The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI for turning raw data into queryable, analysable, actionable insight. Streaming analytics, MLOps, and feature engineering are good practice with proven deployments at scale. The bulk sits at leading-edge, held back not by tooling but by data quality and governance gaps — 60% of AI projects stall on data readiness. Nearly all practices are stalled in trajectory.
The headline: The tools work; most organizations can't yet use them. Across the board, AI projects are failing not because the models are weak, but because the underlying data is messy, ungoverned, and untrusted.
A wave of hard numbers landed this month confirming what the field has suspected for a year: the bottleneck is data, not AI. Gartner found 72 percent of enterprise AI projects fail or underperform. An MIT study found 95 percent of generative-AI pilots — projects using tools like ChatGPT — delivered no measurable financial return. Yet a small group of companies are pulling clear returns, and they have one thing in common: they spend the bulk of their AI budget fixing data foundations (the quality, organization, and documentation of their data) rather than buying flashy analytics tools. Most companies are doing the reverse, and discovering they can't trust what the AI tells them. If you are buying AI tools faster than you are cleaning up your data, you are in the failing majority, not the winning minority.
The failure rate is now documented, not anecdotal. Three large studies converged — Gartner's 72 percent project failure rate, MIT's 95 percent of pilots showing zero financial return, and a 10,000-business survey finding only 5 percent consider their data AI-ready. The pattern is consistent: the cause is data quality and governance, not the AI itself, so the highest-return investment right now is fixing data before buying more AI.
Autonomous AI is being pulled back out of production. A survey of 2,527 decision-makers found that 74 percent of deployed "agents" — AI that acts on its own without being prompted — had to be rolled back, and one widely reported incident involved an AI agent deleting a live company database despite instructions not to. Treat autonomous AI as something to pilot under tight supervision, not deploy unsupervised.
Vendor lock-in is emerging as a real cost. A market analysis found 94 percent of enterprises worried about being locked into their AI vendor, and 47 percent said they could not switch their main vendor without halting core operations. As you sign deeper AI deals with platforms like Snowflake or Databricks, push for exit terms now — switching later is expensive and disruptive.
The "smarter model" myth got punctured. New testing showed leading AI scoring 82 percent on lab tests but as low as 11 percent answering questions on real company data — while a well-organized data layer scored near-perfect on the same questions. The lesson: buying a more powerful model rarely fixes a data problem.
The EU AI Act bites on data lineage. High-risk AI systems will need documented data lineage (a traceable record of where data came from and how it changed), with obligations landing through 2026 and 2027. If you operate in or sell to the EU, start mapping which of your AI systems are "high-risk" and whether you can prove your data's history.
Agentic AI hype meets a hard wall. Gartner projects 40 percent of autonomous-AI projects will be cancelled by 2027 on cost and governance grounds. Budget for governance and human oversight as a line item in any agentic project, not an afterthought — the projects that survive will be the ones that planned for it.
The streaming-cost correction spreads. Vendors are now openly advising customers to drop expensive real-time data processing unless they genuinely need sub-second speed. Review whether your "real-time" systems actually need to be — many can move to cheaper approaches with no business impact.
You cannot buy your way out with better tools. The constraint is your own data — its quality, ownership, and documentation — which no vendor can fix for you. This is organizational work: clear data ownership, governance, and discipline, none of which a purchase order delivers.
AI confidently produces wrong answers. Systems return plausible, well-formatted results that are simply incorrect, and they do it without flagging any error. Without a human reviewing outputs (a person who checks each result before it's used), wrong answers flow silently into decisions.
The companies ahead are compounding their lead. Those that fixed their data foundations first now deploy AI reliably while competitors stall. The gap is widening, and catching up takes years of unglamorous data work, not a single big purchase.
Go deeper: the full Data & Analytics briefing — the longer analytical write-up, plus every practice we track in this domain with its maturity rating, the tools to consider, and the evidence behind our assessment.