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: AI customer service is now mainstream — nearly half the Fortune 500 runs it — but three in four enterprises that gave agents free rein had to pull them back after launch. The winners kept a person reviewing the high-stakes work.
Customer service is the single most common place companies put AI to work: 78 percent of all corporate AI deployments live here. Most organizations have something running, and a small group is pulling ahead with real savings — 40 percent lower cost per interaction, hundreds of thousands of staff hours freed. But the gap between what vendors promise and what customers actually experience has become the defining problem. A survey of 2,527 enterprises found 74 percent rolled back their fully autonomous agents (software that answers customers without a person checking first). The uncomfortable twist: the best-governed companies pulled them back the most — not because their AI was worse, but because they could actually see it failing. If you are deploying here, the question is no longer whether the technology works. It is whether your data, your knowledge base, and your escalation process are ready for it.
The safest, most proven AI pattern in customer service has hit its ceiling — and that is good news. "Auto-draft" — where AI writes the reply and a human approves it before it sends — now handles a majority of customer conversations and delivers roughly a third faster handling. It stopped advancing this cycle not because it failed but because it is finished maturing: keeping a person in the loop is now a permanent feature, not a training-wheels phase. Treat the review step as architecture, not a cost to cut later.
A wave of named failures landed at once. Air Canada was held legally liable for a chatbot that invented a policy; one retailer's agent issued $2.3 million in unauthorized refunds; a support bot was tricked into hijacking 20,000 social accounts. Stanford logged 362 AI incidents in 2025, up 55 percent. The pattern is consistent — agents given the power to act without a check on each action. Audit where your AI can take irreversible actions (refunds, account changes) without human sign-off.
Vendor numbers and customer reality have split wide open. One telecom reported 78 percent of calls "contained" by AI — but only 41 percent of customers actually got their problem solved. One bank's voice bot scored 4.6 out of 5, yet 91 percent of those customers hung up, asked for a human, or called back within a day. Stop accepting "deflection" or "containment" as success metrics; ask vendors for true resolution and repeat-contact rates.
EU enforcement starts in August. The EU AI Act's rules on autonomous decision-making begin landing, and Britain's competition regulator has issued guidance carrying fines of up to 10 percent of global turnover for non-compliant agents. If you operate in Europe, confirm now that your AI's customer-facing decisions are documented, auditable, and reversible.
The bottleneck is your data, not the AI. The same model jumped from 25 to 79 percent of issues resolved at one company purely by cleaning up the knowledge base behind it. Budgets are flowing to AI tools while the unglamorous work — fixing fragmented data and stale help content — goes unfunded. Fund the knowledge and integration work first; it is what actually moves the numbers.
Platforms are bundling autonomy in by default. Microsoft now ships fully autonomous email handling, and Zendesk is folding agent features into base plans. Capability is arriving whether or not you are ready for it. Decide your autonomy boundaries deliberately before a default setting decides them for you.
Better oversight reveals more problems, not fewer. The companies with the strongest monitoring rolled back AI at the highest rate (81 percent) because they could see failures others miss. Investing in governance tooling will surface uncomfortable truths before it delivers comfort — plan for that.
AI confidently makes things up, and it cannot be fully fixed. Hallucination (when AI invents facts) runs from 22 to 94 percent depending on the task; even the best models top out around 85 percent accuracy on live calls. This is why a human review gate on consequential decisions is not optional — it is the cost of doing business safely.
The value is real but concentrated. Only about one in ten organizations has a mature, fully working deployment, even though most have invested. The difference is not the software — it is operational discipline, clean data, and change management, none of which arrive with a purchase order.
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.