AI adoption is outpacing control: what CIOs need to fix now
AI spend is surging, but control is slipping. A new survey of 1,000 global tech leaders finds organizations racing ahead on AI while the governance, skills, and alignment needed to run it at scale lag behind.
More than half (51 percent) say adoption is moving faster than they can manage. Fewer than half report their AI strategy is fully tied to business plans or KPIs - a gap that explains why pilots start strong and stall later.
The strategy gap: ambition without alignment
AI initiatives are being greenlit without mature ways to measure value, prioritize investment, or define success. That creates a pile-up of pilots, scattered tools, and unclear ROI.
- Link every AI use case to a single business KPI and an accountable owner.
- Define "value" upfront: revenue lift, cost per task reduced, cycle time, or risk avoided.
- Stage-gate funding: move from idea → pilot → limited rollout → scale only with evidence.
Governance and risk: accountability without a playbook
CIOs report being on the hook for both delivery and integrity of AI deployments, yet only 36 percent feel they have strong guidance and practices in place. Two-thirds (67 percent) also worry about an "AI bubble," and many admit they lack continuity plans if a key provider disappears.
- Adopt a recognized framework (e.g., the NIST AI Risk Management Framework) and make it your standard.
- Stand up model risk controls: model registry, data lineage, bias/quality checks, human-in-the-loop for material decisions.
- Vendor risk basics: exit clauses, data portability, model fallback plan, and a tested "kill switch."
- Security by default: prompt injection defenses, PII controls, and red-teaming for critical workflows.
Skills: the real blocker
Funding isn't the choke point - skills are. Nearly 9 in 10 organizations say lack of internal capability is holding back progress. That shows up as scattered tooling, slow delivery, and overreliance on vendors.
- Build an AI Center of Excellence with product, data, MLOps, security, and legal at the table.
- Upskill engineers, analysts, and product leads on prompt design, evaluation, data pipelines, and model lifecycle.
- Standardize reference architectures for common patterns: retrieval-augmented generation, prediction, and automation.
If you need a structured track for leadership and delivery teams, see the AI Learning Path for CIOs and AI for Executives & Strategy.
Scaling beyond pilots: 65 percent lack confidence
AI works best where data ownership and processes are already clean - like forecasting and predictive analytics. The challenge is moving from clever demos to durable, organization-wide services.
- Create a productization checklist: data contracts, evaluation metrics, latency/SLA targets, observability, rollback plan.
- Centralize platforms (feature stores, vector DBs, model gateways) to avoid one-off solutions.
- Budget for post-launch: model drift monitoring, re-training, and continuous evaluation.
Don't neglect the basics
Leaders have warned that core IT services can suffer while teams chase generative AI infrastructure. That trade-off shows up later as incidents, blown SLAs, and unhappy business partners.
- Protect service quality: ringfence capacity and staffing for foundational IT.
- Track AI infra spend against total IT performance and reliability metrics.
- Adopt a weekly operating review for AI programs that includes platform health and service risks.
Sustainability: the blind spot
Only 39 percent of organizations measure the environmental impact of AI, and just 41 percent prioritize energy efficiency in deployment - despite estimates that datacenter electricity use could double by 2030, with much of it powered by fossil fuels.
- Set energy KPIs per workload (kWh/inference, kWh/training epoch) and include in project approvals.
- Prefer efficient patterns (RAG, fine-tuning smaller models) over retraining large foundations.
- Use workload scheduling, right-sizing, and greener regions to cut emissions.
For context, see the International Energy Agency's analysis of data center electricity trends: IEA: Data centres and networks.
Investment isn't slowing - it's shifting
Nearly three-quarters (72 percent) plan more AI investment in the next 12 months. Sixty percent expect to fund agentic AI - systems that plan tasks and act across tools - which raises new governance, reliability, and vendor risk questions.
- Set boundaries for autonomous actions: approved tools, spending limits, audit logs, and human approval thresholds.
- Pilot agent workflows in low-risk domains before touching revenue or compliance-sensitive processes.
A 90-day plan for CIOs
- Publish your AI policy v1.0: data use, model approval, human oversight, security, and incident response.
- Create a single AI portfolio with value hypotheses, KPIs, owners, and stage-gates for every use case.
- Stand up evaluation: bias, quality, cost, latency, and business impact - reviewed before scale.
- Run a vendor continuity drill: export data, switch endpoints, and validate a fallback model.
- Launch an executive dashboard: ROI, risk, sustainability metrics, and service impact - updated monthly.
- Fund skills: targeted training for product, data, engineering, and risk teams; define the AI CoE charter.
The message is clear: AI isn't going away, but success depends on managing it with the same discipline you apply to security, reliability, and finance. As Logicalis's CEO Bob Bailkoski put it, organizations have plenty of appetite for AI - what's missing are the frameworks, skills, and confidence to run it at scale.
Your membership also unlocks:
