EXIN BCS Gen AI Award study plan

Define generative AI and distinguish it from discriminative and classical machine learning using at least five concrete contrasts you can state from memory

Explain how transformer architectures, tokenisation, embeddings and next-token prediction produce output, and describe the role of training data, fine-tuning and system prompts

Classify at least eight everyday generative AI use cases by modality (text, image, audio, code, multimodal) and by task type (summarise, draft, classify, translate, generate)

Identify the main categories of generative AI capability limits: hallucination, knowledge cut-off, context window, reasoning gaps, arithmetic weakness and non-determinism

Diagnose why a given generative AI output is wrong by attributing the failure to data, prompt, model or task-fit causes

Compare model families and deployment options (hosted API, private tenant, on-premise, open-weight) on cost, latency, privacy and control

Describe evaluation basics: golden test sets, human review, automated scoring and when each is appropriate

Explain how retrieval augmentation and tool use change the reliability profile of a generative AI system

Write prompts that specify role, task, context, constraints, output format and audience for five different business tasks

Apply decomposition and step-by-step instruction to break a complex task into a reliable prompt chain

Use few-shot examples and structured output requests (tables, JSON, bullet schemas) to make results consistent across repeated runs

Set and use system-level instructions and reusable prompt templates that separate fixed rules from variable inputs

Iterate on a weak prompt through at least three revision cycles and document what each change improved

Test a single prompt across two or three different models or settings and explain the differences in output quality

Apply grounding techniques: supplying source text, asking for citations to that text and instructing the model to say when it does not know

Recognise and correct prompt anti-patterns such as vague verbs, conflicting constraints, overloaded context and unstated output format

Map the main legal exposures of generative AI use: copyright in inputs and outputs, licensing of training data, confidentiality and contractual terms of use

Explain transparency and disclosure obligations, including labelling AI-generated content and the limits of AI detection tools

Identify bias, fairness and accessibility risks in generative AI outputs and describe a review step that reduces them

Describe data protection principles that apply when personal data enters a prompt, including minimisation, lawful basis and cross-border transfer

Explain prompt injection, data leakage, insecure output handling and model poisoning, and state one concrete control for each

Apply a data classification rule to decide what may never be pasted into a public generative AI tool

Draft acceptable-use guidance covering permitted tools, prohibited data, review requirements and accountability for final output

Describe human-in-the-loop and audit-trail requirements for high-impact generative AI decisions

Identify and size generative AI opportunities in a chosen business process using a simple value-versus-feasibility screen

Estimate the cost of a generative AI solution including model usage, integration, review labour and ongoing maintenance

Define success metrics that separate adoption measures from outcome measures such as cycle time, rework rate and quality score

Describe how to run a small pilot with a baseline, a control group and a decision gate before scaling

Consolidate all four domains into a single revision sheet of key definitions, risks and prompt techniques

Answer timed practice questions across all domains and review every wrong answer to its underlying concept

Rehearse explaining a generative AI use case to a non-technical stakeholder in under three minutes, covering value, risk and controls

Close remaining knowledge gaps by re-testing the two weakest areas identified in your practice run