagement begins with a concise description of the organisation, system, policy, workflow or AI use requiring review.

An initial enquiry should ordinarily identify:

  • the organisation or system concerned;
  • the operational or assurance question;
  • the AI system or use case, where relevant;
  • the reason independent review is being sought;
  • the principal documentation available;
  • any significant deadline or constraint.

SWANK AI reviews the enquiry to determine whether the proposed work is suitable for independent assurance and whether the question is sufficiently defined.

Further information may be requested before scope is agreed.

02 — Scope Definition

The question comes before the analysis

Where an engagement is suitable, its scope is recorded in writing.

The agreed scope may identify:

  • the precise assurance question;
  • the organisational system or process included;
  • relevant AI systems or workflows;
  • timeframe;
  • relevant SIAAF domains;
  • evidence expected;
  • exclusions;
  • known limitations;
  • intended deliverable;
  • timetable;
  • fees.

Scope protects both analytical independence and proportionality.

A review should be no broader than necessary to answer the question properly.

Choosing the Appropriate Review

The scope will ordinarily correspond to one of SWANK AI’s three principal assurance engagements.

Short-Form Operational Review

For one clearly defined institutional or AI-governance question.

Selected SIAAF domains are applied according to relevance.

Communication & Feedback Systems Review

For questions concerning how evidence, information, uncertainty, correction, communication and challenge move through an institution.

This ordinarily concentrates on:

Domain 03 — Evidence & Traceability

Domain 04 — Communication & Feedback Integrity

Domain 05 — Escalation, Challenge & Contestability

Full Institutional AI Assurance Review

For a broader assessment of a defined organisational AI environment.

All seven SIAAF domains are applied.

View Assurance Services →

03 — Engagement Confirmation

Once scope is agreed:

  • engagement terms are confirmed;
  • fees and payment arrangements are recorded;
  • relevant materials are identified;
  • transfer arrangements are agreed where necessary;
  • and substantive review begins.

The commissioning organisation defines the question.

It does not define the analytical conclusion.

04 — Evidence Intake

Relevant material is collected and registered according to scope.

Evidence may include:

  • policies;
  • procedures;
  • correspondence;
  • decision records;
  • AI-use documentation;
  • system documentation;
  • governance minutes;
  • training materials;
  • risk assessments;
  • incident records;
  • complaints;
  • contracts;
  • internal guidance;
  • workflow material;
  • version histories;
  • existing review records.

Not every engagement requires every category of material.

Evidence collection should remain proportionate to the question being examined.

Evidence Is Not Treated as a Single Category

SWANK AI distinguishes, where relevant, between:

Documented Fact

Directly supported by available evidence.

Corroborated Observation

Supported by multiple independent sources or records.

Analytical Inference

A conclusion reasonably derived from the available evidence but not directly documented.

Unresolved Issue

A material question for which sufficient evidence is unavailable.

Contradiction

Two or more material sources cannot presently be reconciled.

Recommendation

A proposed improvement arising from the analysis.

These distinctions remain visible throughout the review.

05 — Evidence Mapping

Before reaching conclusions, SWANK examines the structure of the evidence itself.

Evidence mapping may identify:

  • supporting records;
  • contradictory evidence;
  • missing material;
  • disputed facts;
  • source provenance;
  • chronology;
  • assumptions;
  • later corrections;
  • unresolved questions;
  • gaps between policy and operational record.

This stage is important because a coherent institutional narrative is not necessarily the same thing as a demonstrated one.

Where evidence conflicts, the disagreement may remain visible rather than being artificially resolved.

Source Traceability

Material findings should remain connected to the evidence supporting them.

Where relevant, SWANK may examine whether a pathway can be followed from:

institutional conclusion → analytical proposition → underlying source

This becomes particularly important where AI is used for:

  • summarisation;
  • retrieval;
  • classification;
  • chronology;
  • triage;
  • decision support;
  • record preparation.

A summary should assist access to evidence.

It should not make the evidence disappear.

06 — Institutional Systems Analysis

The relevant SIAAF domains are then applied to the defined question.

Depending upon scope, SWANK may examine:

  • governance and accountability;
  • decision ownership;
  • meaningful human oversight;
  • evidence and traceability;
  • communication systems;
  • feedback mechanisms;
  • escalation pathways;
  • challenge and contestability;
  • correction;
  • reconsideration;
  • risk and operational resilience;
  • AI literacy;
  • accessibility;
  • documentation;
  • organisational memory;
  • vendor dependencies;
  • implementation behaviour.

The analytical focus remains:

implemented behaviour rather than stated aspiration.

A policy may describe how a system is intended to operate.

SIAAF asks whether the available evidence demonstrates that it actually operates that way.

Five Cross-Cutting Tests

Relevant SIAAF reviews also apply five cross-cutting tests.

Reality Test

Does actual practice match written policy?

Evidence Test

Can the organisation demonstrate the claim it is making?

Challenge Test

Could a reasonable person question or correct this decision?

Failure Test

What happens when the AI, human or process gets something wrong?

Reconstruction Test

Could an independent reviewer later determine what happened, why it happened and who was responsible?

07 — Challenge Analysis

SWANK AI does not simply assemble evidence supporting the first apparent conclusion.

A provisional finding may be tested against:

  • alternative interpretations;
  • contradictory evidence;
  • missing records;
  • ambiguity;
  • changed circumstances;
  • later corrections;
  • plausible failure scenarios;
  • untested assumptions;
  • reasonable competing explanations.

The purpose is not to create artificial disagreement.

It is to determine whether the apparent conclusion remains reliable when subjected to challenge.

The analytical question is:

What would make this conclusion wrong?

A finding that survives reasonable challenge is stronger than one produced only from confirming evidence.

08 — Findings

Findings are structured using a consistent analytical chain:

OBSERVATION → EVIDENCE → ANALYSIS → SIGNIFICANCE → RECOMMENDATION

Observation

What was identified?

Evidence

What material supports the observation?

Analysis

What does the available evidence reasonably establish or indicate?

Significance

Why does the finding matter within the defined institutional system?

Recommendation

What proportionate improvement follows from the analysis?

This structure keeps the reasoning pathway visible.

Assurance Status

Where appropriate, relevant SIAAF domains may receive an assurance status.

A1 — EFFECTIVE

The institutional system is clearly defined, evidenced and functioning consistently.

A2 — PARTIALLY EFFECTIVE

Core arrangements exist, but identifiable weaknesses limit reliability or consistency.

A3 — WEAK

Material deficiencies exist in design, implementation, evidence or operation.

A4 — NOT DEMONSTRATED

Available evidence is insufficient to establish that an effective system exists.

Not Demonstrated does not automatically mean that something did not occur.

It means the available evidence was insufficient to demonstrate it.

Evidence Confidence

Findings may separately receive an evidence-confidence assessment:

HIGH
Strong, direct and consistent evidence.

MODERATE
Credible evidence exists but material limitations remain.

LOW
Evidence is incomplete, indirect or materially disputed.

Effectiveness and evidential certainty are therefore assessed separately.

Recommendation Priority

Where appropriate, recommendations may be classified:

P1 — Immediate

P2 — High

P3 — Planned

P4 — Enhancement

Priority indicates the significance and timing of recommended action.

It does not change the evidential status of the underlying finding.

09 — Factual Accuracy Review

Where appropriate, the commissioning organisation may be given an opportunity to identify:

  • factual errors;
  • incorrect dates;
  • misidentified documents;
  • missing records;
  • demonstrably inaccurate factual descriptions.

The purpose of this stage is accuracy.

It is not negotiation of the analytical conclusion.

A client may correct a fact.

A client may not rewrite SWANK’s independent analysis because the conclusion is inconvenient.

Read Standards & Independence →

10 — Final Report

The completed review records the analytical outcome within the agreed scope.

A substantial report may include:

  • Executive Assurance Summary;
  • Scope;
  • Methodology;
  • Evidence Base;
  • Evidence Limitations;
  • Institutional Systems Map;
  • Findings;
  • Domain Assessment;
  • Material Risks;
  • Recommendations;
  • Unresolved Questions;
  • Evidence Appendix.

The precise structure is proportionate to the engagement.

Institutional Systems Mapping

Where useful, SWANK may map how relevant parts of the organisation interact.

A systems map may show relationships between:

information

AI processing

human review

decision-making

communication

challenge

escalation

correction

accountability

This can make institutional weaknesses visible that are difficult to identify when each process is considered separately.

What Clients Can Expect

SWANK AI engagements are designed to provide:

  • a clearly defined analytical question;
  • written scope boundaries;
  • structured evidence review;
  • visible evidence limitations;
  • independent challenge analysis;
  • separation between evidence and inference;
  • proportionate findings;
  • identifiable recommendations;
  • written documentation;
  • a clear end point to the engagement.

The objective is not maximum documentation.

The objective is sufficient documentation to make the analytical process understandable and reviewable.

Client Responsibilities

Effective assurance depends upon the evidence made available.

Clients should identify, where known:

  • relevant records;
  • material evidence gaps;
  • disputed information;
  • known factual corrections;
  • significant system changes;
  • relevant incidents;
  • applicable restrictions on information;
  • evidence that may contradict an apparent institutional conclusion.

Clients remain responsible for ensuring that materials are lawfully supplied.

SWANK AI does not assume that client-provided information is complete merely because it has been provided.

AI-Assisted Materials

Clients may use AI to help:

  • organise documentation;
  • structure an enquiry;
  • summarise their own information;
  • identify questions;
  • prepare material for review.

AI assistance does not automatically make information reliable or unreliable.

Material claims remain subject to evidential assessment.

Where AI-generated material contains consequential:

  • factual claims;
  • citations;
  • statistics;
  • technical propositions;
  • legal propositions;
  • or source references,

verification may be required before reliance.

Confidentiality & Information Handling

Information should be collected and handled proportionately to the scope of the engagement.

SWANK AI does not require unnecessary personal or confidential information merely because it is available.

Where an engagement involves sensitive, confidential, restricted or personal material, appropriate handling arrangements may be agreed before transfer.

The review question should determine the information required.

Scope Changes

A defined engagement should not silently become open-ended.

If substantial new:

  • questions;
  • systems;
  • evidence sets;
  • objectives;
  • or review requirements

arise after work begins, they may fall outside the existing scope.

Where appropriate:

  • the original engagement may be completed as agreed;
  • scope may be extended in writing;
  • or a separate review may be proposed.

This preserves clarity, proportionality and analytical discipline.

Completion

An engagement concludes when the agreed review and deliverable are complete.

Unless separately agreed, completion does not create:

  • continuous monitoring;
  • operational management;
  • indefinite advisory responsibility;
  • or an ongoing assurance relationship.

Subsequent work may be commissioned through a new or extended scope.

What SWANK AI Does Not Do

Unless separately stated and appropriately qualified, an engagement does not provide:

  • legal advice;
  • regulatory approval;
  • statutory certification;
  • ISO certification;
  • compliance sign-off;
  • financial audit;
  • cybersecurity penetration testing;
  • source-code auditing;
  • marketing endorsement;
  • predetermined narrative support;
  • automated credibility determinations;
  • guaranteed outcomes.

SWANK AI provides independent analytical assurance concerning institutional systems and available evidence.

Five Working Principles

SCOPE BEFORE ANALYSIS

Define the question before examining the answer.

EVIDENCE BEFORE CONCLUSION

The strength of a finding should remain proportionate to the evidence supporting it.

CHALLENGE BEFORE CERTAINTY

A conclusion should be tested rather than merely confirmed.

IMPLEMENTED BEHAVIOUR OVER STATED ASPIRATION

Policies describe intended systems.

Assurance examines what happens in practice.

HUMAN ACCOUNTABILITY OVER AUTOMATION

AI may assist institutional judgment.

Responsibility must remain identifiable.

Begin an Enquiry

For an initial enquiry, please provide:

  • a brief description of the organisation or system;
  • the question requiring review;
  • the AI use or proposed use, where relevant;
  • the principal evidence available;
  • any material deadline;
  • and the intended purpose of the review.

Email: ai@swanklondon.com

View Assurance Services →

Explore SIAAF →

Read Standards & Independence →

View Pricing →

SWANK AI

Defined scope. Evidence discipline. Independent analysis.

We do not just review AI. We review the institutional systems responsible for governing it.

Scroll to Top