In 2025 only 26% of organisations had a Chief AI Officer. In IBM's 2026 CEO Study that share jumped to 76%. The new role is spreading faster than almost any executive title before it, but a title alone does not create value. IBM's data shows that what matters is the operating model behind it: organisations whose AI leader runs a centralised or hub-and-spoke model report 36% higher ROI on AI. This guide explains what a Chief AI Officer actually does, how the three AI operating models compare, and how a mid-sized company in Croatia or the DACH region can organise AI ownership without hiring a new C-level executive.
Why "who owns AI?" became a board question in 2026
Two years ago most companies treated generative AI as an IT experiment or a set of individual ChatGPT and Claude subscriptions. In 2026 AI touches budgets in every department: Claude Enterprise or ChatGPT Enterprise seats, API spend on models from Anthropic, OpenAI and Mistral, agent platforms, data projects and compliance work under the EU AI Act. When nobody owns the whole picture, the result is predictable: dozens of pilots, duplicate tools, unclear spend and almost nothing in production.
From 26% to 76% in one year
The IBM Institute for Business Value surveyed more than 600 Chief AI Officers in 22 countries in early 2025 and found that 26% of organisations had one. The 2026 CEO Study, based on 2,000 CEOs across 33 geographies and 21 industries (fieldwork February to April 2026), reports that 76% of surveyed organisations now have a CAIO. The two surveys use different samples, so read the jump as a strong direction rather than a precise growth rate. Either way, the CAIO has gone from a niche title to the default within a year.
Sources: IBM Institute for Business Value, CAIO study (July 2025, 600+ CAIOs, 22 countries) and 2026 CEO Study "Rewiring the C-suite" (May 2026, 2,000 CEOs, 33 geographies). Different samples.
What a Chief AI Officer actually does
The role differs from company to company, but in IBM's 2025 data a clear profile emerges. Most CAIOs are promoted from inside (57%), most come from a data background (73%), and 61% control their organisation's AI budget. Only 48% say direct AI implementation is a primary responsibility: the job is less about building models and more about deciding what gets built, by whom, and how it is measured.
- Portfolio and budget: one prioritised list of AI use cases instead of scattered pilots, with business cases and funding decisions in one place.
- Platform and vendor choices: which models (Claude, GPT, Mistral or open-weight models), which enterprise plans, which agent and integration standards such as MCP, and which data may go where.
- Governance and compliance: AI policy, risk classification and documentation under the EU AI Act, and since February 2025 the Article 4 obligation to ensure sufficient AI literacy among staff.
- Measurement: baselines before a project starts and agreed KPIs after go-live (see our guide to measuring AI ROI).
- Adoption: training, champions and change management, because a licence nobody uses is pure cost.
Operating model beats org chart: the 36% finding
The most useful number in IBM's CAIO study is not the adoption rate. Organisations with a CAIO report 10% higher ROI on AI spend than those without one. Where the CAIO runs a centralised or hub-and-spoke operating model, ROI on AI initiatives is 36% higher than in decentralised models. These are survey correlations, not controlled experiments, but they match what we see in projects: the structure around the AI leader explains more than the title itself.
Source: IBM Institute for Business Value with Oxford Economics, CAIO study (July 2025). Survey-based correlation; each bar is compared with its own baseline.
The same study shows how much work remains. Globally, 60% of organisations are still at pilot stage with limited deployment, 68% start AI projects without being able to measure their effects, and 72% say they risk falling behind if they do not measure AI impact. 30% of CAIOs find implementation very difficult.
Source: IBM Institute for Business Value, CAIO study (July 2025), global figures.
Three AI operating models compared
| Model | How it works | Strength | Risk |
|---|---|---|---|
| Centralised | One AI team builds and runs everything | Strong standards, security and cost control | Becomes a bottleneck; far from business problems |
| Hub-and-spoke | Central hub owns platform, standards and governance; business units own use cases with AI champions | Reuse plus business ownership; best path to production | Needs clear rules on who decides what |
| Decentralised | Each department buys and builds its own AI | Fast local experiments | Duplicate tools, shadow AI, no measurement, compliance gaps |
Assessment by AI Workshop / Blocksize based on project experience, November 2026. Higher is better on every axis.
In practice, most companies we work with land on a hub-and-spoke model. The hub is small: an accountable AI owner, one or two engineers or an external partner, and someone responsible for security and compliance. It runs the shared platform (for example Claude Enterprise or ChatGPT Enterprise with SSO and admin controls, an API gateway, approved MCP connectors and logging). The spokes are the business units, each with a named AI champion who knows the processes and owns the results.
The CEO gap: expectations run ahead of the workforce
IBM's 2026 CEO Study shows why a strong operating model matters now. CEOs expect the share of codifiable operational decisions made by AI without human intervention to rise from 25% today to 48% by 2030. Yet according to the same CEOs only 25% of the workforce uses AI regularly, while 86% believe employees have the skills to work with AI. 83% say AI success depends more on people's adoption than on technology, and organisations that redesigned five core business areas were four times more likely to have delivered on their business objectives.
Source: IBM Institute for Business Value, 2026 CEO Study (May 2026). Left: share of codifiable operational decisions made by AI, today vs CEOs' 2030 expectation. Right: workforce using AI regularly vs CEOs who believe employees have the skills to collaborate with AI.
A 90-day plan for mid-sized companies
Most companies in Croatia and the DACH region do not need a new C-level hire. They need clear ownership. A pragmatic sequence:
- Days 1–30: name the owner and take stock. Appoint one accountable AI lead (often the CIO, COO or a digital director) with budget authority. List every AI tool, subscription and pilot in use, including personal accounts.
- Days 31–60: build the hub. Choose an enterprise AI platform, publish an AI policy, set up a risk register in line with the EU AI Act and, if useful, an ISO 42001 management system. Start AI literacy training for all staff.
- Days 61–90: activate the spokes. Name an AI champion per department, pick two or three use cases with a measured baseline, and agree how results are reported each quarter.
Our AI transformation programme follows this structure, and the guide to EU AI Act compliance covers the regulatory side in detail.
The bottom line
The Chief AI Officer has become the norm in large organisations within a single year. The data says the value comes not from the title but from the operating model: one owner with a budget, a central hub for platform, security and standards, and business units that own their use cases. For mid-sized European companies, the hub-and-spoke model with a lean central team, or an external partner in that role, is the fastest way from scattered pilots to measurable results.
Who owns AI in your company?
AI Workshop helps you set up a hub-and-spoke AI operating model: ownership and governance, platform choice between Claude, ChatGPT and Mistral, EU AI Act and AI literacy, and the first use cases with measured ROI. Our Claude Certified Architects can also act as your interim AI hub.
Talk to an AI consultant