How Data Modeling Influences Decision Making

A lot of product teams yearn for a concept of "Isolated Bet Impact." The challenge lies in understanding how individual initiatives influence key metrics within a complex system where multiple factors—both internal and external—interact simultaneously. One of your bets might have a positive impact even though your KPIs are down overall. Conversely, you might have placed a bad bet that is obscured by overall growth in the business. Teams risk falling into "resulting," as the poker player Annie Duke describes, where they mistakenly judge the quality of their decisions solely by lagging business outcomes rather than understanding the true impact of their bets. Running multivariate experiments is part of the solution, but not the entire solution. While an experiment can isolate the impact of a product change on an input metric, the cumulative impact on lagging business KPIs remains unclear. Without a clear way to isolate the impact of bets, teams often give up on being metrics-driven and instead rely solely on intuition to prioritize bets or choose which experiments to run. However, there is a structured approach that radically improves how teams use data to make decisions: deterministic KPI trees, a tool for bringing clarity and rigor to bet evaluation. These trees define relationships between metrics mathematically, providing a framework to estimate the contribution of individual bets to overarching business goals. The attached screenshot showcases DoubleLoop's new Bet Simulator prototype (development just kicked off today!), which uses a deterministic KPI tree to estimate the impact of bets on a business's KPIs. Through the deterministic KPI tree, each factor can be analyzed in isolation. The relationships between metrics—such as Sales = Revenue per Visitor × Total Visitors—provide a structured way to attribute changes in the root metric (sales) to specific bets and external factors. While deterministic KPI trees offer clarity, there are limitations. Many metric relationships are probabilistic, not purely mathematical. That said, even an approximate sizing of bet opportunities with deterministic KPI trees is far superior to not estimating bet impact at all. Using deterministic models enables teams to: - Use data to discuss the relative importance of different bets - Avoid overestimating or underestimating impacts - Ensure assumptions align with realistic expectations and the nuances of the business

For years, I thought most companies made decisions based on data and careful analysis. Then I got closer to the inside of those decisions. I saw supply chain executives fighting over spreadsheets with 20 tabs, each one producing a slightly different answer. I saw managers defaulting to “the way we’ve always done it,” even when the stakes were in the millions. I saw incredibly smart teams chasing gut instincts because the data wasn’t trusted, the process wasn’t clear, or the models weren’t explainable. That changed the way I thought about my own work. It wasn’t enough to just build a solver model, or an elegant piece of code. The real question was: 👉 Does this decision process give leaders confidence that they’re not leaving money on the table? I’ve come to believe three things: 1️⃣ Most organizations don’t measure the cost of being wrong. They underestimate how expensive “good enough” really is. 2️⃣ Consistency is underrated. A process that gives a repeatable, explainable answer beats a one-off “heroic” decision every time. 3️⃣ Bias creeps in quietly. Without structured frameworks, politics and personalities decide more than we admit. Looking back, some of the most impactful projects I’ve been part of weren’t the flashiest. They were the ones where we gave decision-makers clarity: Here is why this is the best choice. Here is what it costs if you do otherwise. Here’s the confidence level behind it. That’s why I work in optimization today. Not because I love algorithms (though I do), but because I’ve seen what happens when organizations fly blind. So here’s my challenge to you: When your team makes its next critical decision… pause and ask yourself: ✅ Could I defend this choice if a board member or regulator asked me “why this?” ✅ Do I know the cost of being wrong? ✅ Am I confident this is the best decision, or just a reasonable one? Because if you don’t know the answers, you’re not really making decisions. You’re just hoping.

Your data strategy has a blind spot. It's missing critical connections. Most organizations store data in ways that hide their most valuable insights. This leads to missed opportunities and undetected risks. Graph modeling changes this by representing data the way your business actually works. Instead of forcing relationships into rigid tables, graphs capture the natural connections between customers, products, suppliers, and transactions. Your data finally reflects reality. Graphs have four key elements: 🎯 Nodes represent your business entities like customers, products, and locations. 🔗 Relationships show how they connect through purchases, partnerships, and dependencies. 📋 Properties store relevant details such as dates, amounts, and contact information. 🏷️ Labels organize similar entities for easy identification. When fraud investigators model accounts and addresses as connected nodes, criminal rings become visible instantly. When supply chains are mapped as graphs, single points of failure emerge before they cause disruptions. The connections reveal what spreadsheets hide. The real power emerges when you ask business questions that span multiple relationships.  ↳ Who are your most influential customers?  ↳ Which suppliers create the biggest risk?  ↳ What paths do successful transactions take? Organizations using graph modeling detect fraud faster, optimize supply chains more effectively, and identify opportunities competitors miss. That's why at data² we built the reView platform on a foundation of graphs. 💬 What hidden relationships could transform your business decisions? Share your thoughts below. ♻️ Know someone wrestling with complex data relationships? Share this post to help them out. 🔔 Follow me Daniel Bukowski for daily insights about analyzing data using graphs + AI.

Start with the End in Mind: A Key to Better Data Models When it comes to real-world data modeling, always start with the end goal. Ask yourself What questions does this model need to answer? Here’s why this approach works: * It keeps your model focused on business needs. * You avoid unnecessary complexity by only including relevant tables and fields. * Stakeholders get the insights they need faster. At a previous company, we needed to analyze customer churn. By focusing on this specific outcome, we identified key tables: Customer, Sales Transactions, and Product Returns. This targeted approach allowed us to create a concise model that directly addressed our churn analysis needs Tips for new users: * Define key metrics/KPIs first like total sales or monthly trends. * Build a functional model quickly and refine it as needed. * Create measures early to guide relationships and validate results. Remember: A model that serves its purpose is better than a perfect one that doesn’t. #DataModeling #PowerBI #BusinessIntelligence #DataAnalytics #BIBestPractices #DataVisualization #CustomerChurn #KPIDriven #DataDrivenDecisions #EfficientModeling #BIInsights #DataOptimization #PowerBIModels #ChurnAnalysis #AnalyticsTips

One of the biggest mistakes data teams make is underestimating the importance of data modeling. If you take a closer look within a cloud data platform - data modeling is an 𝒊𝒎𝒑𝒆𝒓𝒂𝒕𝒊𝒗𝒆 piece in each data layer. 𝐇𝐞𝐫𝐞'𝐬 𝐡𝐨𝐰: 𝐒𝐭𝐚𝐠𝐢𝐧𝐠 𝐋𝐚𝐲𝐞𝐫: This initial phase involves collecting raw data from various sources. Proper data modeling at this stage ensures that the data is accurately represented and organized for subsequent processes. 𝐂𝐨𝐫𝐞 𝐋𝐚𝐲𝐞𝐫: Integration of data from multiple sources happens here. Data modeling helps in creating consistent and unified entities, facilitating smoother data transformations and ensuring data integrity. 𝐂𝐨𝐧𝐟𝐨𝐫𝐦𝐞𝐝 𝐋𝐚𝐲𝐞𝐫: Applying complex transformations and business rules, this layer benefits immensely from robust data models, which provide a clear structure and standardize data across the organization. 𝐏𝐫𝐞𝐬𝐞𝐧𝐭𝐚𝐭𝐢𝐨𝐧 𝐋𝐚𝐲𝐞𝐫: In the final stage, data is made consumption-ready for analytics and reporting. A well-designed data model ensures that the data is easily accessible and comprehensible for business users, enabling quick and accurate decision-making. Effective data modeling ensures data integrity, quality, and enables quick decision-making. #datamodeling #dataarchitecture #cloud

“Why does the Mode report say one thing and the exec dashboard another?” “Spend matches, but revenue’s off — which one’s right?” “If I can’t trust these numbers, how can I make a decision?” We’ve all heard this. The problem isn’t the culture or the tooling. It’s the modeling. Too often, a dashboard - or so-called “insight product” - is a thin wrapper over the warehouse layer. Transformed tables with charts on top. 📊 DATA PRODUCT LAYERS • Warehouses & Marts → structure the data • Insight Products → curate decision-ready, trusted data products Many teams stop at the warehouse layer, but think they’re shipping insights. 🔄 THE BUSINESS MODELING LOOP Delivering Insight Products requires a different discipline: ➡️ Interview stakeholders — understand decisions & metrics ➡️ Identify definition differences — surface conflicting terms/metrics ➡️ Illustrate differences — show contradictions side by side ➡️ Conform metrics & dimensions — facilitate agreement, document definitions ➡️ Design top-down — start from decision-making products & metric trees, then build the conformed model to support them This is business modeling, not data modeling. 🚦 WHY IT MATTERS • Warehouse-first → clean tables ❗ mistrusted dashboards • Insight-first → alignment ✅ trust ✅ faster decisions ✅ If they can’t decide with it, it’s not an Insight Product. 💡 Being a Data Elbow gets you into the room. Building Insight Products earns you the right to stay.

Sanjeev Mohan dives into why the success of AI in enterprise applications hinges on the quality of data and the robustness of data modeling. Accuracy Matters: Accurate, clean data ensures AI algorithms make correct predictions and decisions. Consistency is Key: Consistent data formats allow for smoother integration and processing, enhancing AI efficiency. Timeliness: Current, up-to-date data keeps AI-driven insights relevant, supporting timely business decisions. Just as a building needs a blueprint, AI systems require robust data models to guide their learning and output. Data modeling is crucial because: Structures Data for Understanding: It organizes data in a way that machines can interpret and learn from efficiently. Tailors AI to Business Needs: Customized data models align AI outputs with specific enterprise objectives. Enables Scalability: Well-designed models adapt to increasing data volumes and evolving business requirements. As businesses continue to invest in AI, integrating high standards for data quality and strategic data modeling is non-negotiable.

Most organisations don’t struggle to build predictive models. They struggle to turn predictions into better decisions. I’ve seen this pattern repeatedly across industries. Teams invest heavily in improving model accuracy, yet the business outcomes barely move. Not because the models are wrong, but because the decision process around them hasn’t changed. Prediction alone rarely creates value. Decisions do. In my latest article, “Decision Intelligence: Moving from Prediction to Action,” I explore why many AI initiatives stall after the modelling phase and what it takes to close the gap between analytical insight and operational impact. The shift is subtle but important. Rather than treating models as the final output, organisations need to think in terms of decision systems, where predictions, policies, workflows, and feedback loops operate together. This is where disciplines like decision orchestration, real-time analytics, and continuous learning become critical. In practice, the most effective AI systems are not defined by how accurately they predict the future. They are defined by how effectively they shape actions in the present. This article is the latest edition of The Data Science Decoder, where I unpack ideas shaping how AI is actually deployed inside organisations. If you’re working on operationalising AI, or trying to move beyond isolated models toward real business impact, you may find the perspective useful. Read the full article here:

🚀 𝐃𝐚𝐭𝐚 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 𝐚𝐧𝐝 𝐌𝐚𝐜𝐡𝐢𝐧𝐞 𝐋𝐞𝐚��𝐧𝐢𝐧𝐠 𝐀𝐫𝐞 𝐎𝐟𝐭𝐞𝐧 𝐓𝐚𝐮𝐠𝐡𝐭 𝐚𝐬 𝐓𝐨𝐨𝐥𝐬 𝐁𝐮𝐭 𝐓𝐡𝐞𝐲’𝐫𝐞 𝐑𝐞𝐚𝐥𝐥𝐲 𝐚 𝐖𝐚𝐲 𝐨𝐟 𝐓𝐡𝐢𝐧𝐤𝐢𝐧𝐠 Many people learn data science as a collection of disconnected topics. Statistics here. ML algorithms there. Visualization at the end. This document reinforces a more important idea: data science is an end-to-end reasoning process, not a toolbox. Everything starts with understanding the problem. Before models, before features, before metrics, there’s a question to be answered and a decision to be improved. Without that clarity, even technically perfect models fail to create value. Statistics isn’t just theory in this process. Descriptive statistics help you understand what the data is saying. Probability and distributions help you understand uncertainty. Inferential statistics help you decide whether patterns are real or accidental. These concepts shape better modeling decisions long before algorithms enter the picture. Machine learning then becomes a natural extension, not a replacement. Regression, classification, clustering, dimensionality reduction, and ensembles are simply different ways to formalize patterns already observed in the data. When fundamentals are strong, model choice becomes logical instead of experimental. Another important takeaway is that data preparation and feature engineering are not “pre-work.” They are core modeling steps. Scaling, binning, handling missing values, and managing imbalance often matter more than switching between algorithms. Validation is where maturity shows. Understanding bias–variance tradeoff, cross-validation strategies, and appropriate evaluation metrics is what separates models that look good in notebooks from models that hold up in production. What I appreciate most about this material is that it treats data science as a lifecycle from data understanding to modeling to communication. Visualization and storytelling aren’t optional add-ons; they’re how insights actually influence decisions. I’m uploading this document because it captures the full picture clearly. If you’re learning data science, preparing for interviews, or revisiting fundamentals after working in the field, this kind of structured perspective compounds over time. Good data science isn’t about knowing more algorithms. It’s about making better decisions with data. #DataScience #MachineLearning #AI #ArtificialIntelligence #Statistics #Analytics #MLFundamentals #DataVisualization #TechCareers #LearningInPublic #BuildInPublic #FutureOfWork

In data science we build AI/ML models that "make decisions", for example you build a model that determines the likelihood of a customer churning. Ok so what? Now that you know, what actions should you take? Now that you know the customer is likely to leave, or their estimated LTV is below some threshold you need to take an action that leads to a desired outcome. This is where things get murky for data teams, things are often done haphazardly and businesses wonder what's the point of data. So what do you do? What's the desired outcome? Should you call them? What if they use the call as an opportunity to cancel? Should you send them a promotion? Do nothing? This "action that leads to a desired outcome" is at the heart of Decision Intelligence (DI). DI concerns itself with clarifying the action => outcome causal chain. Here's where human intuition and artificial intelligence get integrated into a coherent whole. It also happens to be the key element businesses care about when it comes to the value of data, because if data isn't helping you take the best possible actions to achieve your desired outcomes, then what's the point? DI tries to answer these types of questions: - If I take this action, what outcome will it lead to? - What actions should I take to get the outcome I want? - Given these actions that are available to me, what outcomes can I expect? DI does this through a simple diagram, called a Causal Decision Diagram (or CDD) where the actions (or levers), intermediate effects, external factors, outcomes and their cause-effect relationships are explicitly laid out for everyone to see. The ML/AI model is only one of a multitude of inputs to this diagram. My point is that this should be the key deliverable for a data team. An agreed upon CDD model is often more valuable than the ML/AI model you were planning to build. In fact, you might even discover that the ML/AI model doesn't even need to be that complicated. A few simple rules is all it takes.