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 cross-functional workflow automation, document processing, and business process optimisation. Evenly split between good-practice and leading-edge: RPA and document extraction are mature; intelligent process mining and autonomous workflow orchestration are still proving out. One practice remains at research stage. Momentum is low — most practices are stalled, with gains coming from incremental automation rather than architectural shifts.
The headline: The software that automates your operations now works and the vendors selling it are finally profitable — but four in five companies that say they've adopted it cannot run it at scale. The bottleneck is your data and your processes, not the AI.
Automating operations — invoice processing, maintenance, scheduling, quality checks, document handling — is no longer an experiment. The tools are bought-and-deployed, and a small group is getting spectacular returns: the best invoice-processing setups cut cost per invoice from roughly $13 to under $3, and the top quarter of agent (software that acts on its own without being prompted) deployments earn back eight times their cost. McKinsey now counts 45% of the Fortune 500 running these agents in production, up from 8% two years ago. But here is the trap that defines the field: 79% of companies say they've adopted AI agents, and only 11% actually run them at scale. Most are stuck in pilots. The dividing line between the winners and everyone else is no longer which software you bought — it's whether you fixed your messy data and redesigned your processes first.
The biggest automation vendor turned its first-ever profit, signaling the technology has matured. UiPath posted its first operating profit and said customers are moving "from pilot to production"; a rival reported its agents ran 6 million tasks this year versus 1,500 a year ago. The capability is proven — which means a stalled project is now your execution problem, not the vendor's.
Three separate studies landed describing the same wall. The research firm Gartner published its first forecast for this market and predicts 40% of these projects will be cancelled by 2027; a survey of 4,625 IT leaders found 77% of live deployments stalled, with three-quarters blaming fragmented data rather than the AI itself. Before greenlighting your next pilot, ask whether the data it depends on is clean and connected — that is what decides success.
A documented failure shows the new risk: confident, invisible mistakes. One large company's document-reading system quietly approved $4.2 million in invoices it should have caught, because its confidence settings drifted without anyone noticing. Insist that every automation flags its own uncertainty and routes doubtful cases to a person (human-in-the-loop — a person reviews each output before it ships).
Connecting systems together got harder, not easier. Despite new tools from Microsoft and SAP, only 27% of the average company's applications are linked, and half of deployed agents work in isolation. Treat your integration backlog as the gating constraint on every automation ambition above it.
New rules are turning "a human must review this" from best practice into law. US drug-safety rules already require human sign-off on AI quality decisions, and the EU's AI Act adds obligations from August 2026 covering workforce scheduling and document trails. Map which of your automated processes touch regulated decisions now, before the deadline forces a scramble.
The market is standardizing on "boring but reliable" plumbing under the AI. After multi-agent systems failed up to 87% of the time in testing, buyers are demanding durable, auditable workflow engines underneath. Make "does it have a full audit trail and recover cleanly from failures" a hard requirement in your next vendor selection.
Expect a wave of write-downs and quiet cancellations through 2027. With Gartner forecasting 40% of these projects abandoned, the question shifts from "should we automate" to "can we finish what we start." Concentrate spending on a few bounded, high-volume processes you can actually complete rather than spreading thin across many pilots.
The problem is your data and processes, not the software. Across every operation we track, 70–85% of failures trace to fragmented data, silos, and processes nobody redesigned — issues no vendor can sell you out of. The companies winning bought the same tools as everyone else and did the unglamorous cleanup first.
The mistakes that hurt most are the ones you don't see. Modern automation rarely fails loudly; it approves the wrong invoice or cites a source that doesn't exist, quietly, at scale. Auditors found error rates above 60% when measured by actual financial exposure rather than headline accuracy.
Value is real but narrow. Automation pays off handsomely in bounded, repetitive, well-governed tasks — invoices, claims, IT tickets, defect detection — and stalls everywhere that requires crossing systems or handling genuinely messy work. Knowing which of your processes is which is the whole game.
Go deeper: the full Operations & Process Automation 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.