The State of Play

Feature prioritisation is the practice of systematically ranking product features and roadmap initiatives based on customer signals, business impact, and effort estimates. Rather than manually juggling priorities or defaulting to loudest-voice decision-making, teams use frameworks like RICE (Reach, Impact, Confidence, Effort) or ICE (Impact, Confidence, Effort) to make explicit trade-offs. AI is emerging as a tool to accelerate this process—synthesising customer feedback into prioritisation signals, modelling roadmap scenarios, and suggesting effort estimates based on historical data. However, the practice remains constrained not by AI capability but by organisational execution discipline. The core tension: vendor tooling matured to production-grade agentic workflows by 2026, yet only 20-28% of organisations achieve measurable ROI from AI-assisted prioritisation. Successful teams share three practices: they encode organisational context (workflows, decision gates, compliance rules) before deploying AI, they measure success against business outcomes not adoption metrics, and they treat AI as a synthesis layer feeding human judgment rather than as a decision-maker. Teams that skip these prerequisites see 80%+ of AI pilots deliver zero bottom-line impact.

By June 2026, feature prioritisation remains trapped in a capability-execution paradox: agentic tooling reached production maturity (Productboard Spark agents, airfocus Intelligence Platform with strategic drift detection, ServiceNow RICE/WSJF, Atlassian velocity scoring), yet only 20-28% of organisations achieve measurable ROI. Gartner's June survey of 782 infrastructure leaders reported only 28% of AI use cases meet ROI expectations, 20% fail outright; IBM's Q4 2025 CEO study found only 25% deliver expected ROI with 75% failing financially due to organisational barriers not technology gaps. Real-world outcomes vary dramatically: EXANTE (fintech) compressed competitor research cycles from 1 week to 1 day and interview synthesis from days to hours; text analytics deployments achieved 35% backlog reduction with 12 NPS gains and $50M recall prevention; CloudSync compressed research cycles from 3 hours to 15 minutes. Yet 95% of AI pilots deliver zero P&L impact because velocity without prioritisation discipline ships unused features—the constraint since 2024.

Verification emerged as the binding constraint on AI-assisted prioritisation in 2026, compounded by tool adoption barriers. PM work shifted from creation (now automated by AI) to judgment: verification of synthesis accuracy, assessment of strategic alignment, risk-checking across functional boundaries. When inference cycles accelerate from weeks to days, human review capacity becomes the bottleneck. Critical gap revealed in category-leading tool adoption: Productboard's structure makes feedback capture effortless but prioritisation requires manual PM scoring per feature; when backlogs exceed 100 items, scoring stops and features accumulate with no signal, creating a "feature graveyard" documented in real user reviews. Framework selection now determines value realisation: McKinsey research shows organisations choosing a single value path and aligning execution achieve 2-4x ROI; McDonald's failure case (85% technical accuracy but catastrophic 15% error rate) exemplifies the risk of misaligned success metrics. Successful teams treat AI as a synthesis layer feeding human judgment: they retain cross-functional review that adjusts for strategic alignment, platform health, compliance, and opportunity cost—factors AI systematically underweights.

Operating model readiness remains the blocking constraint. McKinsey analysis of 80% of failed agentic AI investments identified three missing architectural decisions: encoding organisational context (decision gates, compliance rules, pricing logic) before deployment, building reusable skill components that compound expertise, and designing governance into architecture from the start. Field reports document the cost of skipping these: AI generates 50% revision work by skipping undocumented process gates and optimizing for completion over correctness. The vendor ecosystem bifurcated in 2026: Productboard evolved Spark as agentic copilot (AI augments PM-driven workspace), while competitors like Ferrix positioned agentic-native architectures (agents synthesise and score, PM approves). Framework assumptions invalidated by AI: build effort is no longer the constraint; ongoing inference costs (monitoring, moderation, rollback planning, compliance review) and data readiness now gate prioritisation decisions. Honest net ROI accounting for implementation and change management reveals realistic returns at ~10% vs. vendor-claimed figures; 95% of AI projects fail to show measurable returns within 6 months. The practice stalled because organisations optimised for adoption breadth (73% weekly AI use reported) rather than decision quality (only 11.5% report confident prioritisation decisions), exposing a strategic execution gap that tooling maturity alone cannot solve.

• 2023-H1: Productboard and other vendors launch AI capabilities for feedback analysis and prioritisation. Path & Planning case study documents real deployment using Productboard for OKR-aligned roadmap centralisation and improved forecasting. Broader survey data shows widespread caution about AI project ROI and high failure rates in enterprise AI initiatives.

• 2024-Q1: Vendor ecosystem maturation: Productboard Spark (weighted scoring) enters beta, Strive and competitors expand. Société Générale case study shows structured deployment with 100+ use cases and closed-loop value tracking, but enterprise adoption remains cautious—surveys show ~70% of generative AI projects fail to deliver value. Product strategists emphasize that framework effectiveness depends on strategic clarity and disciplined implementation, not AI alone.

• 2024-Q2: Adoption accelerating to mainstream: 61% of PMs now report using AI/ML in their workflows. Zefi AI launches as dedicated VoC platform with roadmap prioritisation as explicit use case. Documented efficiency gains show 25-30% improvement in product development cycle speed. Cautionary data persists: enterprise deployments continue to struggle with data quality and ROI validation; priority framework effectiveness tied to strategic clarity rather than tool capability.

• 2024-Q3: Vendor consolidation and cautionary signals: Productboard re-architects platform and scales for enterprise deployment (Salesforce, Zoom, Pitney Bowes confirmed as users). However, Gartner forecasts 30% of GenAI projects will be abandoned by EOY 2025 due to poor data quality, cost, and unclear ROI. Generative AI adoption reaches 39% of U.S. workforce (Harvard Kennedy School survey, Aug 2024), but real-world deployment of AI-assisted prioritisation continues to be hampered by data quality issues and integration complexity. The gap between vendor capability and reliable enterprise implementation widens.

• 2024-Q4: Vendor maturity and production readiness gap widen: Productboard launches Pulse AI for Voice of Customer integration. Enterprise AI spending surges to $13.8B (6x from 2023); 85% of enterprises testing GenAI. Yet deployment stalls: only 22% confident in IT architecture; 60% of UK enterprises not in production; AI project ROI declined to 47.3% from 56.7% in 2021; data quality and governance cited as leading obstacles. Workforce sentiment cools (excitement drops 47%→41%), with 48% of workers uncomfortable admitting AI use. Feature prioritisation remains trapped between vendor maturity and operational complexity.

• 2025-Q1: Vendor momentum continues but deployment crisis deepens. Productboard releases Spark AI suite with agentic prioritisation capabilities; Productboard CEO emphasizes strategic integration over AI-first hype at SaaStr Summit. Simultaneously, S&P Global reports failure rates surge to 42% (up from 17%), with 46% of AI pilots failing to reach production. Industry analysis reveals 60-95% of AI initiatives stalled in "Pilot Purgatory"; tech leaders cite reliability concerns (45%) and integration challenges as top barriers. The market reached inflection point: vendor tooling matured while enterprise execution deteriorated, widening the proof-of-concept-to-production gap.

• 2025-Q2: Failure acceleration and ROI crisis materialize. Product leaders report 70% are investing in AI/ML, with 75% recognizing AI/data fluency as critical PM competency—yet simultaneous collapse in execution: 42% of companies scrapped most AI initiatives (vs 17% year prior), and 46% of POCs abandoned. MIT research shows 95% of AI pilots fail to scale; 70% of all AI initiatives never escape pilot phase. Only 4% of companies achieve significant AI returns; average ROI 3.7x but 66% struggle with positive ROI. The practice reaches a critical juncture: vendor maturity is proven (Productboard, Airfocus, Craft.io firmly established), feature prioritisation frameworks are understood, but enterprise execution remains fundamentally constrained by data quality, integration complexity, and measurement discipline—not technology capability.

• 2025-Q4: Inflection reached. Deloitte survey of 1,854 execs: only 6% hit satisfactory AI ROI within a year. Productboard survey: 99% of PMs experimenting with AI but only 8% say it's core to prioritisation. High-profile failures documented: Volkswagen Cariad ($7.5B loss), Taco Bell drive-thru (viral failures). UserIntuition analysis: 64% of delivered features miss adoption targets because frameworks amplify bad input, not solve it. The paradox crystallizes—tooling matured, but organisations remained trapped by the same constraint: quality of data and strategic clarity, not technology capability.

• 2026-Jan: Vendor production readiness confirmed, adoption barriers harden. Productboard publishes case study of Pulse AI (processing 200k-1M feedback items) and Spark agentic system in production, but UC Berkeley research finds only 5% of enterprises see P&L impact from gen-AI and AI tooling can increase task completion time 19%. Data quality, governance gaps, and organisational misalignment emerge as primary adoption blockers—not technology maturity.

• 2026-Feb: Deployment evidence and ROI reality collide. P&G field experiment shows AI-enabled teams 3x more likely to produce top-tier ideas with 13-16% faster ideation cycles, confirming the capability exists. Yet KPMG's 2,500-executive survey reveals only 24% achieve ROI across multiple AI use cases, with high performers at 4.5x ROI and the majority struggling. Leading product discovery teams demonstrate workflow compression (50+ steps to 18), but strategic misalignment persists: feature prioritisation remains constrained by vendor lock-in fears (94% of IT leaders), roadmap optimization misconceptions, and the fundamental tension between tool maturity and organisational execution discipline.

• 2026-Mar: Vendor ecosystem expansion and deployment validation. ServiceNow embeds RICE/WSJF scoring, Koji launches AI-moderated research conversion, CloudSync case study shows ARR-based prioritisation ($513K SSO > $490K API). IdeaPlan survey of 1,200+ PMs: 73% weekly AI use, 31% for roadmap narratives, 5-8 hrs/week savings. Yet organisational readiness gaps widen: Deloitte survey (3,235 leaders) shows 88% use AI but only 20% achieve revenue growth; governance and data readiness declining. Wire analysis reveals critical flaw: AI tools access 1 of 5 context dimensions, causing failure modes (keyword frequency ranking SSO above onboarding without contract context). The paradox persists: capability maturity confirmed, but organisational execution barriers (data quality, strategic clarity, cross-functional alignment) remain unchanged since 2024.

• 2026-Apr: Framework inadequacy, governance failures, and vendor confidence peaks. IdeaPlan documented that standard RICE frameworks fail for AI features—proposing RICE-A with an AI Complexity dimension covering data readiness, model maturity, and operational overhead—amid evidence that 80%+ of AI projects fail and fewer than 20% scale to production within 18 months. Product-Led Alliance's 2026 PM survey found only 11.5% report confident prioritisation decisions despite 73% weekly AI use, confirming that adoption breadth has not translated into decision quality. Mustafa Kapadia's April benchmark follow-up revealed that core product work (roadmap prioritisation, strategic planning) remains <10% of AI use despite 73% weekly adoption—negative signal of stalled strategic execution. Systemic "research breakage" documented as structural governance failure: findings disappear through organisational fog or silent mid-roadmap abandonment. ITONICS analysis showed frameworks fail under organisational pressure (RICE scores become political, MoSCoW politicised, Value/Effort suffers from political feasibility bias). Productboard announced 30% workforce reduction and shift to "AI-only" operating model, signaling vendor conviction in production maturity. MetaCTO consulting documented that 88% of AI POCs never reach production, traditional 12-36 month roadmaps are obsolete, and 80.3% of AI projects fail to deliver intended value (RAND 2025)—critical negative signals revealing execution barriers persist despite vendor maturity.

• 2026-Jun: Verification costs, value concentration, and agentic tool divergence sharpen the ROI reality. Analysis documents the PM work shift: AI removed the creation bottleneck, moving the constraint to verification and judgment—review capacity now limits effective AI-assisted roadmapping. PwC/BCG/McKinsey synthesis confirms the value divide: the top 20% of AI-investing organisations captured 74% of returns, with CEO ownership and workflow redesign as separating factors. Honest ROI accounting—incorporating implementation and change management costs—puts realistic returns at ~10% versus vendor-claimed figures, with 95% of projects failing to show measurable returns within six months. Feature voting boards documented as a structural roadmap liability: 1% engagement dominance biases prioritisation toward vocal minorities; 64% of delivered features see low usage. Productboard Spark GA'd an agentic layer compressing week-long discovery cycles to 90 minutes (Bill.com, Praxedo case studies), while a 2026 architectural fork emerges: Productboard evolves toward AI-copilot in PM workspace, Ferrix and competitors position as agentic-native (agents synthesise, PM approves). IBM CEO study reconfirms 75% of AI initiatives fail financially due to organizational barriers not technology gaps. airfocus launched strategic drift detection surfacing roadmap–OKR misalignment with MCP server exposing prioritisation data to Claude/ChatGPT.

• 2026-May: Agentic deployment evidence, operating model crisis confirmed. Productboard published case studies of Principal PMs at Amplitude and Productboard deploying AI agents for autonomous discovery briefs, metric analysis, and opportunity detection—demonstrating sophisticated production-ready workflows. New prioritisation methodologies emerged: outcome-first KPI scoring frameworks for AI initiatives, text analytics deployments delivered quantified outcomes (35% backlog reduction, 12 NPS gains, $50M recall prevention). Airfocus survey of 500 product professionals confirmed the core paradox: 48% struggle to separate signal from noise in prioritisation despite 73% weekly AI use; Productboard analysis established that traditional discovery frameworks break at scale and AI-synthesized continuous discovery is required to maintain decision currency. Analysis from The Independent quantified the execution failure rate—95% of AI pilots deliver no P&L impact because velocity without prioritisation discipline ships unused features. Userpilot proposed the structural fix: shift from fixed delivery plans to quarterly bet-based decision systems with explicit assumptions and outcome measurement. FAANG framework adoption persisted (57% RICE at Meta/Airbnb/Dropbox, 22% throughput gains with WSJF at Spotify/Amazon). The paradox hardened—agentic capability matured, deployments delivered outcomes, but strategic leverage remained <10% of AI use. Organisational execution barriers (governance, data quality, strategic clarity, cross-functional alignment) remained the constraint.