AI value the wrong way. Instead of asking “What new capabilities does this unlock?”, the conversation quickly turns into questions such as: How many hours can we save? How many people could this replace? While efficiency is an important source of AI value, it is only part of the picture. Many successful AI systems do not primarily replace human work (and those that do are likely to trigger resistance rather than enthusiasm). Instead, they upgrade existing workflows, amplify human capabilities, or enable entirely new business opportunities. For example, a customer support copilot may not reduce headcount, yet it can dramatically improve resolution quality and customer experience. Trying to evaluate it through the efficiency lens alone is a non-starter.
This article analyzes value creation across three types of AI opportunities:
- Automation: AI replaces operational tasks previously performed by humans.
- Augmentation: AI supports humans in performing complex tasks and making better decisions.
- Innovation: AI enables new capabilities, products, or operating models.
Looking across more than 200 AI use cases collected in our AI Radar, AI value appears across nine performance areas which can be grouped into three categories: process improvements, capability improvements, and financial outcomes (cf. Table 1). Timing matters — AI value rarely appears in a single step but emerges in a chain, starting with process and capability improvements and eventually showing up in financial outcomes.
Table 1: Performance areas where AI creates value; cf. the AI Radar to see how they are affected by different use cases.
Let’s examine how value emerges for each opportunity type, and where you should focus to maximize it.
Automation
In automation, the system takes over an existing task and executes it with minimal human intervention. This is especially useful when large volumes of similar decisions must be made quickly and consistently. The AI system evaluates structured inputs and produces classifications or decisions at scale. Humans might still be involved to compensate for AI inaccuracies through two mechanisms:
- Verification: Humans can approve or reject AI outputs after reviewing them.
- Escalation: AI handles common cases where it has a high confidence, handing off more complex cases to the human.
However, the end game for automation initiatives is to completely remove manual work from a process. The central challenge is therefore reliability: can the system perform the task accurately enough to remove humans from routine execution?
As an example, let’s look at fraud detection for financial transactions. Banks process millions of transactions each day. AI systems can analyze these streams in real time and flag suspicious patterns. Most transactions pass automatically, while a small subset is escalated to human analysts for further investigation. The system therefore performs the operational screening, while human experts focus on ambiguous or high-risk cases.
Figure 1: AI-driven fraud detectionaims at automating the first, time-consuming step of screening all incoming transactions.
Where value emerges
Automation is the most intuitive form of AI value — if a human workload disappears, the impact is easy to quantify and measure.
Leading indicators
The earliest signal is usually Efficiency. In our example, once the fraud detection system is deployed, most transactions can be screened continuously without manual review. This allows organizations to process large volumes of transactions with far less manual effort.
Caveat: When estimating efficiency gains, it is important to not assume perfect performance. Your AI system will likely still make mistakes. The effort of finding and fixing these mistakes means additional effort which needs to be subtracted from your value equation.
A second leading indicator is Speed to Insight. Suspicious transactions can be detected immediately rather than after delayed manual analysis, allowing investigators to react faster and reduce potential downstream harm.
Lagging indicators
Over time, a more efficient process leads in Cost Savings and improvements in Risk & Compliance. Automation also improves Scalability — as the system handles increasing volumes of transactions, organizations can scale operations without expanding investigation teams.
Strategic value
Automation rarely creates lasting differentiation. Once the technology becomes widely available, competitors quickly catch up. Its real strategic role is foundational: automation removes large amounts of routine work, improves employee experience, and frees up human capacity for more complex, creative, and strategically relevant activities.
Where value can be amplified
The value of automation systems hinges primarily on the accuracy and reliability of the AI system, which determines how much human intervention is still needed. In the example of fraud detection:
- The key lever is model accuracy. It determines how well the system distinguishes between legitimate and fraudulent transactions.
- A second lever is data coverage and a smooth data pipeline. Fraud patterns evolve constantly, so the system must learn from diverse and up-to-date transaction data, including feedback from human investigators.
- Finally, value depends on the accuracy of escalation decisions. The system must determine when to handle a transaction automatically and when to involve a human analyst. Setting this boundary correctly is crucial: too many escalations reduce efficiency, while too few increase risk.
Based on the AI System Blueprint, the following figure summarizes the value logic of automation systems.
Figure 2: The value logic of automated AI systems
For more examples of automation scenarios, take a look at these use cases:
Augmentation
In the augmentation scenario, AI doesn’t fully replace human work but supports human experts in performing their work. Typically, these are complex, multi-step tasks where each step can branch out into different directions depending on the outcome of the previous step.
The use of AI for UX research illustrates this mechanism. Companies collect large volumes of user feedback across surveys, interviews, product reviews, etc. AI systems can analyze these data sets, identify recurring themes, and generate structured summaries. Product teams can guide the analysis, interpret the insights and translate them into design decisions or roadmap priorities. The AI system expands the information available for decision-making, while humans remain responsible for evaluating and acting on the insights.
Figure 3: With AI, the UX research process can be made more flexible and interactive, leading to more accurate and objective insights.
Where value emerges
Value emerges in better decisions, which eventually compound into better customer experience and financial performance.
Leading indicators
A common leading indicator is Quality & Accuracy, which can improve for several reasons:
- When AI handles routine tasks such as data processing, experts can dedicate more time to deeper interpretation and judgment.
- Human–AI interaction makes the process more iterative: users can refine questions, explore alternative perspectives, and revisit intermediate results when necessary.
- AI can act as an impartial sparring partner that surfaces patterns or arguments the human expert might overlook, helping to reduce bias and broaden the analytical perspective.
A second indicator is Speed to Insight. As AI takes over time-consuming data processing and analysis tasks, experts can work with larger, more diverse datasets and reach relevant insights more quickly.
Augmentation systems also improve Work Experience. Analysts and product managers spend less time on mechanical tasks and more time interpreting insights and translating them into creative, actionable outcomes.
These indicators are qualitative and hard to measure objectively. Trust and alignment between management, expert users, and engineering is crucial to agree on what meaningful improvements look like and how they should be interpreted in practice.
Lagging indicators
Over time, improvements in decision quality translate into broader business outcomes. Better insights lead to better products, services, and operational decisions. Depending on the context, this may improve Customer Experience, reduce operational costs, and contribute to Revenue Growth through better product–market fit and more effective strategic choices.
Unlike automation, where financial impact is often visible quickly, the value of augmentation tends to compound indirectly through a series of improved decisions.
Strategic value
Augmentation can create meaningful differentiation because it amplifies existing talent and domain expertise. AI systems allow experts to analyze larger volumes of information, test ideas more systematically, and explore alternative perspectives. Organizations that combine AI capabilities with strong domain knowledge can gradually turn this interaction into a powerful competitive advantage.
Where value can be amplified
In augmentation, the end game is not about removing humans from the process, but about optimizing the division of labor between human and machine. Each side should play to its strengths while compensating for the limitations of the other.
Figure 4: In the augmentation scenario, we aim to optimize the synergy between human user and AI.
The most important lever for increasing value is human–AI interaction design. In the long term, augmentation systems are only adopted if they are seamlessly embedded into the workflows they support. Insights should therefore appear at the moment when teams make decisions — for example during product reviews or roadmap planning. The user experience should also be highly flexible so workflows can be adjusted at each stage. Conversational and agentic experiences allow to accommodate this versatility.
For broader adoption, augmentation systems must be able to retrieve and operate on relevant context and domain knowledge. The system should “speak the language” of its users, incorporating the terminology, metrics, and conceptual frameworks that structure their work. Often, this requires a structured feedback loop through which users can gradually enrich the domain knowledge of the system.
The figure below summarizes value creation and measurement for augmentation systems.
Figure 5: The value logic of augmentation systems
For more examples of augmentation use cases, review the following:
Innovation
AI is coming for traditional business models. To stay competitive, companies will need to transform themselves in the coming years and decades — the runway depends on the industry. According to McKinsey’s The State of AI in 2025, high performers use AI not only to optimize their “business-as-usual,” but to drive innovation and growth. They discover and add new capabilities that were previously infeasible or economically impractical.
Generative design in industries like construction and automotive illustrates this mechanism. Traditionally, architects and engineers develop a small number of design alternatives and refine them through iterative analysis. Generative design systems transform this process by removing the human bottleneck. Engineers define constraints such as materials, cost limits, environmental conditions, and performance targets, and the AI generates thousands of possible designs that satisfy these constraints. Human experts then focus on evaluating the options and selecting the most promising candidates. This capability fundamentally expands the design space and reshapes how new products are conceived and engineered.
Where value emerges
While automation and augmentation improve existing processes and therefore have a clear baseline for measuring value, innovation benefits are more uncertain because the value of new capabilities must first be discovered and proven.
Leading indicators
The earliest signals appear at the capability level. AI enables organizations to perform tasks that were previously infeasible or economically impractical. In the case of generative design, the new capability lies in exploring vast design spaces automatically and evaluating thousands of possible configurations under defined constraints.
Innovations that restructure internal workflows often amplify Quality & Accuracy and Speed to Insight. For example, engineers can identify promising design alternatives more systematically and converge on viable solutions faster than through manual exploration.
Leading indicators can be different for innovation at the product or business model level. Here, the focus shifts toward early market signals, such as improvements in Customer Experience and customers’ willingness to pay for new features.
Lagging indicators
As the capability becomes embedded in workflows or offerings, its impact begins to appear in broader business outcomes. The specific performance areas depend on how the innovation is used. Operational innovations may translate into improvements in efficiency, scalability, or product quality. Successful product and business model innovations manifest through Revenue Growth, new service categories, or expanded market reach.
Strategic value
By enabling capabilities that competitors may not yet possess, organizations can shape new products, services, or operating models. Over time, such innovation initiatives can redefine how value is created in an industry, and early movers are in a good position to capture the benefits of that transformation.
Where value can be amplified
The success of innovation initiatives depends on how organizations discover new AI-enabled capabilities that are both feasible and valuable. The primary levers are therefore not technical, but organizational:
- Companies need a structured discovery process that encourages broad exploration of potential AI applications while still allowing promising ideas to be specified and prioritized efficiently. Innovation requires both creativity and discipline: the ability to explore new possibilities and the ability to translate them into concrete use cases.
- Organizations must be able to move forward under uncertainty. The value of new capabilities is rarely obvious from the start, and initiatives need to evolve through experimentation, iteration, and learning. Companies that succeed in AI innovation embrace this process through methods like rapid prototyping, iterative development cycles, and continuous feedback from users and customers.
- Innovation depends heavily on organizational culture. Teams need the freedom to experiment, question existing assumptions, and explore unconventional ideas. Otherwise, many AI-enabled opportunities will never be discovered or pursued.
For more examples of innovation use cases, review the following:
Key takeaways
Let’s summarize:
- AI value goes beyond efficiency. Many high-impact AI systems augment human work or enable entirely new capabilities rather than replacing labor.
- Value emerges across multiple layers. Process improvements often appear first, followed by capability improvements and eventually financial outcomes.
- Timing matters. Some benefits appear immediately after deployment (leading indicators), while others materialize only after wider adoption (lagging indicators).
- Different opportunity types create value in different ways. Automation, augmentation, and innovation follow distinct value logics.
- Maximizing AI value requires focusing on the right levers. Model accuracy matters most for automation, human–AI interaction design for augmentation, and discovery and experimentation for innovation.
The organizations that succeed with AI will not be those that automate the most tasks, but those that understand where AI creates value over time, and which levers they need to pull to maximize it.
Note: All images are by the author.
Instagram Engineering
Shopify Engineering