Evan Shellshear spends most of his time working on AI, optimization, and large-scale transformation projects at BCG X, the technology innovation arm of Boston Consulting Group. His work spans mining, retail, supply chains, and industrial operations, often on problems large enough that technical complexity alone becomes an organizational challenge. Currently, he focuses on what repeatedly separates successful AI deployments from expensive pilots that never scale.
One of Shellshear’s strongest arguments is that most organizations focus on the wrong part of AI transformation. At BCG X, teams follow what he calls the 70-20-10 principle. Roughly 70% of the effort lies in people and organizational change, 20% in data and infrastructure, and only 10% in algorithms. Even technically strong systems fail if frontline teams do not trust them, workflows are not redesigned, or leadership has not prepared the organization to operate differently.
That lesson becomes more visible in projects on an industrial scale. Shellshear discussed work with Rio Tinto’s mining operations in Western Australia, spanning an area roughly the size of Spain, with autonomous trains and trucks, thousands of kilometers of rail, and dozens of schedulers coordinating activity around the clock. The technical problem of optimization was difficult, but he argues that adoption mattered more. New systems had to run alongside existing ones long enough for operators to trust them before the organization could switch over.
Shellshear is also unusually direct about pilot purgatory, a problem familiar to most enterprise AI teams. Organizations run dozens of pilots, see technical promise, and still fail to create measurable value. His argument is that the failure often begins too early. AI projects should start with business strategy and commercial value, not with a tool searching for a use case. Before building anything, BCG X often works to prove where value leakage exists and whether the opportunity is large enough to justify change. Once executives understand the financial upside, change management becomes easier to justify.
In practice, he suggests meaningful performance improvement usually requires redesigning how teams work, how decisions are made, and how responsibilities shift over time.
AI programs rarely fail because the models are weak. They fail because organizations underestimate the operational changes required to make them work. Technical capability matters, but sustained value often depends on adoption, incentives, training, leadership support, and the willingness to redesign systems around new ways of working.
Market Pulse
Research from MIT found that 95% of enterprise AI investments fail to generate measurable profit-and-loss impact. The issue is rarely just model capability. Most failures stem from weak integration into workflows, unclear business objectives, and poor change management. The companies seeing stronger results tend to focus on specific operational bottlenecks and redesign how work gets done around the technology.
Celonis’ 2026 Process Optimization Report points to a growing gap between AI ambition and operational readiness. Among 1,649 business leaders surveyed, 85% want to become an agentic enterprise within three years, but 76% say their operations are not ready, and only 19% are currently using multi-agent systems. The pattern is familiar. Most organizations are moving faster on AI strategy than on the underlying process clarity required to make automation work at scale.
Salesforce’s latest State of IT Security report suggests security teams are becoming more comfortable with AI while governance challenges continue to grow. Based on responses from more than 2,000 security, privacy, and compliance leaders, the report highlights rising pressure from identity complexity, regulatory requirements, and distributed environments. AI adoption in security is rising, but operational discipline around permissions, visibility, and accountability is becoming harder.
PwC’s 2026 AI predictions argue that enterprises are entering a more disciplined phase of adoption. Broad experimentation is giving way to narrower investments tied to measurable business outcomes, operational metrics, and defined ownership. Companies are becoming more selective about where AI is applied, with a stronger focus on workflow redesign and proof of value.
Resources and Events
📅 The AI Conference 2026 (San Francisco, CA - September 29-October 1, 2026)
A technical and enterprise-focused AI conference covering production AI systems, deployment reliability, infrastructure, evaluation, and operational implementation. The 2026 program includes sessions on enterprise AI adoption, applied machine learning, model evaluation, and the operational challenges of scaling AI systems inside large organizations. Details →
📅 Data & AI Conference Europe 2026 (London, UK - November 2-6, 2026)
A European conference focused on enterprise analytics, AI implementation, business intelligence, governance, and operational AI systems. The event brings together analytics leaders, AI practitioners, data teams, and enterprise technology executives to discuss deployment patterns, data strategy, decision-making systems, and large-scale AI adoption across industries. Details →
📊 Report Spotlight: 2026 CEO Study (IBM)
IBM’s 2026 CEO study suggests the companies moving fastest on AI are changing how leadership works. Based on a survey of 2,000 CEOs across 33 geographies and 21 industries, the study found that 69% of CEOs say AI is already changing core parts of their business, while 77% say technology and leadership roles are converging. The strongest signal is organizational. Companies with an AI-first leadership structure have scaled 10% more AI initiatives, and 76% of organizations now have a Chief AI Officer, up from 26% in 2025. CEOs also expect AI to make 48% of operational decisions by 2030, up from 25% today, suggesting that operating models, accountability, and decision rights may matter as much as the technology itself. Read →
The DecideWise Edge
Benjamin Baer, Founding Member of the DecideWise community, argues that decision management and knowledge management should work together, but most enterprise KM systems have drifted toward AI-enabled search and chat rather than true decision memory. The core point is that companies do not only need faster access to documents. They need a way to capture why decisions were made, who owned them, what changed, what outcomes were expected, and whether those outcomes actually happened. Without that archive, organizations repeat old mistakes because the reasoning behind past choices disappears from the system. For enterprises, the opportunity is to connect decision intelligence with change history, simulation outputs, and business context so future decisions are informed by what the organization has already learned. Read More →
For the Commute
AI Is Only as Smart as Your Documentation (Decision Intelligence Lab)
Carlos Zetina, an industrial engineer, former Amazon research scientist, and pre-sales consultant at FICO, explains why AI systems depend heavily on the quality of the business knowledge they are given. If policies, rules, workflows, and decision logic are poorly documented, AI will reproduce that confusion rather than fix it. The takeaway is that better AI starts with clearer documentation, stronger process knowledge, and explicit decision rules before teams automate work.
