Driving the Systemic Change for AI – with Deborah Golden of Deloitte

Golden identifies two critical roadblocks causing AI project failures: the 'corporate immune system' that rejects novelty, and a 'crisis of imagination' among leaders. 86% of executives report outdated technology infrastructure, yet the deeper issue is trying to run adaptive AI on rigid systems designed for stability and conformity, not probabilistic outcomes.

• •AI thrives on uncertainty but is being force-fitted into business systems built for predictability and determinism
• •Corporate processes, compliance frameworks, and culture are optimized to slow down variance, causing the system to instinctively reject novel AI capabilities
• •Leaders get stuck in 'pilot purgatory' by failing to rethink how work itself should change, focusing on buying AI rather than transforming business models
• •The fundamental mismatch: trying to run an F1 engine on freight train tracks - core infrastructure fights innovation

Organizations that overcome AI stalls fundamentally re-architect three systems: funding models, team structures, and data governance. The shift from project-based to portfolio-based funding enables staged investment with VC-like discipline, while cross-functional 'fused teams' with shared accountability move decision-making to the edges and dramatically increase speed.

• •Move from annual budget certainty to portfolio-based funding with staged investments, balancing near-term results with long-term capability building
• •Create hybrid cross-functional teams (human and digital) where business, tech, and data experts are co-located with shared accountability from day one
• •67% of employees cite silos as major barriers - fused teams eliminate this by moving decision-making to the edges
• •The gap isn't lack of ideas but outdated systems that turn good ideas into missed opportunities

The quality of strategic questions determines the quality of AI strategy. Instead of asking 'how can AI optimize our call center,' leaders should ask 'what business would we be in if we could predict every customer's needs before they knew them.' This shift in narrative automatically generates more far-reaching answers requiring different ROI thinking.

• •Cultural shift requires giving best people license to be 'intelligently wrong' - organizations claim to embrace failure but fundamentally don't
• •Successful companies derisk business transformation, not just technology, by validating business cases with clear paths to production
• •Performance management systems don't allow for valuable learning from failure - they punish being wrong rather than rewarding organizational learning
• •Shifting the conversation quality establishes deliberate ways to get different outcomes from systems, people, and processes

Organizations must distinguish between two types of failure: intelligent failure (desirable, happens at the edge of knowledge through well-designed experiments) and sloppy failure (preventable, from negligence or cutting corners). Success requires metrics that coexist with unpredictability, rapid blameless postmortems, and psychological safety that separates outcomes from people.

• •Intelligent failure results from well-designed experiments pushing boundaries - this should be rewarded; sloppy failure from negligence should be avoided
• •Blameless postmortems often fail because ego gets in the way - requires intellectual honesty separating outcomes from people involved
• •Framework: define hypothesis, test experiment, analyze results, conduct rapid blameless postmortem, integrate adjustments
• •Metrics culture must evolve from deterministic systems to allow for unpredictability while maintaining compliance and goals

Sandboxes fail when organizations don't define their specific purpose upfront. Golden distinguishes between sandboxes for general learning/training versus those for specific R&D hypotheses. Successful sandboxes require controlled environments with clear hypotheses, expected outcomes, and defined variables - whether testing cybersecurity, training super users, or derisking business transformation.

• •Define sandbox purpose before deployment: general learning/training vs. specific hypothesis testing vs. super user training in non-production environments
• •Organizations often change the hypothesis 17 times to get desired answers - sandboxes must isolate variables, measure everything, and time-box experiments
• •Sandboxes derisk entire business transformation (tech, process, people), not just test code in isolation
• •Super user training sandboxes provide hands-on experience after R&D but before production deployment at scale

AI-era leadership requires orchestrating ecosystems rather than commanding from the top. Leaders must play three critical roles: Shield (protecting nascent initiatives from bureaucracy), Translator (bridging technical and business worlds), and Enabler (actively dismantling barriers). Success depends on knowing when to oscillate between these roles based on circumstances.

• •Shield role: protect experimental budgets, create psychological safety for risk-taking, frame failed experiments as valuable learning investments
• •Translator role: bridge technical and business worlds by converting probabilistic AI capabilities into competitive advantage, margin improvement, and revenue language
• •Enabler role: most active non-passive role requiring direct intervention to dismantle friction, break down silos, and rewire outdated policies holding teams back
• •Leaders must oscillate between roles - not be 100% of everything all the time, but adjust percentages based on situation

Golden's final advice: AI isn't fundamentally a technology challenge but a catalyst forcing overdue conversations about disruption. Leaders across industries are anxious about not knowing where to begin, but that's acceptable - AI is built for unpredictability in an unpredictable world. The opportunity is using AI as impetus to build faster, smarter, more adaptive organizations rather than just deploying another technology.

• •It's okay to not know where you're going - AI is built for unpredictability, and the world is in a moment of unpredictability
• •Use AI as a mirror to examine legacy structures, siloed cultures, and slow decision-making processes
• •This isn't about deploying yet another new technology - it's about building fundamentally more adaptive organizations
• •Leaders should embrace that others are equally confused and use that shared uncertainty as starting point for transformation