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 supporting, retaining, and understanding customers after the sale. The highest concentration of good-practice tiers: chatbots, ticket routing, sentiment analysis, and voice-of-customer are deployed at scale in most industries. Bleeding-edge frontiers include autonomous resolution without human escalation and real-time emotion detection. Momentum is steady but churn prediction and proactive outreach remain stalled.
The headline: The AI support market just stopped counting pilots. Reported adoption halved in a year, the survivors are the ones who kept a person in the loop, and two large breaches showed that support chat logs are now a serious security exposure.
Around 88 percent of contact centers now have AI somewhere in customer service, but only a quarter have wired it into how they actually work. A small group is pulling well ahead: they measure whether the customer's problem was solved rather than whether the conversation ended, they own their help content properly, and they route anything complicated to a human. Everyone else sits between an expensive pilot and a rollback — roughly three-quarters of companies that launched fully automated service agents have since pulled them back. The window is closing because customers, regulators, and auditors are now applying standards most deployments were never built to meet.
Two breaches showed AI support systems are a soft target. Researchers at Wharton documented that Sears Home Services left 3.7 million customer chat transcripts sitting unencrypted, and that McKinsey's internal AI assistant was broken into and its instructions rewritten — exposing 46.5 million messages. Ordinary security review caught neither. Ask your security team a specific question: who has read and write access to your AI agent's conversation logs and to the instructions that govern its behavior.
Reported adoption fell from 95 percent to 54 percent in a year. A Roland Berger survey of 550 leaders across ten countries did not find companies retreating from AI — it found them dropping the pretence about pilots that never worked. The ones still standing report solid double-digit gains on cost and satisfaction. If your own AI programme is still described as a pilot after twelve months, it belongs in the discontinued column, not the roadmap.
Vendors moved to charging per problem solved, and the price is misleading. Salesforce and HubSpot both shipped agents charging roughly $0.50 to $2.00 per resolved case, which sounds like the vendor taking on your risk. Independent analysis puts the real all-in cost nearer $5 once integration, content upkeep, and human escalation are included. Model your own total cost before signing an outcome-based contract; the headline rate is not the number you will pay.
Job cuts became public and specific. Commonwealth Bank saved over $10 million a year, Microsoft cut customer-service headcount from 50,000 to 40,000 for $750 million in annual savings, and Hyatt cut 30 percent of in-house support. Forrester expects half of customer-service roles to be affected by 2030. Note the counterexample: Klarna cut deeply, watched satisfaction fall, and rehired — so tie any headcount decision to measured service quality, not projected savings.
EU chatbot disclosure rules took effect on 2 August. Any chatbot serving EU customers must now tell them they are talking to a machine, with a further deadline on 2 December for machine-readable labeling. The rules apply to non-EU companies serving EU customers. This is a one-line interface change and a documentation task; confirm it is done rather than assuming it is.
Legacy chatbots start losing support on 31 August. Zendesk stops developing its old rule-based bot builder at the end of this month and removes it entirely on 10 December. Some 71 percent of software companies still run legacy scripted bots of some kind. If you are one of them, the migration is a rebuild, not a switch — start scoping it now rather than in November.
Live call translation is about to become standard. Among CX leaders surveyed, 28 percent have deployed AI voice translation and 60 percent plan to within six to twelve months, with Zoom, Google, and Krisp all shipping it this quarter. Accuracy still falls off sharply on mixed-language speech and regional accents, so pilot it on your highest-volume language pair before committing to it as a staffing strategy.
Automated phone agents are drawing regulatory fire. US telemarketing enforcement has produced settlements of $14 million and $9.95 million against companies using AI voice calling, and regulators are drafting broader rules. If any outbound calling in your business is AI-generated, have counsel review your consent handling this quarter.
Your help content, not the AI, decides whether this works. Companies adopt AI agents at 79 percent but AI-assisted knowledge management at 27 percent, and the gap tracks the failure rate almost exactly. About 60 percent of these systems break within 6 to 18 months because old policies quietly contradict new ones — a healthcare deployment invented a drug dosage because a warning got split across two documents, costing $4.7 million.
Better monitoring makes the numbers look worse before they look better. Rollback rates are highest — 81 percent — at the companies with the most mature oversight, because they are the ones who can actually see the failures. Expect your first honest measurement cycle to show performance below what your vendor's dashboard reported.
Customers do not want this, unless it works perfectly. Roughly 64 percent say they would prefer companies not use AI for service at all, and 34 percent cut spending after a bad AI experience. But 69 percent say they would happily switch to AI if it fully resolved their issue — the objection is to bad implementation, not automation, which puts the whole burden on your escalation and handoff design.
Go deeper: the full Customer Operations 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.