[NEW] AWS Certified AI Business Strategist AIB-C01 Training

Pick up the basics AI projects assume: renting from a cloud provider, who builds a foundation model, how software calls it, and what a framework is. Then see why an AI solution is more than the model.

Tell an algorithm from a model, and training from inference, the two costs inside every AI proposal. Then plan for wrong predictions, and place generative AI inside machine learning.

See why a foundation model turns an AI build into a buying decision. Then choose between Amazon Bedrock and Amazon SageMaker AI by how much of the solution you want to own.

Sort structured data from unstructured data, and match each shape to the kind of AI that fits it. Then spot the data quality failures, from missing values to coverage gaps, that a model repeats at scale.

Understand why supervised learning cannot predict an outcome that your company never recorded, why live data drifts away from the training data, and how bias can fail one group. Then have a baseline ready before monitoring starts.

Separate three ISO standards for AI: ISO/IEC 23053 describes an AI system, ISO/IEC 42001 governs how an organization runs AI, and ISO/IEC 22989 supplies the AI terms. Then ask a vendor to name the standard it follows.

Take a first look at Amazon Bedrock: its foundation models, what you can build with it, and how on-demand and Provisioned Throughput pricing differ.

Decide when rule-based automation fits better than a model, and when a model costs more than the benefit it returns. Then evaluate what wrong predictions cost, and plan how your team handles the errors.

Spot the difference between a chatbot and an AI agent: an agent works in a loop, and the model decides the next step. Then set how much autonomy an agent gets, and see where tool use and orchestration fit.

Build a prompt from task, context and output format, and add worked examples with few shot prompting when a plain instruction fails. See why a model remembers nothing between requests, and why long chats cost more.

Work with tokens, the unit a model reads and writes, and fit a long document inside the context window. Then use the length of prompts and answers to cut both the cost and the wait.

Compare Retrieval Augmented Generation with fine tuning for putting company knowledge into answers. Then follow the order that tries prompt engineering first and fine tuning last, because each step costs more.

Run a prompt in the Amazon Bedrock chat and text playground, adjust the model settings, and compare two models side by side. Then read the token counts, and create an image with Amazon Nova Canvas.

Upload a fictional company's PDF to the chat with your document feature in Amazon Bedrock, then ask it questions. Check each answer against the file, and watch retrieval work on facts the model never trained on.

Search for high-impact AI use cases across business functions, and write each use case as an AI capability plus a business result. Then pick the category of AI tool that fits your question.

Weigh what building, buying and partnering each trade, and compare a vendor proposal against an internal build. Meet Amazon Quick and AWS Marketplace, and see why buying never moves accountability.

Write a KPI that names a business result rather than model accuracy, and state tangible and intangible benefits together. Capture a baseline first, then measure the later number the same way.

Score AI initiatives on business value, feasibility, sustainability and strategic alignment, then re-score them to scale, pause or terminate. Place that work in Envision, the first phase of the AWS Cloud Adoption Framework.

Answer four questions before you set a switch date, and keep the old path available until the AI path proves itself on real volume. Then test your prompts, vector store, evaluations and integrations again on the new platform.

Turn time savings, cost reduction, revenue growth and productivity gains into money, and avoid counting saved hours that never reach the accounts. Then track leading indicators that move long before ROI does.

Load the cost side with build or license, integration, change management and ongoing run cost. Recognize consumption based, instance based and seat based charges, and estimate a solution in AWS Pricing Calculator.

Read actual spend in AWS Cost Explorer, and see that Amazon SageMaker AI is on the Savings Plans eligible services list and Amazon Bedrock is not. Then compare batch inference, the Flex service tier and Provisioned Throughput.

Measure the company against what rivals ship and what customers now expect, and judge when an advantage turns into the standard. Then size defensive investment against offensive investment.

Contrast transforming a business model with speeding up a process, and place each in the AWS Cloud Adoption Framework transformation domains. Then see why data, processes and customers outlast any model.

Walk through the Amazon Bedrock Infer menu, and see how a batch inference job reads many prompts from Amazon S3 and writes the results back. Use it when volume is high and answers can wait.

Reserve capacity with Provisioned Throughput in Amazon Bedrock: pick a model, the model units and a commitment term. Then see how custom model on demand serves a fine tuned model, paid per request.

Go from a list of six responsible AI dimensions to the ten in the AWS Well-Architected Responsible AI Lens. Then tell fairness, explainability, and transparency apart, and separate veracity from robustness.

Settle competing objectives at design time and see why responsible AI work starts in planning. Then write release criteria as pass-or-fail tests, each with a metric and a threshold.

Place human review where quality is hard to judge or the decision is consequential, and write escalation criteria a system can test. Then staff an AI governance group and record what each group decides.

Name the countries a system will operate in before you fix the design, and see that the rules of a regulated process can still apply. Then manage intellectual property flowing into the model and back out.

Combine likelihood with severity into one risk level, and keep a risk registry across every AI use case. Then attach the rating to the model with Amazon SageMaker Model Cards.

Cut shadow AI with a published list of approved tools, and answer who may call the model, what data it reaches and who reads the outputs. Then switch on Amazon Bedrock model invocation logging.

Configure Amazon Bedrock Guardrails with content filters, denied topics, word filters, sensitive information filters and hallucination detection. Then score a model with Amazon Bedrock evaluations before it carries any work.

Watch data quality, output quality, bias and feature attribution drift against a recorded baseline, and see why a fair model can turn unfair. Then route monitoring alerts to a person with the authority to act.

Create two guardrails in Amazon Bedrock: one blocks a denied topic on conspiracy theories, and one masks names and email addresses. Test each one in the console and watch it block or mask the content.

Tour Amazon Bedrock evaluations: automatic scoring, a second model as judge, and human reviewers. Then see how a retrieval evaluation checks both what a knowledge base finds and what the model writes.

Explore the Configure and learn area of the Amazon Bedrock console, where access policies limit which teams can use which models. Then turn on model invocation logging and see what it records.

Assess readiness across the six perspectives of the AWS Cloud Adoption Framework, and follow its four phases: envision, align, launch and scale. Then see what the align phase produces before any pilot starts.

Locate the phase a company works in, and read maturity for each capability, not only for the whole company. Then rank the capability gaps worth closing, and sequence them so each layer enables the next.

Ask whether data is both good and reachable, and meet the product prerequisites behind accessibility. Then give every dataset a named owner, catalog data as a product, and let it move through a modern data architecture.

Clarify what executive sponsors are for, and find AI champions instead of appointing them. Then run change acceleration as a program, and answer the question staff really ask: what changes in my role?

Identify the cultural barriers AWS names, and close them with fluency training and agreed interface words. Then choose the right development format, and prepare roles that move from producing output to judging it.

Draw the line between launch and scale, and pick a first win for what it proves. Then learn the role of an AI center of excellence, and expect workloads to change as volume grows.

Close the loop with user feedback that changes the system, and add automated deployment, monitoring, retraining and rollback when the company builds its own model. Then plan for incidents instead of assuming none will come.