AI Strategy and Business Value Creation

Tying use cases to outcomes

AI initiatives start not from the technology but from the business problems of each department. For example, in customer service, you shorten handling time with automated responses to inquiries, and in software development, you shorten the development period with code generation; in this way, you tie AI capabilities to measurable business outcomes. A use case whose outcome cannot be stated cannot show its effect, and it also becomes hard to keep its budget.

→ Working Backwards from the Customer

Build, buy, or partner decisions

For work that does not need differentiation and that you want to adopt quickly, the basic approach is to buy off-the-shelf products or products from AWS Marketplace and use managed services. Partner when expertise is lacking, the timeline is limited, or you want to transfer skills into the organization. Build your own on top of Bedrock or SageMaker AI when you need core differentiation, proprietary data, or compliance with strict regulations.

→ Procurement on AWS Marketplace / → Amazon Bedrock Glossary

Prioritizing initiatives and the portfolio

Compare multiple AI initiatives from the viewpoints of business value, feasibility, sustainability, and strategic alignment, and rank them. Even after they start, evaluate the results regularly: scale initiatives that are producing results, pause those whose assumptions have broken down, and end those with no prospects. Rather than continuing every initiative that has started, it is important to set the criteria for these decisions in advance.

Situations where AI is not suitable

AI is not the answer for processes that can be clearly solved with rules, processes with no data available for training or reference, processes where the cost of AI exceeds the benefit obtained, and processes where no errors at all are acceptable. The exam scope includes questions in which the option of not using AI is the correct answer. The goal is not to use AI but to solve business problems with the best means.

KPIs and technical metrics

AI KPIs are defined in terms of both tangible benefits, such as cost reduction and revenue growth, and intangible benefits, such as customer satisfaction and employee productivity. Model accuracy and F1 score are technical metrics and are not business outcomes in themselves. Translate them into how improved accuracy leads to shorter handling times or fewer cancellations, and measure that.

Baselines and ROI

To show the effect of AI, measure baselines such as current processing time and cost before adoption, and compare them with the values after adoption. ROI is calculated by subtracting the cost from the value of the benefits obtained and dividing by the cost, and the cost includes not only inference usage fees but also operating costs such as monitoring, retraining, and governance. Time saved becomes a benefit only when it is redirected to other valuable work.

Leading indicators

Lagging indicators such as revenue and ROI take time to show results, so leading indicators are used to foresee success or failure at an early stage. Typical examples are usage rate (adoption rate), usage frequency, satisfaction, and data readiness. If the leading indicators are not growing, you can revisit the initiative without waiting for the lagging indicators.

Pricing models and cost management

Consumption-based pricing where you pay only for what you use, such as Bedrock on-demand, suits stages with large fluctuations in usage and PoCs. As of October 2026, Bedrock's pay-for-what-you-use pricing has Standard / Flex / Priority tiers, so you can choose the lower-cost Flex for processing that can tolerate delays and Priority for processing that needs fast responses. Once volume becomes stable and predictable, lower costs with instance-based pricing, Provisioned Throughput, or the Reserved tier, which reserves throughput in tokens per minute, together with long-term commitments, and send large volumes of processing that need not be immediate to discounted batch inference. If you want to forecast costs by the number of users, use per-seat pricing such as Amazon Quick; the basic approach is to start small, measure actual usage, and then move to long-term commitments.

→ AWS Cost Explorer Glossary

What Savings Plans cover

Savings Plans are a mechanism that discounts prices in exchange for committing to a certain amount of spend (an hourly amount) for a 1-year or 3-year term, and they can lower the cost of long-term use of SageMaker AI. As of October 2026, Bedrock is not covered by Savings Plans, and the basic ways to lower Bedrock costs are batch inference, the Flex tier, and Provisioned Throughput or the Reserved tier. The exam tests whether you can distinguish which means lowers the cost of which service.

→ AWS Savings Plans Glossary

Sustainable competitive advantage and the level of investment

Simply using foundation models that anyone can use lets competitors quickly do the same. A sustainable advantage comes from data only your company has, deep integration into business workflows, and mechanisms that improve the more they are used. Set the level of investment by looking at how widely AI has spread in the industry and at competitors' moves, and avoid committing to large facilities or long-term contracts from the start.