AI's 'Proof Of Value' Gives CTOs The Business Case Technical Performance Cannot

An AI program can hit every technical target and still fail to win funding to scale. A CTO can build the platform, connect the data, and prove the model works. But the decision to deploy it across the company belongs to the CEO and business unit leaders, who judge it by the business result it can deliver. The CTO’s job is to help define that result at the start, before there's a system to build or a model to measure. Technology validates feasibility. Proof of value validates business commitment.

Curtis (Jungsoo) Kim is a Founding Partner at THE CRUX, a consulting firm that helps companies turn unmet customer needs into new products and business opportunities. He has spent more than 30 years in technology and business roles at Google, Motorola, Samsung, SK, Hanwha, and LG U+. Previously, as Head of Digital Innovation for Asia Pacific at AWS, he helped more than 70 enterprises translate business needs into technology initiatives, leading sessions where technology and business leaders defined the purpose and intended outcomes of each project. That experience taught Kim that the technology itself is only one part of the work, and rarely the part that determines whether transformation succeeds.

"AI tools like LLMs, agents, or chatbots are the method, not the goal. The real object of AI transformation is the enterprise itself, its operating model, decision-making processes, and the way value is created. The entity of the transformation is the enterprise itself," he said. That work reaches into parts of the company no single executive controls.

• Factory floor friction: Kim describes enterprise AI as an AI Factory which turns a company’s data into predictions and automated decisions. Just as a manufacturing factory transforms raw materials into finished goods, an AI Factory transforms enterprise data into predictions, decisions, automation, and ultimately business outcomes. It pulls information from across the business, processes it through models, and sends the results back to the people and systems that act on them. Technology teams build the factory, but business systems, workflows, and data sources must be orchestrated across departments before AI can generate meaningful business outcomes. The challenge is not building AI. It's orchestrating workflows across the enterprise so that AI can continuously generate measurable value. “While it is a CTO’s responsibility to build the AI factory, determining what kind of data to use for business growth extends far beyond IT,” Kim said. Without that input, technology teams naturally optimize what they can measure, such as model accuracy, latency, or inference cost. Business leaders, however, approve investments based on revenue growth, productivity gains, customer experience, or operational efficiency. Model accuracy alone won’t secure business approval. AI programs are more likely to advance when every part has a named owner and a shared measure of success.

• Four levers, four owners: Kim effectively describes AI transformation as a new leadership operating model. Rather than placing responsibility on IT, he distributes ownership across four executives: the CEO owns strategy, the CTO owns architecture, finance and risk own governance, and HR owns organizational adoption. In his view, strategy belongs squarely to the CEO and board, who commit the company to products or services that improve as more people use them. The first turn of that cycle costs the most and returns the least, so it needs the CEO to commit before there's anything to show. Architecture is shared between the CTO and business unit leaders. Governance sits with finance and risk. Trust cannot be added after deployment. It must be designed into the architecture from the beginning through governance, privacy, explainability, and responsible AI practices. Culture falls to the CHRO, because the same change lands differently across a workforce. Kim described people who panic, people who stay indifferent, people who resist, and people who take to it. Each group needs to hear something different, and Kim assigns that work to HR. "All four leaders must pull their levers at the same time," Kim said. A company that moves on one and leaves the other three ends up with new tools running an unchanged operating model.

• Brakes make speed: A system running across the company makes far more decisions than a pilot, with fewer people reviewing each one. Kim sets what it is allowed to do while it is still being designed. "A vehicle without brakes cannot run fast. A supercar can run fast not only because of a great engine, but because of great brakes," Kim said. The controls he specifies cover privacy protections and checks for bias in the system's outputs. Organizations that postpone governance until after deployment often find themselves redesigning production systems precisely when they need to scale them.

Hanwha Vision makes security cameras for businesses. When Kim worked there, hotels primarily used them to monitor theft and accidents. But after speaking with front desk and housekeeping staff about the delays they faced each day, he saw another use. Cleaners often waited for confirmation that a guest had checked out before turning over a room, while managers had little visibility into when facilities like the restaurant and gym were busiest. The cameras already captured those events, so Kim’s team configured them to record the activity as data without storing the underlying video. Housekeeping could then receive an alert as soon as a room was vacant, and managers could track when and how guests used hotel facilities. The cameras stopped being just a security expense and became part of the hotel’s operating system. The breakthrough wasn't the camera. It was redesigning the hotel's operating workflow around AI-generated events.

• Front desk first: Kim's takeaway is to put those conversations first. "If we just hand the project to the engineering teams, they will focus on technical optimizations like making higher-fidelity cameras or a more highly trained model. Business users don't describe AI features. They describe operational pain points. That's where successful AI projects begin," Kim said. The answers from hotel staff shaped what the cameras needed to do. They also set what the hotel would measure to judge whether the work had paid off, and that number is what a company weighs when it decides to fund more.

• Six months to yes: Kim used the same approach with a large logistics company. Packages were getting damaged while machines sorted them inside its warehouses, and the company put the cost of that at more than ₩200 billion (about $137.5M USD). Kim's team used cameras to find the exact points where the packages were being hit. He ran the system as a six-month trial, agreeing with the client in advance how far damage had to fall for the work to count as a success. The system met that number, and the company installed it across all of its warehouses. The technology did not convince the customer. The agreed business outcome did. "Every client undergoing AI transformation wants to go through proof of value sessions. They want to see the KPIs met before they decide to invest in a full-scale deployment," Kim said.

Kim has worked through both digital transformation and the rise of AI, but his starting point hasn’t changed: define the value before building the technology. Kim has seen enterprise AI fail because teams never establish a shared definition of business success before the implementation or adoption begins. He argues that once a definition exists, technology becomes an execution problem instead of a strategic one. "Let’s write down what value we can promise to our customer before anyone starts writing code or training algorithms," Kim said. Without a shared definition of business value, even the best AI model remains only a technical achievement. With it, technology becomes an execution challenge rather than a strategic debate.

Proof of value is not about proving AI works. It is about proving the business is worth transforming.