The question that most often blocks organizations at the start of the AI automation journey isn't "can we do it?" but "where do we start?" The wrong answer to this question has concrete consequences: 72% of CIOs reported to Gartner in a May 2025 survey that their organization is breaking even or losing money on AI investments. This isn't a technology failure. It's a selection failure.
The main criterion: high value, high feasibility
Gartner provides a clear operating principle for I&O leaders who want to adopt AI intentionally: don't chase big AI projects, but start with high-value, high-feasibility pilots with flexible upgrades. The value/feasibility matrix is the most direct prioritization tool available: processes in the upper-right corner, where potential benefit is high and implementation conditions are favorable, are the natural starting point.
High-feasibility processes for AI automation share precise characteristics that Gartner has identified in the specific context of agentic AI: high volume of interactions and transactions, high-quality and well-structured data, APIs and process orchestration already available, governance and security controls already in place, and a user base already familiar with AI tools. The more these conditions are met, the more predictable the implementation and the more manageable the risk.
The signals that indicate a process is ready
Gartner has identified the signals that make a process particularly suited for AI automation. The first is high volume: processes handling thousands of similar cases every month produce enough data to train and validate the system, and deliver measurable impact even with marginal per-unit improvements. The second is low variance: processes where the majority of cases follow a predictable pattern with limited exceptions lend themselves better to automation than those with high variability, where contextual judgment is frequently required.
The third signal is behavioral verifiability: it must be possible to check whether the agent made the right decision and to do so efficiently. Processes where the correct outcome is ambiguous or difficult to evaluate make system governance impossible. The fourth is early return on investment: Gartner indicates that processes where ROI is expected within twelve months are those that build the internal political credibility needed to secure budget for subsequent projects.
72% of AI investments are not producing returns
In a May 2025 survey of 506 CIOs and other technology leaders, Gartner found that 72% of organizations are breaking even or losing money on their AI investments. Key causes identified include difficulty in identifying adequate use cases, unrealistic expectations about initiatives, lack of qualified professionals, insufficient governance, and integration difficulties. 48% of I&O leaders cite integration difficulties as the main obstacle to AI adoption, and 50% cite lack of budget.
These numbers don't describe a technology failure: they describe a selection and governance failure. Organizations producing concrete results on their AI investments start from clear use cases, with measurable objectives and a validation framework defined before implementation.
By 2030, the entire IT function will involve AI
In a July 2025 survey of over 700 CIOs, Gartner captured expectations about future work models: by 2030, CIOs expect that 0% of IT work will be done by humans without AI, 75% will be done by humans augmented with AI, and 25% will be done by AI alone. This scenario isn't an abstract goal: it's the direction organizations are moving in, and it demands thinking today about how to build the automation sequence that gets there.
The correct sequence isn't determined by available technology but by process maturity, data quality, and the organization's capacity to govern automated systems. Starting with simpler processes doesn't mean giving up on ambition: it means building the foundations of competence, trust, and governance on which more complex projects can rest.
What to avoid: the most common traps
Gartner identifies three recurring mistakes in selecting processes to automate. The first is chasing big projects instead of starting with high-value, high-feasibility pilots: ambitious projects require organizational maturity that often isn't yet in place. The second is automating processes that don't work well even without AI: AI amplifies what it finds, it doesn't fix it. An inefficient process that's automated is a faster inefficient process, not an improved one. Gartner states this explicitly: you need to optimize data and processes before scaling AI deployments, not during.
The third mistake is choosing processes based on technological excitement rather than business value. The fact that a process can be automated isn't a sufficient argument for doing so. The right question is: if this process is automated well, which business outcomes change, by how much, and how quickly? If the answer is vague, the process isn't the right starting point.
