Prepare and test your skills
Prepare and test your skills
Data Readiness and Infrastructure Foundation is the combination of trusted data and scalable cloud services that allows generative AI (opens in a new tab) to create reliable business value. Data readiness means organizational data is accurate, secure, and accessible to approved teams. Infrastructure foundation means compute, storage, and pre-trained models are ready so teams can test and deploy without building everything from scratch.
Assessing data quality, security, and accessibility is the review that decides if organizational data can safely power AI. Quality checks confirm that data is accurate, complete, and relevant for training models and grounding their outputs. Security checks confirm that access controls, privacy protections, and industry and country regulatory requirements are met. Accessibility checks confirm that approved data scientists and practitioners can discover and reuse data instead of repeating engineering work.
Implementing data governance with Google Cloud is a structured way to manage data through its lifecycle with defined policies, procedures, and roles. A governance framework improves data quality, strengthens security, improves utilization, and reduces redundancy across AI workloads. Google Cloud applies the framework without requiring all datasets to be consolidated in one place, so distributed data can be unified, discovered, and shared securely.
The catalog depends on access controls to make discovery safe, and sharing depends on the catalog to make curated assets findable before they are exchanged.
Making datasets and features reusable across the ML lifecycle means publishing data once so many projects can discover and use it. Reuse avoids redundant data engineering and feature development, which speeds up model development and lowers infrastructure costs. Google Cloud stores each data type where it can be governed and shared while keeping links between data, features, and embeddings.
Leveraging Google Cloud infrastructure and pre-trained models means using managed compute and ready-made models instead of building all capability from scratch. Vertex AI (opens in a new tab) hosts Google foundation models such as Gemini and Imagen and provides Vertex AI Model Garden for select open-source and third-party models, including Hugging Face text generation models. Scalable training and serving run on Google Kubernetes Engine (GKE) as a customizable AI platform and on Compute Engine virtual machines with attached GPUs for accelerated workloads. The choice of model approach depends on task type and training data availability:
| Approach | Best for | Data requirement |
|---|---|---|
| Pre-trained traditional AI model such as Document AI, Vision AI, Natural Language API, and Video Intelligence API | Classification or detection when an existing API meets requirements | No custom training data needed |
| Traditional predictive AI | Forecasting and classifying from historical patterns, especially effective with structured tabular data | Requires sufficient historical training data for custom training |
| Generative AI model such as Gemini | Summarization, content generation, advanced transcription, and multimodal inputs across text, images, video, and audio | Works without a large training dataset using prompt engineering, with optional tuning on data examples |
Exam tip: When no training dataset is available or datasets are too small for custom training, use generative AI models with prompt engineering rather than custom training a traditional model.
Responsible AI Governance and Change Management is the organizational foundation for integrating generative AI into workflows. Governance means AI systems run on data that is secure, governed, and high quality, with clear control over who can access data and workloads. Change management means planning how people, identities, and operating practices move to the cloud so adoption succeeds and daily work becomes more efficient.
Governed data pipelines carry data from raw sources to AI-driven actions under security and policy controls. A borderless open data lakehouse establishes a highly governed, secure pipeline from raw borderless silos to AI and agentic driven actions, so fragmented data becomes usable for analysis and agents without losing oversight. Secure data environments in Google Cloud secure sensitive datasets against accidental exposure and malicious exfiltration, which places protection on each side of the trust boundary between sensitive data on one side and users, applications, and agents on the other. Together, the governed pipeline depends on the secure environment because data moves forward only when exposure and exfiltration controls are in place.
Quality and operational governance with MLOps applies consistent testing, monitoring, and updates as AI moves from experiment to production. Machine learning operations (MLOps) (opens in a new tab) connects model development to deployment and operation, so the same practices cover classification models, generative AI, and agentic AI systems. Google Cloud supports this control with best practices for implementing machine learning on Google Cloud and with guidelines to help assess, ensure, and control quality in machine learning solutions. When MLOps owns the lifecycle, quality checks apply to data, models, and operations before AI outputs reach business workflows.
Google Cloud organizes this planning knowledge into four peer resources that teams use at different stages:
Exam tip: A design guide gives prescriptive, cross-product planning guidance, while a reference architecture gives a detailed example for a specific workload.
Managing organizational change for adoption means preparing identities, accounts, and landing zones before scaling generative AI. A Google Cloud landing zone (opens in a new tab) guides teams through high-level decisions about identity onboarding, resource hierarchy, network design, and security, so business transformation does not move forward without ownership and boundaries. Cloud Identity and Google Workspace provide the managed accounts for workforce users, and teams assess existing user accounts when planning to manage identities across an organization. Onboarding then brings both existing identities and new identities into Cloud Identity or Google Workspace, which lets the organization consolidate accounts and apply identity and access management (IAM) consistently.
Change follows an ordered migration path from current operations to cloud operations:
This order matters because design depends on the plan, and implementation depends on the design. When identity onboarding, account planning, and landing zone decisions come first, workforce users can authenticate and consume services in a consistent way.
Strategic use case identification and value assessment is the practice of finding generative AI opportunities that serve measurable business goals. A use case is only worth pursuing when it supports a defined business outcome, such as higher employee efficiency, lower cost, or improved customer satisfaction, rather than existing as a standalone technology experiment. Each candidate is judged by feasibility, alignment with executive priorities, and expected value.
The business value-driven decision process is the five-step method that starts from the business outcome and works backward to the technology choice. Business use cases (opens in a new tab) are defined by measurable goals, user expectations, and required process changes before a model is selected.
Choosing between generative AI and traditional AI means matching the problem type to the model type that fits it best. Traditional AI learns from historical data to classify information or predict future outcomes, while generative AI creates summaries, uncovers hidden correlations, and generates new content such as text, images, or video.
A decision tree that branches on classification, predictive, and generative tasks plus data availability and control needs to choose pre-trained, traditional, or generative AI.
| Type | Best-fit tasks | Example use cases |
|---|---|---|
| Traditional AI | Classification, regression, time series forecasting, clustering on structured or image data | Spam filtering, defect detection in manufacturing, house-price prediction, sales forecasting, customer segmentation |
| Generative AI | Content creation, conversational search, document understanding, code generation, multi-modal input | Marketing content, chatbots, summarization, translation, captioning, developer assistance |
For classification or detection, first check whether a pre-trained traditional AI model such as Document AI, Vision AI, Natural Language API, or Video Intelligence API meets the requirement. If not, and enough training data exists, choose between custom training when high control over algorithms, loss functions, explainability, and hyperparameters matters and generative AI when faster go-to-market matters. If training data is missing or too small, use generative AI models with prompt engineering, optionally refined through supervised tuning with data examples.
Exam tip: Model selection is not a binary choice — when a pre-trained model fits, use it; when training data is scarce, generative AI with prompt engineering is the recommended path.
Combining generative AI with traditional AI means using both model types together to reach a business goal. A common pattern feeds the output of a traditional AI model into the prompt of a generative AI model. For example, traditional predictive AI can forecast customer churn probability, and a generative AI chatbot can let the sales team explore those predictions in natural language. Traditional AI can also identify customer segments that generative AI then uses to produce personalized marketing content, or forecast risks that generative AI turns into simulated mitigation scenarios. Computer vision can detect and classify sign language into text, while generative AI adds context and generates voice output for two-way communication.
Stakeholder roles in use case selection describe who shapes the decision about which model to use. Although data scientists and machine learning engineers commonly lead model selection, several other stakeholders shape the decision.
Example use cases in Google Cloud show how data services, retrieval, and Gemini models connect into working applications. A personalized marketing campaign use case ingests user data into BigQuery, derives insights with Dataflow, triggers a Cloud Run service through Eventarc, and prompts a Gemini (opens in a new tab) model to generate audio, video, and text assets stored in Cloud Storage. A customer-support use case uses retrieval-augmented generation (RAG): a knowledge retriever service asks Gemini to identify relevant resources from a Cloud Storage knowledge base, then a solution generator service prompts Gemini to produce step-by-step answers. A product recommendation use case processes clickstream data through Dataflow into BigQuery, then a recommender service performs a vector similarity search and prompts Gemini to generate tailored recommendations for the storefront. Each example can be improved with a feedback loop and, where brand or safety standards matter, a human review step before content is served.
Traditional AI learns from historical data to classify information or predict future outcomes and fits classification, regression, time series forecasting and clustering, while generative AI creates summaries, uncovers hidden correlations and generates new content such as text, images or video and fits content creation, conversational search, document understanding, code generation and multimodal input. For classification or detection, first check whether a pre-trained traditional AI model such as Document AI, Vision AI, Natural Language API or Video Intelligence API meets the requirement. If training data is missing or too small, use generative AI models with prompt engineering, optionally refined through supervised tuning with data examples.
Use generative AI models with prompt engineering when no training dataset is available or datasets are too small for custom training. Traditional predictive AI requires sufficient historical training data for custom training, especially for forecasting and classifying from historical patterns with structured tabular data, while generative AI such as Gemini works without a large training dataset using prompt engineering for summarization, content generation, advanced transcription and multimodal inputs across text, images, video and audio. If enough training data exists, choose custom training when high control over algorithms, loss functions, explainability and hyperparameters matters and generative AI when faster go-to-market matters.
Knowledge Catalog and BigQuery sharing support data governance and reuse by organizing, securing and exchanging distributed data without requiring all datasets to be consolidated in one place. Knowledge Catalog organizes distributed data into logical data lakes and zones, controls access with Google Groups and Knowledge Catalog roles, streamlines quality checks with auto data quality, ingests technical metadata automatically and makes assets findable through search for users with sufficient permissions. BigQuery sharing exchanges curated data assets across organizations through data exchanges viewed as listings, protects sensitive collaboration with data clean rooms and grants view and publish rights for data scientists and practitioners, and sharing depends on the catalog to make curated assets findable before they are exchanged.
An enterprise has rolled out generative AI tools across its business departments, but employee adoption has stalled at low engagement levels due to confusion around ethical usage and a lack of practical skills. Leadership wants to execute an effective organizational change management and governance strategy to ensure widespread adoption while adhering to responsible AI principles.
Which strategy should the organization implement to achieve successful AI adoption and governance?
Mandate immediate tool usage across all departments through strict performance metrics without formal training or governance policies.
Automate all internal business decisions using autonomous AI agents to bypass the need for human review and employee training.
Establish an enterprise-wide workforce upskilling program paired with human-centered responsible AI guidelines and ethical review practices.
Restrict generative AI access exclusively to specialized software engineering teams to minimize organizational compliance risks.
Last updated