Perly Consulting │ Beck Eco

The State of Play

A living index of AI adoption across industries — where established practice meets the bleeding edge
UPDATED DAILY

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.

The Daily Dispatch

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.

AI Maturity by Domain

Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail

DOMAIN
BLEEDING EDGEESTABLISHED

Customer health scoring & churn prediction

GOOD PRACTICE

TRAJECTORY

Stalled

AI that scores customer health, detects churn signals, and triggers proactive intervention workflows. Includes usage-based health scoring and early warning systems; distinct from customer journey analysis which maps experience rather than predicting outcomes.

OVERVIEW

Customer health scoring and churn prediction has matured into a proven capability with GA tooling, documented ROI, and a deep evidence base — yet a persistent gap between technical readiness and execution quality defines the practice today. The underlying problem is well understood: customer success teams need both a prioritisation signal (which accounts need attention now) and a forecasting signal (which accounts will churn). Vendor platforms from Gainsight, ChurnZero, Salesforce, and Microsoft now ship both capabilities as standard features, and peer-reviewed research across 200+ studies confirms ensemble ML models routinely achieve 89-97% prediction accuracy in production. The question is no longer whether the technology works. It is whether organisations can implement it well enough to realise the 15-25% churn reductions that top deployments demonstrate. Most cannot: subjective score weighting, poor data integration, and insufficient calibration mean the majority of deployed health scores underperform simple churn-rate baselines. For enterprises with clean data pipelines, the practice delivers measurable retention and net-revenue impact. For everyone else, the rollout challenge — not the technology — remains the constraint.

CURRENT LANDSCAPE

Gainsight's Staircase AI Health Score, generally available since early 2026, represents the current state of the art: real-time 0-100 scoring that blends sentiment analysis, engagement patterns, and response-time signals with customisable weighting. ChurnZero offers structured health scores mapped to specific churn archetypes, while Salesforce Einstein and Microsoft Dynamics 365 Customer Insights provide embedded prediction engines with automated model retraining. Microsoft's Fabric platform surfaces production-ready tutorials on churn modeling using gradient-boosted trees, demonstrating democratization of technical implementation. The vendor ecosystem is feature-complete and mature. Deployment results at well-resourced organisations are compelling: a mid-market B2B SaaS company reduced first-year churn from 34% to 11%, increased NRR from 96% to 118%, and attributed $8.4M in retained and expanded revenue to unified health scoring with automated interventions; Waystar deployed ChurnZero and achieved 20% churn reduction. A feedback-driven approach using AI analysis of support tickets and survey data achieved a 56% churn reduction (8% to 3.5% monthly) by identifying behavioral patterns and triggering psychological interventions. A systematic review of 142 studies found predictive health scoring achieving 89%+ accuracy with 34-47% NRR improvements in production settings. G2 survey data across platforms documents 15-25% churn reductions (Chargebee up to 25%, Velaris averaging 15%). The global AI-enhanced churn scoring market reached $2.53B in 2025 and is projected to grow 24.5% annually to $3.15B in 2026 and $7.48B by 2030.

The execution gap remains significant, however. Eighty percent of customer success teams remain experimental with AI-driven scoring despite years of vendor investment. Year 1 total cost of ownership runs $60K-$99K for ChurnZero to $90K-$140K for Gainsight, with realistic setup demanding 150+ hours and specialist resources for ongoing calibration. Practitioner assessments reveal that most deployed scores fail to outperform churn-rate baselines — a consequence of subjective weighting and poor signal selection that erodes CSM trust through persistent false positives. Critical implementation analysis shows that 83% precision models fail in production because prediction capability does not automatically translate to intervention execution; deployment gaps include cold-start reliability issues (models unreliable on customers under 14-30 days old), prediction-window mismatches (30-day models surface signals too late for multi-touch retention campaigns), and silent model drift requiring frequent retraining cycles. TSIA analysts warn that many health-scoring pilots lack ROI visibility, and disconnected data systems remain the primary blocking constraint. The practice delivers proven value at enterprise scale but has stalled at the boundary of mid-market adoption, where cost, complexity, and data quality challenges compound.

TIER HISTORY

ResearchJan-2017 → Jan-2017
Bleeding EdgeJan-2017 → Jan-2021
Leading EdgeJan-2021 → Jan-2024
Good PracticeJan-2024 → present

EVIDENCE (123)

— Comprehensive churn metrics framework: logo churn, revenue churn, NRR calculations, and health score construction. Details leading vs. lagging indicators; health scores as proactive intervention signal; 15 documented churn-reduction strategies.

— 8-platform comparison emphasizing signal quality (behavioral data earliest indicator), prediction methods (ML models vs configurable scoring), and activation strategies. Highlights prediction-to-action workflow as differentiator.

— Peer-reviewed research: XGBoost achieved 93.97% accuracy and 0.98 AUC on telecom dataset. Demonstrates practical integration of churn prediction with behavioral segmentation enabling differentiated retention strategies by customer segment.

— Market sizing: $1.62B (2025) to $10.74B (2036), 19% CAGR. SaaS platforms 61.4% revenue. Churn prediction/retention 37.2% of use cases. Large enterprises 64.8% adoption. Confirms ecosystem maturity and vendor consolidation.

Customer Success Software - ChurnZeroProduct Launches

— GA platform with AI health scoring and churn prediction. Vendor-reported outcomes: 60% churn reduction, 65% onboarding improvement, 21% GRR increase, 2x CLV improvement; demonstrates platform feature maturity and ROI potential.

— Critical assessment documenting intervention failures: accurate model (70%+ risk) yielded no better retention than unflagged accounts. Identifies three gaps: scores lack causal drivers, models ignore 80-90% enterprise unstructured data, no actionable next-step guidance.

— Critical analysis: typical health score accuracy ~85%; composite scores (4+ dimensions) show 34% better accuracy. Identifies specific churn signals (usage drop 20% over 90 days, support escalation 30%+, exec champion departure 51-65% risk). CSM data fragmentation barrier.

— Multi-year deployment framework with specific outcomes: 7% renewal lift, 15% accuracy improvement via monthly recalibration, 3% fraud reduction. Insurance survey: 64% of leaders view dynamic scoring as critical competitive advantage.

HISTORY

  • 2017: Customer health scoring emerges as a distinct practice with product-level support (Gainsight Sally bot) and real-world deployments (Feedvisor); academic research advances churn prediction methodologies; industry recognition of fundamental tension between subjective CSM prioritization and objective forecasting models limits broader adoption.
  • 2018: Churn prediction research expands beyond telecom into e-commerce and general SaaS; deep learning techniques (Keras, XGBoost) enter mainstream practitioner adoption; novel methods (PU learning) address real-world data challenges; adoption remains constrained by data scarcity and the unresolved CSM-vs.-forecasting tension.
  • 2019: Major CRM vendors (Salesforce, Microsoft) launch productized churn prediction capabilities; custom deployments accelerate across multiple sectors with 90%+ model accuracies; vendor platforms evolve to separate health scoring (CSM visibility) from churn propensity (forecasting), reducing the core architectural tension; adoption broadens beyond early-adopter SaaS into financial services and telecom, though data scarcity and skills gaps remain barriers to widespread custom implementation.
  • 2021: Churn prediction reaches commodity status in enterprise SaaS and CRM platforms; Salesforce Einstein and Microsoft Dynamics 365 embed prediction engines as standard features; practitioner guides document multi-dimensional health scoring approaches and implementation maturity; research literature reviews compare statistical methods for production deployment; adoption barriers shift from technical capability to organizational readiness and data quality rather than algorithm innovation.
  • 2022-H1: Vendor platforms continue feature expansion: Salesforce Einstein adds multiclass prediction and temporal awareness (Projected Predictions) in Spring/Summer releases; Gainsight publishes D.E.A.R. framework for operationalizing health scores at scale; SaaS vendors report 20%+ churn reduction with ML-powered health scoring; segmentation-driven churn models validated across telecom operators; vendor innovation emphasizes lifecycle-stage health scoring to improve CSM targeting and forecast accuracy.
  • 2022-H2: Microsoft Dynamics 365 announces GA predictive churn capabilities in Customer Insights; academic research validates advanced data transformation methods (26% AUC improvements via feature selection); practitioner roundtables surface real-world implementation challenges including data hygiene and tool integration complexity; vendor consolidation accelerates with health scoring becoming standard across SaaS and CRM platforms; market barriers shift from technical capability to organizational readiness and data quality.
  • 2023-H1: Peer-reviewed research validates health scoring as established B2B CS metric with production deployments across SaaS, banking, and subscription sectors; Notion publishes segment-specific D.E.A.R. framework implementation; Salesforce Einstein confirms 80 billion daily predictions including churn; Gainsight redesigns health scoring feature replacing traffic-light models with multi-dimensional approaches; practitioner adoption documented across service industries with churn prediction as core use case; sector expansion includes banking and OTT subscription services with 97%+ model accuracy achieved in production.
  • 2023-H2: Industry adoption metrics confirm 60% of 400+ North American companies use health scores (Gainsight Oct 2023); Klaviyo launches churn prediction in CDP with 70% baseline churn data; peer-reviewed research validates production-grade churn models in banking (GA-XGBoost) and telecom (0.832 accuracy); systematic ML/DL review confirms ensemble methods dominate with technical maturity but persistent gaps in interpretability; critical finding: SaaS survey shows no correlation between health scores and upsell revenue, challenging traditional ROI narratives despite evidence of churn control benefits.
  • 2024-Q1: Systematic review of 212 peer-reviewed papers confirms ensemble ML/DL dominance with 93-97% AUC on production datasets; researchers emphasize profit-based evaluation metrics; verified deployments at enterprises (PTC/ChurnZero) demonstrate automation of retention workflows; vendor CAB surveys predict 2024 emphasis on AI-driven CS and expansion-over-acquisition strategy; Totango and ChurnZero maintain 7.6/10 user satisfaction with 90%+ recommendation rates and 70%+ renewal intent; third-party health score analysis tools (nCloud Integrators) emerge for Gainsight ecosystem, indicating vendor consolidation and platform extensibility.
  • 2024-Q2: Multi-org case studies document real-world deployments at BigTime, Logiwa, and Qualtrics with specific health scoring methodologies; peer-reviewed research achieves 91.66% accuracy with Random Forest on telecom data with 30%+ churn rates; industry benchmark established at 13% median SaaS churn; critical assessment reveals adoption barriers—only 7% of companies actively track health scores and naive prediction-based interventions can trigger unintended churn; experts emphasize proactive onboarding signals over order cadence and highlight need for uplift modeling over classification.
  • 2024-Q3: Microsoft GA launches transactional churn prediction in Dynamics 365 Customer Insights with automated retraining; Gainsight acquires Staircase AI to enhance interaction-based health signals, signaling continued vendor consolidation in AI-driven CS; adoption metrics show 42% of CS teams track health scores; practitioner guides highlight behavioral alternatives when survey data unavailable; McKinsey reports AI churn reduction potential of 15%.
  • 2024-Q4: Cross-sector deployments validate production maturity: NBFC in Sri Lanka achieves 90% accuracy and 20-30% churn reduction with behavioral feature engineering; telecom operators (Viatel) achieve 97.92% precision with LightGBM; analysis across 67 B2B SaaS companies confirms 82% accuracy and 5.2x median ROI from proactive intervention. Vendor innovation continues: Gainsight releases Scorecard Optimizer for improved health scoring. Critical assessment emerges: JoySuite analysis warns of lagging-indicator risk and false-confidence traps in all-green health scores, reinforcing need for qualitative validation. Practice solidifies as table-stakes feature across enterprise platforms with demonstrated business impact, though execution barriers persist around data quality, intervention timing, and comprehensive implementation.
  • 2025-Q1: Vendor platforms integrate AI-driven interaction analysis: Gainsight launches Atlas AI agents and deepens Staircase AI integration for sentiment monitoring and proactive risk detection. Case studies validate mid-market deployments: SmartReach achieves 35% churn reduction (27% to 17.5%) through weighted health scoring; ChurnZero implementation cases document custom model development and automation. Industry adoption survey reveals expansion limits: 70% adoption at enterprises vs. lower mid-market penetration; only 21% incorporated AI despite 87% planning to do so, indicating implementation gap. Critical assessments surface methodology limitations: advocates argue quantitative-only health scores miss sentiment signals and relationship changes, recommending integrated qualitative approaches. Practice demonstrates broad adoption among large companies with emerging focus on behavioral signal integration and intervention optimization.
  • 2025-Q2: Independent case studies confirm production maturity: GitLab publishes internal health scoring methodology with use-case adoption tracking; SaaS companies report health score improvements from 40% to 82% prediction accuracy with 60% fewer false positives; fintech deployments reduce churn from 18% to 14% via early warning systems. Consultant analyses document real-world implementations including global financial services platforms shifting from reactive to proactive CS via segment-specific models and CSM sentiment integration. Industry adoption shows persistent execution gaps: data quality remains primary barrier (incomplete journeys, legacy CRM silos), churn benchmarks vary significantly by sector (6.9% digital media to 25% finance), and qualitative signal integration emerges as critical gap in quantitative-only approaches. Practice demonstrates table-stakes maturity with proven deployment patterns, yet execution barriers and data quality challenges limit broader adoption below enterprise tier.
  • 2025-Q3: Vendor platforms accelerate AI integration: Gainsight's Insight Agent (Staircase AI) GA delivers automated health scoring (0-100) with real-time churn signals and Executive Dashboard for NRR tracking; ChurnZero Success Insights GA enables ML-powered risk detection; Totango launches Unison AI (though Gartner cautions execution lags behind static legacy systems). Market growth accelerates: customer health scoring AI market reaches USD 1.48B with 25.7% projected CAGR through 2033. Deployment evidence remains strong: London fintech achieves 60-day early warning churn reduction (18% to 14%); SaaS companies demonstrate 40-82% accuracy improvements. Critical signal emerges: analyst reviews highlight platform maturity gaps—Totango's AI remains limited despite roadmap claims, signaling uneven vendor execution. Practice solidifies as table-stakes with accelerating AI-driven capabilities, yet mid-market adoption lags enterprise tier; data quality and qualitative signal gaps persist as core barriers to intervention effectiveness.
  • 2025-Q4: Deployment evidence expands across sectors: peer-reviewed research validates 95.13% accuracy in telecom churn prediction; consulting case studies document 260%+ conversion improvements from predictive analytics (Hydrant with Pecan AI); implementation guides establish quantified operational metrics (88% renewal accuracy, 90% health scoring precision, 50% churn reduction). Vendor deployments confirm signal viability: ChurnZero demonstrates engagement-churn correlation through community platform integration. Critical assessment surfaces implementation reality: realistic setup requires 150+ hours and $20K-$50K consulting costs, with data quality challenges and sector-specific churn variation (6.9% to 25%) demanding segment-specific models. Practice demonstrates strong enterprise adoption and deployment maturity with consistent churn reduction outcomes, yet mid-market penetration gaps (21% AI adoption despite 87% planning), execution barriers around data integration and cost, and unresolved ROI clarity on upsell correlation remain key scaling constraints.
  • 2026-Jan: Vendor platforms deepen AI signal integration: Gainsight Staircase AI Health Score (GA) delivers 0-100 scoring with sentiment, engagement, open items, and response time analysis; ChurnZero structures health scores for specific churn scenarios. New case studies validate enterprise deployments (18% to 11% churn reduction, 42% to 68% save rate improvement). Practitioner consensus emphasizes outcome-based design and segment-specific models. Vendor guidance moves toward hybrid scoring (25-50% AI weighting) balanced with traditional metrics. Execution barriers persist: implementation demands specialist resources, false positive rates remain problematic, and opaque scoring erodes CSM trust. Global AI infrastructure adoption remains early-stage (6% of large enterprises, 13.4% of Fortune 500 with LLM tools), constraining sophistication of AI-driven interventions.
  • 2026-Feb: Independent research and practitioner assessments surface critical implementation gaps despite vendor feature maturity. Academic systematic review (142 studies) confirms predictive health scoring achieves 89%+ accuracy and 34-47% NRR gains in production settings, validating capability potential. However, adoption reality diverges sharply: 80% of CS teams remain experimental with AI despite aggressive vendor investment; Year 1 TCO barriers of $60K-$99K (ChurnZero) to $90K-$140K (Gainsight) persist alongside ongoing configuration demands. Practitioner consensus surfaces accuracy and design failures: most deployed scores underperform churn baseline rates due to subjective weighting and poor signal selection. Platform complexity and false positive rates erode CSM trust. Systematic assessments (Gartner, TSIA, G2) highlight that widespread pilot programs lack ROI visibility and data unification remains the blocking constraint. Market evidence of measurable impact (15-25% churn reductions) concentrated at well-resourced enterprises; mid-market adoption remains significantly constrained. The practice demonstrates proven technical capability at scale but persistent barriers to mainstream deployment—execution complexity, specialist resource requirements, and unproven ROI at non-enterprise tiers limit broader market penetration.
  • 2026-Apr: New evidence reinforces both the deployment upside and the execution ceiling. A mid-market B2B SaaS case study documented churn reduction from 34% to 11% with NRR improving from 96% to 118% and $8.4M attributed to integrated health scoring with automated interventions. Market sizing confirmed at $2.53B (2025) growing to $3.15B (2026) at 24.5% CAGR. Against these positive signals, critical practitioner analysis identified core production failure patterns: high-precision models (83% precision) fail because prediction does not guarantee intervention execution, cold-start unreliability persists below 14-30 days of customer tenure, 30-day prediction windows surface signals too late for multi-touch retention campaigns, and silent model drift demands continuous retraining cycles that most teams do not run. Execution gap remains the defining constraint.
  • 2026-May: Latest evidence reinforces persistent execution gap despite improved platform capabilities. Critical practitioner assessments document specific accuracy ceilings (health scores ~85% baseline, 34% better with 4+ dimensions per Gainsight) and intervention failures (accurate models yield no better retention than unflagged accounts without actionable next-step guidance). Broader market evidence confirms adoption concentration: SaaS platforms hold 61.4% of churn prediction market share; large enterprises represent 64.8% of demand. Real-world deployments demonstrate capability: insurance health scoring frameworks yielded 7% renewal lift, 15% accuracy gains, 3% fraud reduction; telecom research achieved 93.97% accuracy with XGBoost. Platform comparison data shows ChurnZero (90% recommendation rate, 71% renewal intent) and Gainsight (86% recommendation, 95% renewal intent) maintain strong user satisfaction; however, 80% of CS teams remain experimental with AI despite years of vendor investment. Implementation reality: typical setup requires 150+ hours and $60-140K annual TCO, with most deployed scores underperforming churn baselines due to poor signal selection and data fragmentation barriers across CRM, usage, and support systems. Practice demonstrates proven technical capability at enterprise scale with documented 15-25% churn reductions, but mid-market adoption remains significantly constrained by cost, complexity, and unresolved ROI clarity on intervention outcomes.