Sangha Penesetti: Reinventing Enterprise’s Future Through Flexible Work

Your next AI pilot will not fix an operating model that cannot scale it. Two rooms. Two cities. One verdict. In this solo episode of Scouting for Growth, I connect the conversations from AI Horizon 2030 in Las Vegas and Majesco’s Frontier Customer Summit in New Orleans, where two very different groups arrived independently at the same conclusion: AI capability is increasingly available. The operating model required to turn it into measurable results is not. The evidence reinforces the point. Research conducted with my students at Bayes Business School found that 26 of 30 leading carriers have yet to cross the halfway mark on both the People and Operating Model transformation axes. In Las Vegas, executives rated their own readiness to run agents in production at just over three out of five. The technology is moving quickly. Most organizations remain stuck in the middle. But readiness is not simply about becoming more technologically sophisticated. Boston Mutual moved from nine-day policy issuance to same-day issuance by redesigning the work around its platform. Manulife transformed a heavily customized model into something repeatable and scalable, while increasing straight-through processing for digital claims from 80% to 98%. What matters is not another isolated capability. It is whether the organization can redesign workflows around the outcomes intelligence now makes possible. I frame the practical response around three moves: name the accountable owner, choose the operating model before the model, and make trust demonstrable. Every agentic initiative needs a named executive with authority over the outcome. AI needs to sit inside an end-to-end workflow rather than beside it. And organizations need to be able to explain who authorized an action, whether it was permissible, what produced the decision, and which human ultimately carries accountability. Because when intelligence becomes widely available, access to intelligence stops being the differentiator. The advantage moves to the organization around it: how decisions are made, how humans and agents share work, how trust is governed, and how quickly a promising experiment can become a repeatable operating model. KEY TAKEAWAYS I want to challenge the assumption that another AI pilot will move an organization closer to becoming AI-ready. The conversations in Las Vegas and New Orleans pointed to the opposite conclusion. Most insurers do not appear to be constrained primarily by access to models, vendors, or technical capability. They are constrained by how work is organized around those capabilities. When intelligence becomes increasingly inexpensive and widely accessible, the more important question moves from “Where can we use AI?” to “How must the work change?” That shift starts with the customer workflow and works backwards. This is why accountability, workflow design, and trust have to be treated as part of the operating model rather than as implementation details. Every agentic initiative needs a named owner with the authority to change what happens next. The AI itself needs to operate inside a redesigned end-to-end process, supported by clear identity, entitlements, decision rights, and auditability. And trust must be demonstrable. If an agent takes an action, leaders should be able to explain who authorized it, whether the action was permissible, which data and model informed it, and which human carries accountability. For me, the principle is increasingly human in the lead, not simply human in the loop. The practical implication is deliberately simple: start with one owner, one process, and trust you can prove. Boston Mutual and Manulife show why this matters. Their progress did not come from layering intelligence over unchanged workflows. They redesigned the work and connected operational improvement to a better customer or member outcome. Readiness is therefore not a product an enterprise can purchase. It is a leadership choice about how the organization

Insurance has become extraordinarily good at finding the few. The next competitive advantage may come from clearing the many. In this episode of Scouting for Growth, I sit down with Manjit Rana, Executive Vice President for Insurance across the UK, EMEA and APAC at Clearspeed, to explore a deceptively simple question: can insurers establish trust at the same speed at which they are automating decisions? More than 90 out of every 100 customers coming to an insurer are genuine. Yet much of the industry's fraud infrastructure has historically been designed around identifying the minority that may represent risk. Manjit's argument is not that those systems were wrong. They were built for a different information environment. Today, AI is accelerating both the automation of insurance decisions and the ability to manufacture convincing evidence, making the trust problem more urgent. We explore the emerging concept of a Trust Intelligence Layer: a continuous, voice-derived, regulatory-ready risk signal designed to help genuine interactions move forward quickly while directing scarce human judgement toward the cases requiring closer attention. Manjit's airport metal detector analogy captures the distinction well. The signal does not make the final decision. It identifies where something unexpected deserves a closer look, allowing everyone else to keep moving. The implications extend well beyond claims. Historical data and behavioural profiles can indicate risk, but they cannot necessarily reveal what an individual knows about an application, renewal or claim happening right now. As insurers automate more of those interactions, the challenge is to preserve trust without simply adding more friction, more profiling or more manual intervention. Most importantly, the conversation moves from technology to operating model. If you could clear more genuine customers, identify more risk and make decisions faster, what would that change about your people, processes, workflows and economics? Manjit argues that this is where leaders should start. Because the question is no longer whether insurance will automate. It is whether insurers can tell which of those automated decisions deserve their trust. KEY TAKEAWAYS I want to challenge the way we have traditionally framed fraud and trust in insurance. For decades, we have become increasingly sophisticated at identifying the interactions that may represent risk. But if the overwhelming majority of customers are genuine, there is another side to the equation: how quickly can we establish sufficient trust in those customers to let them move forward? The distinction matters because every genuine interaction held up by unnecessary scrutiny creates friction, cost and a poorer customer experience. The opportunity is not simply to find more needles. It is to become much better at clearing the hay. This is why we explore the Trust Intelligence Layer. The principle is not to replace existing fraud systems or human judgement with another automated decision-maker. It is to introduce an additional signal that helps determine where human attention may be valuable. That also exposes the limitations of relying exclusively on historical data and profiles. Data can tell us a great deal about patterns and previous behaviour, but it cannot necessarily tell us what an individual knows about the interaction taking place right now. For me, that is also where fairness and customer experience begin to converge: the closer we can get to assessing the interaction itself, the less we need to rely on broad proxies for suspicion. The AI implications make this urgent. As underwriting, claims and customer journeys become faster and increasingly automated, trust cannot remain a slow layer sitting outside the process. It needs to operate at the same speed. But leaders should not begin by asking whether a particular technology works. They should first ask what happens to the business if it does. If more genuine customers can

AI is changing the economics of business. But it is not rewriting the fundamentals of strategy. In this episode of Scouting for Growth, I sit down with Professor Gianvito Lanzolla, Professor of Strategy at Bayes Business School, advisor to boards and CEOs, and co-author of Diversification in the World of Data and AI, to examine what actually creates competitive advantage when access to intelligence is rapidly becoming ubiquitous. The uncomfortable truth is that better technology alone does not create a better strategy. Being early to AI can accelerate learning, but scaling before you understand how the technology creates value, how the organization captures that value, and what customer problem it solves can simply mean buying expensive learning. Gianvito makes an important distinction between being early for learning and being early for scaling. We then explore why data in context matters more than data volume. Tesla's understanding of driving behavior can translate naturally into insurance because the data is relevant to risk. Tesco's knowledge of what customers buy did not automatically reveal what they wanted to watch. The strategic question is not how much data you hold, but what the data describes, how it was generated, and which decisions it can improve. That same discipline applies to diversification and governance. Companies still need a credible right to win, supported by continuous data in context, domain expertise, and the ability to orchestrate an ecosystem. And as AI increasingly influences or makes decisions, governance cannot remain a sequence of static gates. Gianvito argues for adaptive governance built around monitoring, feedback loops, escalation mechanisms, and the ability to revisit decisions as the technology evolves. Most importantly, the conversation brings strategy back to fundamentals. If AI ultimately becomes more like electricity than a source of differentiation in itself, advantage will come from the business model built around it. Gianvito leaves leaders with three questions: Value, Visibility, and Viability. What new customer problem can you solve? What can you see or do now that was previously impossible? And what can you organize and execute economically today that you could not before? Because AI may change the context, the economics, and even the architecture of decision-making. But competitive advantage still depends on understanding where value is created, why you have the right to capture it, and how you build an organization capable of doing so. KEY TAKEAWAYS I want to challenge the assumption that moving first on AI automatically creates an advantage. Gianvito's distinction between being early for learning and early for scaling is particularly important. Organizations should experiment early because they need to build the capacity to understand and implement the technology. Scaling is different. That decision requires a credible hypothesis about the business model: how does AI create value, how will the company capture it, and what becomes possible economically that was not possible before? Access to increasingly powerful models will spread. The differentiation comes from what the organization builds around them. This is also why the conversation around data needs to become more precise. It is not enough to say that a company has enormous amounts of data. What matters is data in context: what the data describes, how it was generated, and which decisions it can inform. Gianvito identifies three capabilities behind AI-enabled diversification: generating a continuous flow of relevant data, combining AI with domain expertise, and orchestrating the ecosystem required to deliver the proposition. Even then, traditional strategic questions remain. Do you genuinely know something your competitors do not? Do you have a right to win? And is the opportunity economically large enough to justify pursuing? For me, the governance implications are just as significant. As AI evolves and takes

Discovery is becoming abundant. Customer ownership is not. In this solo episode of Scouting for Growth, I go inside the Distribution Intelligence Layer and examine what happens when AI assistants become a new front door to the market. Customers can increasingly discover products, compare options, generate quotes, and decide what to do next before they ever reach a company website. The strategic question is no longer simply how they find you. It is what happens to the relationship when they arrive. The uncomfortable truth is that AI may not remove the broker. It may remove the search box and relocate the customer's first interaction into an interface the enterprise does not control. As search, comparison, form-filling, and routine administration become cheaper, value moves downstream toward advice, trust, complex judgement, service, and renewal. Walmart shows what this shift looks like when an enterprise treats agentic distribution as an operating model rather than a chatbot. Aviva and Plymouth Rock show the same logic beginning to emerge in insurance, using ChatGPT as part of the discovery and quotation journey while connecting customers back into enterprise systems. The visible assistant is only the shop window. Behind it must sit product data, identity, payments, service, governance, and accountability. I frame the Distribution Intelligence Layer around three essential jobs: be found, be buyable, and be remembered. Product information must be accurate enough for an assistant to understand. Secure connections must allow quotation, purchase, and service. And when the customer moves into the enterprise journey, their details, permissions, context, and choices need to move with them. The hand-off determines whether the organization simply receives a transaction or begins a relationship it can continue through service and renewal. Most importantly, I leave leaders with three practical moves: try to buy your own product through an AI assistant, make one product AI-ready from discovery through service, and map the complete hand-off from the customer's first question to renewal. Because being found creates an introduction. The real commercial advantage lies in receiving the customer, serving them, and earning the right to keep the relationship. KEY TAKEAWAYS I want to challenge the idea that agentic distribution is simply another digital channel. When an AI assistant can understand the customer's problem, narrow the available choices, and send the business a customer with stronger intent, it begins to change where the market starts. Discovery becomes cheaper, and some of the value historically attached to search, comparison, form-filling, and administration moves elsewhere. This is why I do not believe the broker simply disappears. The broker's value moves downstream toward advice, trust, judgement, and ongoing service. This is why I frame the Distribution Intelligence Layer around three jobs: be found, be buyable, and be remembered. Discovery is only the first leg of the journey. The hand-off into the enterprise must carry customer data, identity, permissions, consent, context, and choices while connecting securely to payment, service, auditability, human accountability, and renewal. The distinction is important because customer ownership is not automatically surrendered to the platform. It is shaped by how the journey is designed and by who controls what happens after the introduction. For me, the practical implication is that leaders should stop debating this shift in the abstract and experience the journey themselves. Try to buy your own product through an AI assistant. Make one product AI-ready from discovery through service rather than attempting to transform the entire portfolio at once. Then draw the complete hand-off and identify who owns the data, identity, payment, service, renewal, and next conversation at every stage. For founders, the same test applies: an impressive interface is not enough unless it can co

Four excellent tools can still produce nothing. Intelligence only creates value when it connects, compounds, and scales. In this solo episode of Scouting for Growth, I introduce the Enterprise Intelligence Stack, a practical framework for understanding why promising agentic pilots stall, why different AI agents can operate with completely different views of the same customer, and why the third pilot can end up costing as much as the first. The uncomfortable truth is that most enterprises are not suffering from a lack of capability. They are suffering from a lack of composition. Data, models, platforms, people, and governance may all exist, but they rarely operate as one coherent system. The result is fragmented intelligence, duplicated infrastructure, weak accountability, and pilots that work beautifully in isolation but struggle to scale. I break the Enterprise Intelligence Stack into four layers and two rails: Data and Judgment; Models and Orchestration; Distribution and Access; Customer and Experience; supported by the People and Trust rails. I then introduce a four-part Composition Test built around one shared source of truth, one registry, one audit trail, and one accountable human. Most importantly, I leave leaders with three practical moves: draw the stack, run the Composition Test, and fund the rails before funding the next layer. Because in the age of agents, the winners will not necessarily be the firms with the most intelligence. They will be the ones whose intelligence connects, compounds, and scales. KEY TAKEAWAYS I want to challenge the way we continue to diagnose enterprise AI problems. We often blame procurement, technology, models, talent, or adoption when an agentic pilot fails to scale. But increasingly, I believe the real constraint is composition. We may already own many of the capabilities we need, but they are not working together as one system. Four excellent tools can still produce nothing if they operate from different data, different logs, different governance models, and different assumptions about the customer. This is why I introduce the Enterprise Intelligence Stack: four layers covering Data and Judgment, Models and Orchestration, Distribution and Access, and Customer and Experience, supported by the People and Trust rails that have to run through the entire system. The distinction is important because layers can be purchased, whereas rails must be designed and governed by the organization itself. The Composition Test then gives us a practical way to challenge every new initiative: is there one shared source of truth, one registry, one audit trail, and one accountable human? For me, the biggest takeaway is that we need to stop asking simply, “What should we buy next?” and start asking, “How will this connect to what we already have?” Whether you are an enterprise leader or an AI venture selling into one, the objective is the same: build intelligence that can connect, compound, and scale. The next move is not necessarily another purchase. It may simply be to draw the stack and understand what you actually own.