Machine Learning Applications in Strategy

Strategic planning just got an AI upgrade – and it's a game-changer. Thrilled to share my latest #Fortune column, co-authored with some of my former colleagues at Boston Consulting Group (BCG). The reality: Even the best strategic planning suffers from human limitations – our biases, groupthink, and tendency to anchor future scenarios in past experience. When volatility rises, these constraints become dangerous blind spots. The breakthrough: Multi-agent AI platforms that simulate complex strategic scenarios with human-like behavioral patterns, but without human cognitive limitations. Think of it as having a boardroom full of AI agents – each playing regulators, competitors, customers, and other stakeholders – stress-testing your strategy 24/7 at a fraction of traditional costs. What we're seeing in practice: AI simulations identifying the same strategic moves as human workshops – plus new options humans missed entirely "Unknown unknowns" becoming "known unknowns" through expanded scenario modeling Strategic planning becoming more frequent, scalable, and accessible across organizations Leaders building confidence through pattern recognition across multiple simulation runs This isn't about replacing human strategic thinking. It's about augmenting it with tools that can explore a vastly wider range of futures, faster and cheaper than ever before. In an era where resilience drives outperformance, the organizations that upgrade their strategic planning capabilities first will have the advantage. Read the full piece: https://lnkd.in/eUNDT2WZ #AI #StrategicPlanning #BusinessStrategy #Leadership #GenAI #ScenarioPlanning #DigitalTransformation Leonid Zhukov, Ph.D, Maxwell Struever, Alan Iny Elton Parker David Zuluaga Martínez

AI won’t just automate strategy work. It will redefine how strategy gets made.   I’ve been advising leaders on strategy for over 20 years. And rarely have I seen the fundamentals shift as quickly as they have in the past two years. AI is accelerating classic strategy work: research, analysis, option generation, storytelling.   A new BCG Henderson Institute perspective by my colleagues Prof. Dr. Ulrich Pidun, Ketil Gjerstad, Nina Kataeva, Gabe Bouslov, and Adam Job, PhD puts big numbers behind this: ~80% of typical strategy tasks are highly exposed to AI automation and augmentation   But the bigger shift is more fundamental: AI will change the system of strategy-making itself – how insights are gathered, where strategy is set, the tempo at which it is adjusted.   Three shifts stand out for large companies: ➡️ Supercharged information gathering across divisions, functions, and geographies ➡️ More decentralized strategy-making (within clear guardrails) ➡️ Always-on strategy to reduce the time lag of annual cycles   One aspect stays the same: the need of a guiding North Star for your mid-term strategy.   To make that work, new roles will matter; especially “social strategists” to orchestrate the human side (buy-in, sequencing, adoption) and “technical strategists” to design the decision systems, escalation criteria, and governance.   This is where the human factor becomes decisive. Many strategic choices require judgment and accountability, and the decisions fundamental to competitive advantage must remain human-led.   At #BCG, we see this shift playing out across large companies. The goal is “superhuman” teams: human judgment + AI power, so it’s clear where automation ends and leadership begins.   👉 Full report: https://lnkd.in/dqZt8g43 #ArtificialIntelligence #Strategy #Leadership #Transformation

AI doesn't wait for your yearly review. Neither should your strategy. Static roadmaps are being replaced by living, evolving systems. The shift isn't about more meetings or bigger decks. It's about embedding agility into the core of how strategy is created, tested, and refined in the age of AI. Here are 13 ways leaders are leveraging AI to shape their strategic planning: 1/ Real-Time Monitoring Systems ↳ AI-powered dashboard integration ↳ Automated trend detection 💡Pro tip: Set up 15-minute daily stand-ups focused solely on emerging AI trends. 2/ Rolling Quarter Framework ↳ 90-day action sprints ↳ Monthly strategy refinements 💡Pro tip: Keep 70% of resources committed, 30% flexible. 3/ Scenario Planning Networks ↳ Multiple future state mapping ↳ Risk-opportunity matrices 💡Pro tip: Create 3 scenarios for every major decision: baseline, accelerated AI adoption, and disruption. 4/ Digital Twin Strategies ↳ Virtual strategy modeling ↳ Quick iteration cycles 💡Pro tip: Test strategic changes in digital environments before real-world implementation. 5/ Adaptive Team Structures ↳ Fluid role assignments ↳ Skills-based reorganization 💡Pro tip: Rotate 20% of team members quarterly across departments for fresh perspectives. 6/ AI Intelligence Streams ↳ Automated competitor analysis ↳ Market sentiment tracking 💡Pro tip: Set up AI alerts for both direct competitors and adjacent industry innovations. 7/ Micro-Learning Systems ↳ Just-in-time training ↳ Adaptive learning paths 💡Pro tip: Schedule 20-minute weekly team sessions on new AI tools. 8/ Decision Velocity Framework ↳ Rapid testing protocols ↳ Fast-fail mechanisms 💡Pro tip: Define your "reversal cost threshold" - the point at which a decision needs more review. 9/ Stakeholder Feedback Loops ↳ Continuous alignment checks ↳ Dynamic priority adjustment 💡Pro tip: Create a weekly survey that takes less than 30 seconds to complete. 10/ Resource Fluidity Models ↳ Dynamic budget allocation ↳ Skill-based resourcing 💡Pro tip: Keep 25% of your innovation budget unallocated for emerging AI opportunities. 11/ Crisis-Ready Culture ↳ Rapid response protocols ↳ Distributed decision rights 💡Pro tip: Run monthly "AI disruption simulations" with different teams leading each time. 12/ Data-Driven Pivots ↳ Automated trend analysis ↳ Predictive modeling 💡Pro tip: Define specific metrics that automatically initiate strategy reviews. 13/ Continuous Communication ↳ Strategy visualization tools ↳ Real-time progress tracking 💡Pro tip: Use AI tools to create strategy briefings under 2 minutes. The most resilient teams aren’t the ones with the perfect plan. They’re the ones built to adapt in real time. Continuous strategy isn’t a trend; it’s the new baseline for staying competitive in an AI-driven market. Which of these shifts are you implementing? Share below 👇 _____ Follow Carolyn Healey for more AI and leadership content. Repost to your network if they will find this valuable.

Can AI replace Strategy Consulting? 🤔 I ran an experiment to find out. In 3 weeks and for $10, I tackled a $20B strategic question—the kind that usually takes weeks of consulting and six-figure fees 💸💼 Here’s what I tested 👇 🔍 The strategic question Industry experts argue that Google has a durable moat over OpenAI because of: - Massive distribution 📱 - A vertically integrated stack (from silicon to apps) 🧱 But I wanted to test a different hypothesis: 👉 Google’s real moat might be decades of investment in intent detection—its ability to understand even poorly formed queries. If true, this could: - Give Google a major edge as AI becomes embedded across devices & software - Become a powerful monetization lever for AI agents relying on search-powered RAG 💡 Why this matters Based on the results, this analysis helps answer questions like: - Should Apple reconsider its ~$20B search revenue share with Google? 🍎 - Could Google offer AI Mode as a standalone service for AI agents? - Should product leaders building AI assistants build their own search stack—or just rely on Google? 🧪 The method In one day, I built: - A side-by-side evaluation app - A dataset of 200 real queries - An LLM-based judge - Scripts to analyze results scientifically - All using Perplexity + Claude Code ⚡ 😮 The surprising finding Despite Google’s heavy investments in search (RankBrain, BERT, MUM): 👉 ChatGPT outperformed Google AI Mode across nearly every metric. I’ve shared the full report + methodology in this post 👇 🚀 The bigger learning Strategic decision-making is no longer gated by expensive consulting firms. The tools are now accessible to everyone. I’d love to hear your thoughts 💬 - What’s your take on Google’s intent detection moat? - What strategic questions are you wrestling with right now? - What would you evaluate if cost and time weren’t barriers? - How are you using AI for strategy work today? Drop a comment or DM me—I’m genuinely curious how others are thinking about this 👀✨

Chapter 3 – Online learning Sequential decision problems use machine learning (typically in an online context) in five different ways:   1) Approximating the expectation of a function F(x,W) of a decision x and random variables W. 2) Approximate a policy X^pi(S|\theta) that depends on the information in the state S, and gives the decision x. 3) Approximating the value V(S) of being in a state S. 4) Learning the underlying model, which can include the cost function, the transition function (either of these might not be known), or the exogenous information process. 5) A class of policy known as a parametric cost function approximation, which is a parameterized optimization model (this is distinct from (2) above which does not have an imbedded optimization problem).   The rest of chapter 3 is a tour of the most widely used machine learning tools that are used for sequential decision problems, divided into the following categories:   o Lookup tables, using both Bayesian and frequentist belief models. We cover the powerful strategy of hierarchical learning using adaptive, dynamic weights that allow us to combine estimates at different levels of aggregation. o Parametric models. Recursive updating for linear models is presented. For nonlinear models, we just present the usual gradient-based search algorithms. o Nonparametric models – These use local approximations that might be constant or linear models. I also include k-nearest neighbor, kernel regression, local polynomial regrerssion, and neural networks. In chapter 5 I give the formulas for computing the gradients of neural networks. o Nonstationary learning – When we use machine learning in the context of sequential decision problems, we are sometimes estimating functions that are evolving while we learn. This is particularly important when estimating value function approximations, as well as policies.   Machine learning is a toolbox – readers need to be aware of what is available, but it is not necessary to absorb this entire chapter on a first read.

Machine learning for dynamic pricing optimization offers businesses a competitive edge by enabling them to adjust prices in real-time, ensuring they remain responsive to market demands, customer behavior, and competition, ultimately maximizing revenue and profitability. Machine learning, a subset of AI, allows systems to learn from data and improve without explicit programming, identifying patterns and making predictions from historical data. In pricing optimization, it helps set prices strategically by considering demand, competition, costs, and customer perception. Fundamental data types used include sales history, market trends, competitor pricing, customer behavior, demographics, seasonality, and search trends. Standard algorithms, such as regression, decision trees, neural networks, clustering, and reinforcement learning, are applied to predict demand shifts. Dynamic pricing then adjusts prices in real-time, boosting revenue and competitiveness. For business implementation, ML models can be integrated with existing systems like sales, ERP, and CRM, allowing for real-time price adjustments. Challenges include maintaining high data quality, investing in technology and skills, and addressing ethical and regulatory concerns regarding dynamic pricing, customer perception, and compliance. #ai #MachineLearning #Pricing #CRO #COO

𝗖𝗮𝗻 𝗔𝗜 𝗵𝗲𝗹𝗽 𝗰𝗿𝗲𝗮𝘁𝗲 𝗯𝗲𝘁𝘁𝗲𝗿 𝗽𝗿𝗼𝗱𝘂𝗰𝘁 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗲𝘀 𝗳𝗮𝘀𝘁𝗲𝗿? That's the question I answer in my latest article. Below is a summary of my findings. 𝗧𝗟;𝗗𝗥: AI-based tools can lead to better and faster strategic decisions, at least for certain products. But they are no silver bullet. ✅ Strategy Benefit #1: AI-based tools can discover user and customer trends using predictive analytics. This can help you create a new strategy and evolve an existing one. ✅ Strategy Benefit #2: AI tools can analyse market data, customer feedback, and emerging trends to suggest new products and features, assuming that enough relevant data is available. ✅ Strategy Benefit #3: AI tools can help make products stand out from the crowd by offering AI-enabled product features like Spotify’s DJ. ✅ Strategy Benefit #4: AI tools can continuously monitor how much value a product is creating and recommend improvements. ❌ Strategy Limitation #1: AI tools require enough good-quality data to generate helpful results. ❌ Strategy Limitation #2: Generative AI tools give the most likely answers, not necessarily the correct ones. Consequently, their results might not be to be good enough. ❌ Strategy Limitation #3: AI’s data dependency makes it hard, if not impossible, to use it for disruptive, transformative products. ❌ Strategy Limitation #4: AI-based tools are no replacement for meeting users and customers and understanding their needs. ❌ Strategy Limitation #5: AI tools may increase the environmental impact and carbon footprint of your company and jeopardise your product’s ethicality. 💡𝗧𝗼 𝘁𝗮𝗸𝗲 𝗳𝘂𝗹𝗹 𝗮𝗱𝘃𝗮𝗻𝘁𝗮𝗴𝗲 𝗼𝗳 𝗔𝗜, 𝘆𝗼𝘂 𝗵𝗮𝘃𝗲 𝘁𝗼 𝗵𝗮𝘃𝗲 𝘁𝗵𝗲 𝗿𝗶𝗴𝗵𝘁 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝘆 𝗽𝗿𝗮𝗰𝘁𝗶𝗰𝗲𝘀 𝗶𝗻 𝗽𝗹𝗮𝗰𝗲. Otherwise, you risk being tool-led. In the worst case, you will make the wrong decisions faster. To understand if your strategy practice is in good shape, ask yourself the following questions: 1️⃣ Is it clear who contributes to strategic decisions and who has the final say? Are the right people involved? 2️⃣ Has your current strategy been captured using a tool like the Product Vision Board? 3️⃣ Are strategic decisions based on empirical evidence rather than opinions and beliefs? Do you systematically de-risk a new and significantly changed product strategy? 4️⃣ Do you regularly review the product strategy using the right KPIs? Do you regularly interact with users and customers to understand if their needs are being met? 5️⃣ Is the product strategy aligned with the business and portfolio strategy as well as the product roadmap? 🚀 If your answers are positive, you’re in a good place. Using the right AI tools will most likely benefit you. Otherwise, improve your strategy practice first before you heavily invest in AI. 👉 𝗥𝗲𝗮𝗱 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝗮𝗿𝘁𝗶𝗰𝗹𝗲: https://lnkd.in/eSZBC2wm #poductmanagement #productstaretgy #ai

𝗠𝗼𝘀𝘁 𝗽𝗲𝗼𝗽𝗹𝗲 𝘁𝗵𝗶𝗻𝗸 “𝗔𝗜 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝘆” 𝗺𝗲𝗮𝗻𝘀 𝗽𝗶𝗰𝗸𝗶𝗻𝗴 𝘁𝗵𝗲 𝗯𝗲𝘀𝘁 𝗺𝗼𝗱𝗲𝗹. But in production, the model is just the brain. The real game is everything around it. This AI Strategy Map shows what modern AI actually needs to work inside real businesses - not as demos, but as reliable systems. It breaks AI into layers: - Foundation models (OpenAI, Anthropic, Google, Meta). - Developer frameworks (LangChain, LlamaIndex, Semantic Kernel). - Agentic AI systems (planning, tools, workflows). - RAG + retrieval infra (Pinecone, Weaviate, Qdrant). - Fine-tuning + customization. - LLMOps + evaluation (LangSmith, W&B, TruLens, Ragas). - Safety + governance guardrails. - Compute infrastructure (GPUs, TPUs, cloud AI). - Data engineering stack (Snowflake, Databricks, dbt). - Knowledge graphs + vector search. - Workflow automation (Make, n8n, Zapier). 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗹𝗲𝘀𝘀𝗼𝗻? If your strategy is only “choose a model”… you don’t have a strategy. You have a prototype. The winners will be the teams who build the full stack — so AI can operate, not just respond. ♻️ Repost this to help your network get started ➕ Follow Prem N. for more

#Machine #Learning (ML) has taken the world of empirical asset pricing by storm. But which design choices should you opt for when constructing an ML investment strategy in the US equity market? In a new working paper, Minghui Chen, CFA, Matthias Hanauer, and Tobias Kalsbach (2024) analyze the variation in seven design choices when implementing an ML strategy for individual stocks based on 207 common features. (1): Algorithm (N=11, based on a neural network, gradient boosting, etc.) (2): Target variable (N=3, return over the risk-free rate, return over market, or risk-adjusted return) (3): Target variable transformation (N=2, continuous or discrete) (4): Post-publication treatment (N=2) (5): Features (N=2, pre-selection or all features) (6): Training window (N=2, rolling window or expanding scheme) (7): Training sample (N=2, with or without micro stocks) Interestingly, the authors find that the design choice strongly influences the resulting trading performance of a long-short trading strategy based on ML. A hypothetical USD 1 investment in 1987 leads to a final wealth between USD 0.94 and USD 2,652 (see figure). What a diffeerence! 💡 The best model is associated with design choices based on an ensemble of the different ML algorithms, a continuous target variable based on return over market, features with no pre-selection and post-publication treatment, and an expanding training window on the sample without micro stocks.