AI Strategy: 62% of Enterprise AI Stuck in Pilot - Even as ROI Rises
AI budgets are up. Returns are showing. Yet most projects don't make it past the pilot. That's the core tension in Kyndryl's second annual Readiness Report, which surveyed 3,700 senior leaders across 21 countries.
Martin Schroeter, Chairman and CEO at Kyndryl, put it plainly: "A readiness gap exists as enterprises grapple with the promise of transformative value from AI. While 90% of organisations think they have the tools and processes to scale innovation, more than half are stalled by their tech stack and less than a third say their employees are truly ready for AI. Closing that gap is the challenge and opportunity ahead."
What the data says
- AI spend is up 33% year-over-year; 68% of organisations are investing in at least one form of AI.
- 54% report positive ROI from AI investments.
- 62% still can't push projects beyond pilot.
- Cybersecurity is the top AI use case.
- 90% believe they can test/scale ideas quickly, yet more than half say their tech stack blocks innovation.
- In 2024, 90% claimed "best-in-class" infrastructure, but only 39% felt ready for future disruption-a confidence gap that persists.
The people gap is bigger than the tech
Leaders expect AI to reshape jobs within 12 months (87%), but only 29% say their workforce is ready to use it effectively. Employees aren't using AI consistently and lack the technical skills to make it stick.
Cultural drag compounds the issue: 48% of CEOs say their organisation stifles innovation, and 45% say decisions take too long. That's how promising pilots stall in committees and never reach production.
Pacesetters do it differently
The report highlights "Pacesetters" that move faster and break fewer things. They invest in culture, training, and leadership coordination alongside tech.
- 32% less likely to cite the tech stack as a barrier.
- 20% less likely to have experienced a cyber-related outage in the past year.
- 30 percentage points more likely to say their cloud can adapt to new regulations.
Infrastructure is the bottleneck (and the leverage)
AI workloads are exposing fragile foundations. 70% of CEOs describe their current cloud setup as "accident rather than design." Three in four leaders worry about geopolitical risks tied to data in global clouds.
65% have already shifted strategy via data repatriation, vendor reassessment, or moves toward private cloud. The message: AI value depends on clear data boundaries, architectural intent, and compliance-readiness-not just model performance.
Why pilots stall despite ROI
- Disconnected stacks: Data isn't production-ready; pipelines are brittle; MLOps is immature.
- Unclear ownership: No single accountable leader to drive use cases across business, data, and IT.
- Risk posture: Security, privacy, and regulatory controls added late instead of built-in.
- Talent and culture: Teams lack skills, incentives, or time to adopt new workflows.
- Value translation: Early wins don't translate into repeatable playbooks and funding models.
Use case spotlight: Cybersecurity leads
Security is where AI is landing first-threat detection, incident triage, identity risk, and policy enforcement. It's measurable, high stakes, and already data-rich. If you're looking for an entry point with clear ROI signals, start here.
Useful reference for aligning controls: NIST Cybersecurity Framework.
From pilot to production: a 90-day operating plan
- Pick 2-3 production-worthy use cases: Tie them to a clear metric (cost out, risk down, revenue up). Kill the rest for now.
- Assign an accountable owner: One leader across business + data + engineering, with weekly decision rights.
- Harden the data path: Define source-of-truth, quality checks, lineage, PII controls, and model monitoring before go-live.
- Stand up MLOps/GovOps basics: Versioning, CI/CD for models and prompts, approval gates, audit trails, rollback plans.
- Security-first patterns: Threat models, isolation for model endpoints, secret management, policy-as-code.
- Right-size the cloud footprint: Map workloads to GPU/CPU needs, storage tiers, and data residency. Reduce accidental multi-cloud sprawl.
- Prove value early: 30/60/90-day KPIs, agreed with finance. Lock a funding model for phase two if targets are hit.
- Skill the team, fast: Targeted upskilling for product owners, data engineers, SRE, and developers. Bake AI into everyday tools and SOPs.
- Codify the playbook: Template the process-use case intake, risk review, deployment pattern-so the second and third projects move faster.
Where Kyndryl fits
The report pulls from 3,700 leaders and operational telemetry from Kyndryl Bridge, the company's digital business platform. It's a reminder that AI success is less about a single model and more about the system around it: architecture, controls, people, and incentives. Additional background: Kyndryl.
Action for General, IT, and Development roles
- General (Exec/PMO): Set a single-threaded owner and tie AI initiatives to one business metric each. Establish a stop/scale cadence.
- IT/Ops: Prioritise data contracts, access patterns, observability, and SRE support for AI workloads. Treat models as first-class deployables.
- Developers: Ship small, production-grade increments. Add evals, prompt tests, feature flags, and guardrails from day one.
Upskill your teams
If your workforce readiness is the constraint, make training part of the operating plan-not an afterthought. Curated paths by role can speed this up.
Explore role-based AI training to close the skills gap while you build.
Your membership also unlocks:
