AWS AIB-C01 Glossary

Every term below appears in, or directly underpins, the AWS Certified AI Business Strategist (AIB-C01) exam guide. Each entry gives a business-level definition, why it matters on this exam, and a domain tag showing where it is tested most. Remember the framing that shapes the whole credential: the exam does not assess AWS services knowledge — services appear only at the strategic level, and every entry here is written for the product manager, consultant or line-of-business leader who works alongside technical teams rather than building anything. See the course overview, the exam logistics and the AWS exam guide for how these terms map to the blueprint.

Domain tags: D1 = AI Fundamentals and Literacy · D2 = Strategy and Business Value · D3 = Governance and Responsible AI · D4 = Readiness, Leadership and Transformation.

---

A

Access controls – The rules that decide who can reach which data and which AI systems, and under whose permissions. On this exam you identify appropriate access controls and data-security measures as a governance concern, not as an IAM configuration task. D3

Agent (AI agent) – An AI system that can plan and act toward a goal with some autonomy, rather than answering a single prompt. Core capabilities to recognise are autonomy, tool use, agent-to-agent communication and orchestration strategies. The exam wants you to distinguish an agent from a plain model or a rule-based tool, not to build one. D1

Agent-to-agent communication – The pattern where multiple AI agents coordinate or hand off work to each other. It is one of the four agent capabilities the guide names; on the exam it signals higher autonomy and therefore higher oversight and governance needs. D1 D3

AI (artificial intelligence) – The broad field of systems that perform tasks associated with human intelligence. On the exam AI is the umbrella term; the discriminator is distinguishing it from the narrower ML and from GenAI, and knowing when not to use AI at all. D1

AI Service Cards – AWS’s transparency artefact documenting a service or model’s intended use cases and limitations, responsible-AI design choices, and performance best practices. They matter for transparency and vendor due-diligence discussions; know what they contain, not how to author one. D3

Align (AWS CAF phase) – The second AWS CAF phase, which identifies capability gaps across the six perspectives, cross-organisational dependencies and stakeholder concerns before delivery begins. The exam guide’s loose wording for task 4.4.1 says “experiment” here, but CAF’s own phase is Align. D4

Algorithm – The procedure a model uses to learn from data or produce an output. The exam treats “algorithm” as basic vocabulary you can explain in a business context; selecting or tuning algorithms is explicitly out of scope. D1

Amazon Bedrock – AWS’s managed platform for foundation models and generative-AI applications, offering multi-provider model choice on one platform. At the strategic level it represents the buy/managed end of the build-buy-partner spectrum: fast access to models without running infrastructure. D2 D1

Amazon Quick – AWS’s AI-powered business assistant and business-intelligence surface for research, insights, automation and no-code app building over business data, on a seat-based model. The strategic point: it puts AI in front of business users without a build project. D2 D4

Amazon SageMaker AI – AWS’s service for building, training and deploying custom ML models. At the strategic level it represents the build/custom end of the spectrum, with instance-based pricing where idle endpoints still cost money — the managed-versus-custom decision is the exam-relevant judgment. D2 D1

Automated Reasoning checks – A Bedrock Guardrails capability that uses formal logic to verify model outputs against defined rules, priced per policy. Know it as one of the guardrail types that reduces reliability risk, not its implementation. D3

Autonomy – The degree to which an AI agent acts without step-by-step human direction. More autonomy raises both potential value and risk, which is why higher-autonomy systems attract more human oversight and governance on the exam. D1 D3

B

Baseline (baseline metrics) – The measured level of a metric before an AI initiative begins. Establishing a baseline before implementation is an explicit skill (2.2.2); without it you cannot honestly claim improvement or calculate ROI. A missing baseline is a classic exam trap. D2

Batch inference – Running model requests in bulk when latency does not matter, priced in Bedrock at 50% below on-demand for select models. It is the go-to cost lever for non-real-time workloads; recognise it as a cost-optimisation choice, not a technical setting. D2

Bedrock Data Automation – A Bedrock capability that turns unstructured multimodal content into structured data, priced per page, per minute or per image. Strategically it is a managed way to unlock value from unstructured data without a data-engineering build. D2 D1

Bedrock Guardrails – Configurable safeguards usable with any foundation model in Bedrock or self-hosted, including content filters, denied topics, sensitive-information (PII) filters, word filters, contextual grounding checks and Automated Reasoning checks. AWS states Guardrails blocks up to 88% of harmful content; on the exam they are the archetypal safeguard for governance and human-oversight scenarios. D3

Bedrock Knowledge Bases – Managed retrieval-augmented generation without running infrastructure: managed parsing, embeddings and re-ranking, with index storage and per-call retrieval pricing. Strategically, this is how you ground a model in your own data through a buy rather than a build. D3 D1

Bedrock service tiers – The commercial tiers for Bedrock inference: Standard, Flex (a 50% discount to Standard), Priority (a 75% premium to Standard) and Reserved. Choosing a tier is a cost-versus-latency-versus-predictability trade-off — the kind of judgment the exam rewards. D2

Bias – Systematic unfairness in AI outputs. The exam stresses that bias can arise at multiple stages of the lifecycle (data, training, deployment) and must be monitored over time, including bias drift; it links to the fairness responsible-AI dimension. D3

Bias drift – The gradual worsening of bias in a deployed system as data or conditions change. Monitoring for bias drift in production is an explicit governance skill (3.3.2); it is why fairness is an ongoing operational duty, not a one-time check. D3

Build-buy-partner – The decision of whether to build an AI capability in-house, buy an off-the-shelf solution, or partner with a vendor. Skill 2.1.2 asks you to weigh budget, timelines, capabilities, vendor proposals and regulatory compliance; AWS Marketplace supports the buy/partner side of this analysis. D2

Business case – The structured argument for an AI investment, combining expected value, cost, risk and strategic fit. Building the business case is one of the four things AWS says the exam validates; strong cases rest on baselines, KPIs and ROI, not enthusiasm. D2

Business continuity – Keeping the business running through change. It is a named consideration when transitioning processes to AI or between AI platforms (2.1.5) and when scaling (4.4.6); the exam wants you to weigh disruption risk against expected benefit. D2 D4

C

Capability gap – The distance between the skills, processes, technology or governance an organisation has and what its AI ambitions require. Identifying gaps across people, process, technology and governance (4.1.3) feeds directly into prioritising development investments. D4

Center of excellence (COE) – A cross-functional team that concentrates AI expertise, sets standards and supports the wider organisation as it scales. Establishing AI COEs (4.4.3) is a scaling mechanism, not a pilot activity; recognise it as an enterprise-maturity move. D4

Champion (AI champion) / champion network – Influential individuals who advocate for and accelerate AI adoption within teams. Identifying and empowering champions (4.3.1) and building a champion network is a change-management lever that complements executive sponsorship from above. D4

Change management – The disciplined effort to help people adopt new ways of working. On this exam it covers transparent communication about timelines and role impacts, addressing cultural barriers and fear, and leadership interventions — the human side of transformation. D4

Compensatory scoring – A scoring model where a strong overall score can offset weaker sections; you must pass the exam overall, not each domain. AIB-C01 uses this model, so section-level feedback is directional only and AWS says to interpret it with caution. D1 D2 D3 D4

Compliance risk – The risk of breaching laws, regulations or contractual obligations with an AI system. Identifying and addressing regulatory compliance risks (3.2.2) is a governance skill; the exam asks you to apply a risk-classification approach, not to cite specific statutes. D3

Concept drift – Drift where the real-world relationship the model learned changes over time, degrading accuracy even if the input data looks stable. It is one reason AI needs ongoing monitoring and updates (1.2.3); distinguish it from data drift and model drift. D1 D3

Consumption-based pricing – Paying per unit of use — per token, request, page or image — so cost scales with usage and has no floor. Distinguishing it from instance-based and seat-based pricing (2.2.5) is exam-relevant because each shapes the business case differently. D2

Content filters – A Bedrock Guardrails control that blocks harmful categories of input and output. It is the archetypal safeguard behind AWS’s claim that Guardrails blocks up to 88% of harmful content; know it as a harmful-content control (3.3.3). D3

Context window – The maximum amount of text (measured in tokens) a model can consider at once. When inputs exceed it, older content is dropped or must be summarised, which degrades GenAI performance (1.3.2); recognise the symptom, not the fix. D1

Contextual grounding checks – A Bedrock Guardrails capability that checks a response against a source to catch hallucinations. It is the concrete safeguard behind “hallucination detection” in human-oversight scenarios (3.1.4). D3

Controllability – One of AWS’s eight responsible-AI dimensions: the ability to monitor and steer an AI system’s behaviour. It underpins human-oversight and escalation design; note the exam guide’s shorter list omits it, but you should know all eight. D3

Cost optimisation – Deliberately reducing AI spend without undermining value — via tier choice, batch inference, commitment discounts, prompt routing and monitoring actual cost. Implementing basic cost-factor controls (2.2.5) is an explicit skill. D2

Cross-functional team – A team combining business leads, technical experts, legal and compliance with clear accountability. Building these teams (4.3.2) is how governance boards and COEs get real reach; single-function ownership of AI is a recurring weakness. D4 D3

Cultural barrier – An organisational attitude that resists AI adoption, such as risk aversion, resistance to change or fear of failure. Recognising these and identifying leadership interventions (4.3.4) is a leadership skill, not a technical one. D4

D

Data drift – A change in the statistical properties of input data after deployment, which can silently degrade model performance. It is a core reason for ongoing monitoring (1.2.3, 3.3.4); distinguish it from concept drift and model drift. D1 D3

Data ownership – Clarity over who is accountable for a dataset’s quality, access and use. Describing the importance of data ownership alongside data strategy and data-sharing frameworks (4.2.2) is a readiness skill. D4

Data quality – The fitness of data for its purpose, across dimensions such as accuracy, completeness, consistency, timeliness and validity. Poor data quality degrades AI outcomes (1.1.4) and is a reliability risk in production (3.3.4); “garbage in, garbage out” is the exam intuition. D1 D3 D4

Data-sharing framework – The agreed rules for how data moves between teams, systems or partners while respecting ownership, privacy and compliance. It is named alongside data strategy and ownership as a foundation for AI readiness (4.2.2). D4

Data silo – Data trapped in one team or system, inaccessible to others who need it. Evaluating the effects of data silos (4.2.1) is central to data-readiness assessment; silos are a common blocker to enterprise-scale AI. D4

Data strategy – An organisation’s plan for how it collects, governs, shares and uses data to create value. Describing its importance (4.2.2) is a readiness skill; without a data strategy, AI initiatives stall on access and quality problems. D4

Decision rights – Explicit statements of who may make which decisions about an AI system. Clear decision rights and accountability are the backbone of a governance structure (3.2.1); vague ownership is a governance anti-pattern. D3

Denied topics – A Bedrock Guardrails control that blocks specified subject areas. Know it as one of the safeguard types for managing harmful-content risk (3.3.3), at the level of what it does, not how to configure it. D3

Drift – The umbrella term for a deployed model’s performance degrading over time as data, relationships or bias shift (model drift, data drift, concept drift, bias drift). Detecting and remediating drift is why AI needs ongoing monitoring and updates (1.2.3). D1 D3

E

Envision (AWS CAF phase) – The first AWS CAF phase, which identifies and prioritises transformation opportunities against business objectives. It is where an AI transformation starts before capability gaps (Align), pilots (Launch) and expansion (Scale). D4

Escalation criteria – The predefined conditions under which an AI system hands a decision to a human. Identifying escalation criteria is a named safeguard for human oversight (3.1.4); they turn “keep a human in the loop” into an operational rule. D3

EU AI Act – A risk-tiered regulation classifying AI uses as unacceptable, high, limited or minimal risk, with obligations attached to the tier. On this exam it is an example of applying a risk-classification framework (3.2.4), not a source of article numbers to memorise. D3

Exception path – A defined route for handling cases that fall outside normal AI-governance rules, so edge cases get human judgment rather than silent failure or ad-hoc workarounds. It complements decision rights within a governance structure. D3

Executive sponsorship – Visible, resourced backing for an AI initiative from senior leadership. Establishing executive sponsorship and leadership alignment (4.3.1) is repeatedly the difference between pilots that scale and pilots that stall. D4

Explainability – One of AWS’s eight responsible-AI dimensions: the ability to understand and articulate why a model produced an output. It matters for trust, compliance and human oversight; SageMaker Clarify supports it at the tooling level. D3

F

Fairness – One of AWS’s eight responsible-AI dimensions: treating people and groups equitably and avoiding harmful bias. Applying fairness to business scenarios (3.1.1) and monitoring for bias drift are recurring exam themes. D3

Feasibility – How realistically an initiative can be delivered given data, skills, technology and constraints. Prioritising initiatives weighs feasibility alongside business value, sustainability and strategic alignment (2.1.3). D2

Feedback loop – A mechanism that feeds real-world results back to improve an AI system or the decisions around it. Establishing continuous feedback mechanisms and success metrics (4.4.4) is a scaling discipline; it also underlies how models “improve” over time. D1 D4

Fine-tuning – Adapting a model by further training it on domain-specific examples so its responses fit a particular business need. Recognising when fine-tuning (versus RAG) helps (1.3.3) is the exam-relevant judgment; you do not perform the tuning. D1

Flex (Bedrock tier) – The Bedrock service tier priced at a 50% discount to Standard, trading some latency or priority for lower cost. It is a concrete cost-optimisation lever for non-urgent workloads. D2

Foundation model – A large, general-purpose model trained on broad data that can be adapted to many tasks, offered through Amazon Bedrock. Strategically it represents fast, managed access to capability without training a model from scratch. D1 D2

G

Governance board – A cross-functional body with clear accountability that oversees AI decisions, risk and compliance. Establishing governance structures with cross-functional representation (3.2.1) is an explicit skill; who sits on it matters as much as that it exists. D3

Governance by design – Integrating governance and responsible-AI considerations into project planning from the start, rather than bolting them on later. Recognising when to do this (3.1.3) is a maturity signal the exam rewards. D3

Governance (responsible-AI dimension) – One of AWS’s eight responsible-AI dimensions: the practices, roles and controls that keep AI use accountable across its lifecycle. It connects the responsible-AI principles to the governance structures of Domain 3. D3

GenAI (generative AI) – AI that creates new content — text, images, code — typically using foundation models. Distinguishing GenAI from the broader ML and AI (1.1.2), and knowing its failure modes (hallucination, token limits), is core Domain 1 material. D1

Ground Truth – The SageMaker AI capability for human feedback and data labelling. Business-relevant as the tooling behind human-in-the-loop labelling; you should recognise its purpose, not operate it. D3 D4

Grounding – Anchoring a model’s responses in trusted source data, typically via RAG, so answers reflect real information rather than invention. Grounding reduces hallucination risk and is the point of Bedrock Knowledge Bases and contextual grounding checks. D1 D3

Guardrail – Any safeguard that constrains AI behaviour to keep it safe, compliant and on-topic. On this exam “guardrail” appears both generically (a human-oversight safeguard, 3.1.4) and as the Bedrock Guardrails feature; know the concept first. D3

H

Hallucination – A confident but false or fabricated model output. Detecting and mitigating hallucinations (3.3.4) is a named reliability risk; contextual grounding checks and human oversight are the safeguards, and grounding is the prevention. D1 D3

Harmful content – Outputs that are offensive, dangerous or otherwise damaging. Managing harmful-content risk (3.3.3) is a governance skill; Bedrock Guardrails content filters are the archetypal control, blocking up to 88% of harmful content per AWS. D3

Historical data – Past data used to train a model so it can make predictions about new cases. Explaining training on historical data (1.1.5) is basic Domain 1 vocabulary; the exam links it to why data quality and representativeness matter. D1

Human-in-the-loop – A design where a human reviews, approves or corrects AI outputs before they take effect. It is the practical form of human oversight; the exam asks when it is required (3.1.4) and how roles shift toward oversight (4.3.6). D3 D4

Human oversight – Meaningful human control over AI systems, especially higher-risk or higher-autonomy ones. Recognising when systems require oversight and identifying safeguards (3.1.4) is a core Domain 3 skill, reinforced by the shared responsibility model. D3

I

Inference – The act of a trained model producing an output (a prediction or generation) from new input. It is basic vocabulary (1.1.1) and the thing most AI pricing meters — consumption-based pricing charges per inference request or token. D1 D2

Instance-based pricing – Paying per hour of provisioned compute, predictable but wasteful when the resource sits idle — as with an idle SageMaker AI endpoint. Distinguishing it from consumption-based and seat-based pricing (2.2.5) is exam-relevant. D2

Intangible benefit – A benefit that is real but hard to quantify directly, such as customer satisfaction or employee productivity. KPIs should capture both tangible and intangible benefits (2.2.1); ignoring intangibles understates AI’s value. D2

Intellectual-property (IP) risk – The risk that AI outputs infringe others’ IP, or that proprietary data leaks into a model. Managing IP concerns (3.3.3) is a governance duty; it shapes vendor terms, data handling and use-case approval. D3

Intelligent Prompt Routing – A Bedrock capability that routes requests within a model family by prompt complexity, priced per request; AWS states it can reduce cost by up to 30% without compromising accuracy. It is a concrete cost-optimisation lever. D2

ISO/IEC 23053 – The international framework for AI systems that use machine learning. Awareness of it (1.1.6) supports a unified AI vocabulary and standards-based governance conversations; know its purpose, not its clauses. D1 D3

ISO/IEC 42001 – The certifiable management-system standard for AI (an AI management system), which AWS has stated it supports. On the exam it is the standards anchor for governance and a shared vocabulary (1.1.6, 3.2). D1 D3

K

KPI (key performance indicator) – A metric chosen to track whether an AI initiative is achieving its business objective. Defining KPIs across tangible and intangible benefits (2.2.1) and tying them to a baseline is central to demonstrating value. D2

L

Labelled data – Data annotated with the correct answers, used to train supervised models; Ground Truth supports producing it. On this exam you recognise its role in training and data readiness — you do not label data yourself (that is out of scope). D1 D4

Lagging indicator – A metric that confirms an outcome after it has happened, such as realised cost savings or revenue. It pairs with leading indicators; lagging indicators prove value but arrive too late to steer a project. D2

Leading indicator – An early metric that predicts future success, such as adoption rate or pilot task completion. Identifying leading indicators (2.2.4) lets leaders course-correct before lagging results land. D2

Launch (AWS CAF phase) – The third AWS CAF phase, which delivers pilots in production and demonstrates incremental value. It sits between Align (finding gaps) and Scale (expanding what works). D4

Leadership alignment – Senior leaders sharing a common view of the AI strategy, its priorities and its trade-offs. It is a critical readiness dimension (4.1.1) and a prerequisite for effective sponsorship and change management. D4

M

Maturity model – A staged framework describing an organisation’s AI capability from experimentation to enterprise-scale deployment. Applying AI maturity models (4.1.2) lets you place an organisation and plan a progression pathway. D4

Minimum viable product (MVP) – The smallest version of a solution that delivers real value and can be validated with users. On this exam it sits between a proof of concept and production-grade in the pilot-to-scale journey. D2 D4

ML (machine learning) – A subset of AI in which systems learn patterns from data rather than following hand-written rules. Distinguishing ML from the broader AI and the narrower GenAI (1.1.2) is core Domain 1 vocabulary. D1

ML Governance (SageMaker) – SageMaker AI tooling — model cards, a model registry and dashboards — that supports governing custom ML models. Recognise it as the governance-conversation tooling for the custom-ML path. D3

Model – The trained artefact that turns input into predictions or generations. Explaining what a model is, in business terms (1.1.1), is baseline vocabulary; building or tuning one is out of scope. D1

Model adaptation – Techniques such as RAG and fine-tuning that make a general model perform better for a specific business need (1.3.3). The exam wants you to pick the right adaptation approach conceptually, not implement it. D1

Model drift – The degradation of a deployed model’s accuracy over time as the world moves away from its training data. Detecting and remediating model drift is a headline reason AI needs ongoing monitoring (1.2.3, 3.3.4). D1 D3

Model Evaluation (Bedrock) – A Bedrock capability offering algorithmic scores at no extra charge, plus human and LLM-as-a-judge evaluation options. It supports evidence-based model selection; know its purpose in a build-buy discussion. D2 D3

Model Monitor (SageMaker) – The SageMaker AI capability that detects and alerts on inaccurate predictions from deployed models. It is the concrete tooling behind “monitor for drift and performance changes” for the custom-ML path. D1 D3

Model unit – The unit in which Bedrock Provisioned Throughput is purchased, with no-commitment, one-month or six-month prices. Buying model units is the commit-for-a-discount lever — and a large fixed cost if utilisation is low. D2

N

NIST AI Risk Management Framework (NIST AI RMF) – A voluntary framework structured around four functions — Govern, Map, Measure, Manage — for managing AI risk. On this exam it is a risk-classification and governance reference to apply, not to recite verbatim. D3

NPV (net present value) – The value of an initiative’s future cash flows discounted to today, used to compare investments on a like-for-like basis. It sharpens a business case beyond simple ROI by accounting for the time value of money. D2

O

Orchestration – Coordinating multiple steps, tools or agents to complete a task. It is one of the agent capabilities to recognise (1.2.2); more orchestration means more moving parts to govern and monitor. D1 D3

P

Payback period – The time it takes for an initiative’s cumulative benefits to cover its costs. A short payback strengthens a business case and lowers risk; it is a common ROI-adjacent metric in Domain 2. D2

Pilot – A limited, real deployment used to test value and learn before scaling. Moving from pilot to production-grade (4.4.5) — and deciding to scale, pause or terminate a pilot (2.1.3) — is a recurring exam theme. D2 D4

Prediction – A model’s output about a new case, such as a forecast or classification. It is basic vocabulary (1.1.1); the exam links prediction quality to data quality and drift. D1

Pricing structures – The three commercial models the guide asks you to distinguish: consumption-based, instance-based and seat-based, plus commitment-based discounts. Matching the pricing structure to the workload is a Domain 2 cost skill. D2

Priority (Bedrock tier) – The Bedrock service tier priced at a 75% premium to Standard, buying higher priority or lower latency at higher cost. It is the “pay more for speed” end of the tier trade-off. D2

Privacy and security – One of AWS’s eight responsible-AI dimensions: protecting personal and sensitive data and the system itself. Applying it to business scenarios (3.1.1) and identifying data-security measures (3.2.3) are governance skills. D3

Production-grade – A system reliable, governed and operable enough for real business dependence, as opposed to an experiment. Addressing the transition from experimental to production-grade (4.4.5), including governance and operational requirements, is a scaling skill. D4

Proof of concept (POC) – A quick test that an approach is technically or practically viable, earlier and lighter than a pilot. POC programmes also appear as a workforce-development approach to build AI literacy (4.3.5). D2 D4

Prompt engineering – Crafting inputs to steer a model toward better outputs. Applying basic prompt-engineering principles (1.3.1) is in scope at a conceptual level; the exam does not test advanced prompt syntax. D1

Prompt Optimization (Bedrock) – A Bedrock capability that rewrites prompts for better results, priced per token processed. Recognise it as a managed way to improve GenAI performance without prompt-engineering expertise on staff. D1 D2

Provisioned Throughput – A Bedrock purchasing option bought in model units with no-commitment, one-month or six-month commitments — a discount for committed capacity, but a fixed cost that is wasted if utilisation is low. It is a cost-planning trade-off. D2

Q

Quick (Amazon Quick) – See Amazon Quick. The AI-powered business assistant and BI surface that lets business users get insights and build no-code apps over their data on a seat-based model. D2 D4

R

RACI – A responsibility-assignment scheme naming who is Responsible, Accountable, Consulted and Informed for a decision or task. It is a practical tool for the clear accountability that AI governance structures require (3.2.1). D3 D4

RAG (retrieval-augmented generation) – A model-adaptation technique that retrieves relevant source data at query time and feeds it to the model, grounding responses in your own information. Recognising when RAG (versus fine-tuning) fits a need (1.3.3) is the exam judgment; Bedrock Knowledge Bases deliver it as a managed service. D1 D3

Readiness assessment – A structured evaluation of whether an organisation is prepared for AI across leadership alignment, data quality, cultural preparedness, technical infrastructure and governance (4.1.1). It is the entry point to Domain 4. D4

Reserved (Bedrock tier) – The Bedrock service tier for reserved capacity, a commitment-based option for predictable, sustained workloads. It sits alongside Standard, Flex and Priority in the tier trade-off. D2

Responsible AI – The practice of building and using AI that is fair, safe, transparent and accountable. AWS publishes eight dimensions; applying responsible-AI principles to business decisions (3.1) is the heart of Domain 3, including navigating trade-offs when they conflict with business goals. D3

Responsible AI dimensions (AWS’s eight) – AWS’s published set: fairness, explainability, privacy and security, safety, controllability, veracity and robustness, governance, transparency. Learn all eight; note the exam guide’s shorter list (fairness, explainability, privacy, safety, transparency, robustness) is a subset. D3

Risk register – A living record of identified AI risks with their severity, owners and mitigations. It operationalises governance and supports directing mitigation strategies (3.3); a governance board without a risk register is incomplete. D3

Risk tier / risk classification – Grouping AI uses by risk level so governance effort matches risk, as in the EU AI Act’s tiers. Applying a risk-classification framework to prioritise governance across the lifecycle (3.2.4) is an explicit skill. D3

ROI (return on investment) – The ratio of net benefit to cost, the headline number in most AI business cases. Calculating ROI comprehensively — time savings, cost reduction, revenue growth and productivity gains (2.2.3) — against a baseline is a core Domain 2 skill. D2

Rule-based automation – Automation that follows explicit, hand-written rules rather than learning from data. Determining when to use rule-based automation instead of AI (1.2.1) is a key discriminator: stable, well-defined, high-stakes rules often beat AI. D1 D2

S

Safety – One of AWS’s eight responsible-AI dimensions: preventing harm to people and systems. It drives guardrails, human oversight and escalation criteria in Domain 3 scenarios. D3

SageMaker Clarify – The SageMaker AI capability that detects bias across data preparation, after training and in the deployed model, and supports explainability. It is the tooling that makes fairness and explainability actionable for the custom-ML path. D3

Savings Plans – An AWS commitment-based discount mechanism (commit to a level of usage for a lower rate) named as a cost-optimisation strategy. Recognise it as a lever that trades flexibility for lower cost. D2

Scale (AWS CAF phase) – The fourth AWS CAF phase, which expands production pilots and their business value to the desired scale. It is the enterprise-wide end of the Envision → Align → Launch → Scale journey. D4

Scale, pause or terminate – The three outcomes of prioritising an AI initiative on business value, feasibility, sustainability and strategic alignment (2.1.3). The exam rewards the discipline to pause or terminate, not only to scale. D2

Seat-based pricing – Paying per user per month, predictable and decoupled from usage volume, as with Amazon Quick. Distinguishing it from consumption-based and instance-based pricing (2.2.5) matters because it changes how cost scales with adoption. D2

Sensitive-information filters – A Bedrock Guardrails control that detects and redacts personally identifiable information. Know it as the privacy-and-security safeguard among the guardrail types (3.1.4, 3.2.3), at the level of what it does. D3

Shadow AI – Unsanctioned use of AI tools outside official approval and oversight. Establishing a transparent classification of tools (approved, blocked, under evaluation) mitigates shadow-AI risk (1.2.4); it is both a governance and a literacy concern. D1 D3

Shared responsibility model (for AI) – AWS’s division of security duties: AWS secures the cloud (infrastructure, managed-service operation) while the customer is responsible in the cloud — their data, access, use-case appropriateness, human oversight and business-process compliance. The line moves for managed services but never removes customer accountability for what the system is used for and whether its output is fit for purpose. D3

Standard (Bedrock tier) – The baseline Bedrock service tier against which Flex (a 50% discount) and Priority (a 75% premium) are defined. It is the reference point for tier trade-offs. D2

Strategic alignment – How well an initiative supports the organisation’s actual strategy and objectives. It is one of the four prioritisation criteria (2.1.3) and a readiness dimension; misaligned but flashy projects are a classic trap. D2 D4

Structured data – Data organised in a defined schema, such as rows and columns in a table. Distinguishing structured from unstructured data and explaining why the type matters for AI (1.1.3) is core Domain 1 vocabulary. D1

Sustainability (of an initiative) – Whether an AI initiative can be maintained and keep delivering value over time, given cost, data and operational demands. It is one of the four prioritisation criteria (2.1.3). D2

T

Tangible benefit – A benefit that can be measured directly in money or units, such as cost reduction or revenue growth. KPIs must capture tangible benefits alongside intangible ones (2.2.1); tangible benefits usually anchor the ROI calculation. D2

TCO (total cost of ownership) – The full lifetime cost of an AI solution — licensing, compute, integration, people, monitoring and change — not just the sticker price. A sound business case compares options on TCO, not headline cost. D2

Token – The unit of text a model processes; consumption-based pricing often meters tokens, and the context window is measured in them. Recognising when token limits affect GenAI performance (1.3.2) is in scope. D1 D2

Tool classification – Categorising AI tools as approved, blocked or under evaluation so use is transparent and governed. It is the concrete mechanism for mitigating shadow-AI risk (1.2.4). D1 D3

Tool use – An agent’s ability to call external tools or systems to accomplish a task. It is one of the four agent capabilities to recognise (1.2.2); tool use expands both capability and the governance surface. D1 D3

Training – The process of teaching a model from data, typically historical data, so it can make predictions. It is basic vocabulary (1.1.1, 1.1.5); the exam links training data quality and representativeness to outcomes and bias. D1

Transformation domains (AWS CAF) – The four domains AWS CAF treats as a value chain: Technology → Process → Organization → Product. They frame what an AI transformation changes; know them as the CAF value chain. D4

Transparency – One of AWS’s eight responsible-AI dimensions: being open about how AI is used and how it works, supported by AI Service Cards. It builds trust and supports compliance and human oversight. D3

U

Unit economics – The profit or cost of a single unit of activity (per transaction, per customer, per request). Understanding unit economics keeps an AI business case honest as usage scales, especially under consumption-based pricing. D2

Unstructured data – Data without a fixed schema — free text, images, audio, video. Distinguishing it from structured data and explaining why the type matters for AI (1.1.3) is core Domain 1 vocabulary; much GenAI value comes from unstructured data. D1

V

Veracity and robustness – One of AWS’s eight responsible-AI dimensions: producing accurate outputs and performing reliably under varied or adverse conditions. It connects directly to reliability risks such as hallucination and drift (3.3.4). D3

Vendor risk – The risk introduced by relying on a third-party AI provider — lock-in, viability, security, compliance and terms. Evaluating vendor proposals within a build-buy-partner decision (2.1.2) requires weighing vendor risk explicitly. D2 D3

W

Well-Architected Responsible AI Lens – The AWS Well-Architected Framework lens providing responsible-AI best practices across the design, development and operation of AI workloads. On this exam it is the AWS-native reference for embedding responsible AI, at a strategic level. D3

Word filters – A Bedrock Guardrails control that blocks specified words or phrases, free of charge. Know it as one of the lightweight safeguard types for keeping outputs on-policy. D3

Workforce development – Approaches that accelerate AI literacy across an organisation — POC programmes, hackathons, training programmes and responsible-AI training (4.3.5). It is how leadership turns strategy into capability. D4

Frameworks and phase vocabulary

AWS Cloud Adoption Framework (AWS CAF) – AWS’s framework for guiding cloud and AI transformation, organised into six perspectives, four transformation domains and four phases, with a dedicated CAF for AI, ML and Generative AI whitepaper. On this exam it is the structuring framework for readiness, capability gaps and scaling. D4

CAF perspectives (the six) – The lenses AWS CAF uses to assess capability: Business, People, Governance, Platform, Security, Operations. Identifying capability gaps across these perspectives is an Align-phase and readiness activity (4.1.3). D4

CAF phases (the four) – Envision → Align → Launch → Scale. Envision identifies and prioritises opportunities against business objectives; Align finds capability gaps and stakeholder concerns; Launch delivers pilots in production; Scale expands them to the desired scale. Note the exam guide’s loose wording “envision, experiment, launch, scale” for task 4.4.1 — treat it as loose, not a different framework. D4

Cost and tooling references

AWS Cost Explorer – The AWS tool for viewing actual AI spend, trends and anomalies after the fact. Distinguish it from the Pricing Calculator (forward estimates); Cost Explorer is where you investigate what you have already spent. D2

AWS Marketplace – The AWS catalogue for finding and procuring third-party solutions, useful for evaluating buy-and-partner options against building in-house within a build-buy-partner decision. D2

AWS Pricing Calculator – The AWS tool for forward cost estimates before committing. Distinguish it from Cost Explorer (actuals); the Calculator informs the business case before spend begins. D2

Exam-mechanics vocabulary

Beta exam – An exam whose items are still being validated, with results timing that differs from standard exams and no official practice exam available. AIB-C01 is a beta exam; earning the certification by 15 February 2027 also earns an Early Adopter badge. D1 D2 D3 D4

Multiple response – An item type with two or more correct answers among five or more options, where you must select all correct responses to earn credit. On our mocks these are labelled “Select two”; there is no penalty for guessing and unanswered items score as incorrect. D1 D2 D3 D4

Scaled score – AIB-C01 reports a scaled score from 100 to 1,000 with a minimum passing score of 700 and a pass/fail designation. A scaled score is not a percentage: 700 does not mean “70% correct”, and AWS does not publish the mapping — our mocks show a raw percentage plus an indicative scaled figure as a rough guide only. D1 D2 D3 D4