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-generated simulations for practising professional skills including medical diagnosis, legal argumentation, sales calls, and negotiations. Includes scenario generation and performance assessment; distinct from digital twins which simulate physical systems rather than professional interactions.
AI-powered simulated practice environments have crossed from prototype to production -- but only in narrow, well-defined use cases. Contact center and sales training platforms now run at scale with documented ROI, while healthcare, education, and negotiation applications remain largely in pilot or evaluation phases. The practice extends decades of simulation-based medical training by adding generative AI for dynamic scenario creation, adaptive difficulty, and natural-language interaction. That combination unlocks something traditional roleplay cannot: repeatable, personalised practice at volume. Sales professionals need 20-30 practice repetitions to build confidence; typical training provides two or three. AI simulation closes that gap. The defining tension at this tier is breadth of adoption. Forward-leaning organisations are getting measurable value, but most have not started, and enterprise-wide rollout faces the same systemic barriers -- data quality, integration complexity, change management -- that constrain GenAI deployments more broadly.
Commercial sales and contact center platforms remain the strongest deployment anchor. Oracle NetSuite deployed Second Nature's AI roleplay to 100+ SDRs with 21% first-logo acceleration and 20% onboarding-time reduction; Zenarate's Sallie Mae client cut certification from 2.5 days to 2 hours; TDCX's e-commerce deployment (600 agents) achieved 20% CSAT increase and 50% faster proficiency ramp with 50% attrition reduction. Support Services Group documented 50% training reductions across financial services, travel, and retail. The market now shows 58% Fortune 500 penetration of AI sales roleplay with 91% adoption among high-performing sales teams and 3.2x ROI within 12 months, signaling enterprise mainstream acceptance. Contact center platforms continue scaling: 3.5x year-over-year growth in AI roleplay exercise completions with 97% learner recommendation rates. Yet persistent barriers remain: 61% of contact center leaders report GenAI conversations more challenging than expected despite 98% adoption, highlighting the gap between capability and operational readiness.
Deployment domains are expanding beyond sales and contact centers. Insurance firms are deploying AI-powered compliance training with simulated regulatory scenarios (explaining exclusions, suitability requirements) showing learning gains of 0.73-1.3 standard deviations versus traditional instruction. Home service companies (HVAC, roofing, plumbing) are using 24/7 on-demand AI roleplay with industry-specific objections and realistic voice to address the manager-availability bottleneck (traditional one-weekly practice). Negotiation training is emerging as a new domain: Harvard Program on Negotiation documents AI coaches for eviction advocacy and family caregiver negotiations with 74% skill application rates; higher education is beginning institutional adoption, with ZHAW (University of Teacher Education) launching a leadership communication and negotiation course combining research-based instruction with AI-supported simulations. Careertrainer.ai has deployed AI-powered sales simulations across 20+ industries with parameterized AI customers modeling realistic buying motives, objections, and negotiation styles—directly addressing a core pedagogy gap: sales professionals need 20-30 practice repetitions to develop confidence, but traditional training provides 2-3. New market entrants (Deal/Spar negotiation simulator, Tavus video agents with ASU Thunderbird and Stanford Law deployments) signal ecosystem expansion into professional negotiations and legal training.
Medical education deployments show mixed progress. The U.S. Military Health System has deployed AI-driven standardized patients and personalized tutors across Uniformed Services University and Army Medical Center, indicating large-scale institutional adoption in medical training. Peer-reviewed research shows specific gains: a Charité Berlin study of 162 medical students using role-prompted GPT-4o across 4 clinical scenarios (shared decision-making, motivational interviewing, sexual health, breaking bad news) documented self-assessed communication competence increases of 0.94 points with 7.92/10 feedback utility. A surgical simulation market analysis projects growth from $176M (2025) to $349.4M (2030) at 14.7% CAGR, driven by AI-powered performance analytics. However, critical evidence reveals significant limitations: a meta-analysis of 4 surgical simulation studies (268 participants) found AI tutoring showed minimal clinical benefit (0.20 OSATS improvement, uncertain significance) while increasing cognitive load. A Nature Medicine correspondence documents a fundamental barrier: digital simulations fail to transmit "tacit learning"—embodied judgment, physical intuition, and pressure calibration necessary for clinical competence—despite detailed VR/AR programming. Historical context: 437 clinicians across 18 countries completed AI-powered XR intubation simulations with 71% rating high effectiveness; GPT-4 virtual patient systems showed gains in student comfort; a large oncology RCT (124 residents, 3 hospitals) demonstrated superior mastery and 91% knowledge retention at 3-month follow-up. Yet unstructured LLM integration in trauma training found no performance improvement and lower teamwork scores, confirming that implementation design—pedagogical structure, human integration, scenario fidelity—matters as much as the AI itself. A research synthesis of 45 papers on AI in simulation-based medical education documented applications across scenario development, adaptive feedback, and personalized learning, while explicitly identifying barriers: ethical concerns, cost, infrastructure, and insufficient AI literacy among educators.
Design principles for effective simulations are becoming clearer. Evidence-based practitioner frameworks identify five key factors: authentic scenario fidelity (mirroring real situations), branching with meaningful consequences, immediate granular feedback, calibrated difficulty with deliberate practice, and spaced repetition. These principles apply across modalities—AI-powered conversation simulations, branching video scenarios, AR overlays, and full VR—with ROI highest for sales, customer service, safety, and onboarding contexts. A recognized L&D expert analysis notes that effectiveness requires deliberate practice conditions: effectiveness varies dramatically by context (sales performance improved 7-35% with specific conditions like low prior performance, high goals, strong supervision), and AI augments rather than replaces human coaching. The pedagogical shift is profound: instead of passive content or rare role-play exercises, simulations enable repeatable deliberate practice at scale, a capability decades of traditional role-play could not provide. Adoption metrics show 52% of companies plan AI L&D integration with projected 30% soft skills improvement and 40% training time reduction; market projections show the virtual character market growing from $10.7B (2022) to $107B by 2032.
Regulatory and integration constraints are sharpening. The EU AI Act (fully applicable August 2026) classifies educational AI as 'high-risk,' requiring traceability, data quality, transparency, and human oversight—fundamentally changing deployment pathways for training providers in regulated markets. Beyond regulation, healthcare and education at scale remain in cautious evaluation; systematic reviews confirm positive student experiences but flag cost, validation rigour, and data privacy as unresolved barriers. Teacher education researchers found that while AI feedback can cost-effectively scale training in mixed-reality simulations, human experts still provide more nuanced pedagogical guidance, especially in identifying missed teaching opportunities and balancing classroom dynamics. A critical research finding documents a fundamental design tension: alignment training that improves AI safety systematically degrades human behavior simulation fidelity—a constraint affecting all conversational simulations. The broader enterprise AI context remains constraining: McKinsey found 73% of AI pilots fail to reach production, and MIT analysis showed only 5% of organisations achieve full-scale GenAI rollout. Simulated practice environments outperform that baseline where the use case is tightly scoped and integration carefully designed, but scaling beyond proven niches remains the central challenge.
— Market adoption snapshot: 30% soft skills improvement, 40% training time reduction, 52% of companies plan AI L&D integration by 2025, 300% ROI within 2-3 years for early adopters, virtual character market projected $107B by 2032.
— U.S. Military Health System deployed AI-driven standardized patients across Uniformed Services University and Army Medical Center; scenario-based training for combat medics with personalized tutors. Large-scale institutional deployment integrating AI simulation into medical education pipeline.
— Peer-reviewed intervention (162 medical students, Charité Berlin) showed role-prompted GPT-4o as simulated practice partner across 4 clinical scenarios increased self-assessed communication competence by 0.94 points (Cohen d=0.58) with 7.92/10 feedback utility rating.
— Expert synthesis positioning digital twins as high-fidelity virtual models enabling risk-free practice; Duke University example of surgeons simulating vascular procedures before operating on actual patients, optimizing surgical plans.
— Nature Medicine correspondence (Hebrew University) documents fundamental gap: digital simulations fail to transmit embodied judgment and physical intuition ('tacit learning') necessary for clinical competence despite detailed VR/AR programming—critical limitation for medical simulation fidelity.
— Implementation guide on AI-powered roleplay platforms for customer service training; distinguishes simulation-based practice (active handling of realistic situations) from knowledge transfer, cites HBR research on experiential learning improving skill acquisition.
— Product GA launch (June 2026) for multi-modal negotiation simulator with configurable negotiation styles, scenario customization, AI-generated performance feedback. Signals ecosystem maturity in negotiation practice tooling and commercial viability.
— Meta-analysis (268 participants, 4 studies) found AI tutoring for surgical simulation showed minimal clinical benefit (0.20 OSATS improvement, uncertain significance) while increasing cognitive load—negative signal on limitations of unstructured AI integration.