Technical
Exploring specific data science applications and methods.
-
How do you test an enterprise AI system when performance is distributed across a chain of interdependent components? In this post, I outline a layered evaluation framework for RAG, prompts, GenAI models, deterministic rules, and end-to-end workflows—using a Decision Contract to define what “good” looks like. The goal is not a single AI accuracy score,…
-
Organizations investing in AI often prioritize infrastructure but overlook a critical question: what decision is being improved? This leads to a weak approach that operationalizes conversation rather than intelligence. The concept of a Decision Contract can transform AI from a vague assistant to a governed, decision-oriented system, ensuring effective judgment and measurable economic actions.
-
This post explores how Expected Pipeline Value—a probability-driven, risk-adjusted view of the pipeline—transforms B2B sales pipeline management by applying Decision Science principles to evaluate both deal probability and deal value. By moving beyond simple deal counts and nominal (face) value, this approach delivers a clearer view of pipeline quality, improves resource allocation, strengthens forecast credibility,…
-
The way we do data science has changed forever. GenAI makes coding frictionless—but without structure it gets messy and invites slippage down rabbit holes. I’ve been refining CRISP-AI, a lightweight process (inspired by CRISP-DM) to work smarter with AI.
-
My latest post discusses optimizing Marketing spend using reinforcement learning (RL) and an LLM as a meta-controller—an approach we term Agentic-AI. It details creating a custom RL environment for managing multiple Marketing channels and describes how the LLM meta-controller orchestrates the overall learning process. Rather than making spend decisions directly, the LLM guides iterative training,…
-
Check out my latest blog post, where I set up a simple scenario to show how a reinforcement learning (RL) agent finds the sweet spot in marketing spend—laying the groundwork for Agentic AI. Whether you’re curious about RL fundamentals or integrating LLMs, this mini use case offers valuable insight into the inner workings of RL…
-
Large Language Models (LLMs) are powerful, but they don’t always surface the most relevant or up-to-date information for your needs. Retrieval-Augmented Generation (RAG) changes that by integrating real-time, domain-specific insights, ensuring AI-generated responses are always up-to-date and relevant. Instead of relying on generic training data, RAG personalizes AI with custom knowledge sources, from reports to…
-
This blog explores how saturation curves and incremental returns can guide optimal marketing spend allocation. Using scenario analysis, I show why relying on averages like ROAS can mislead decision-makers. Learn how to balance spend across channels, avoid saturation, and unlock growth by reallocating budgets to underutilized opportunities for higher returns.
-
Let’s sketch some broad strokes to delineate these roles. Though exceptions abound – and every organization is different – this framework might help clarify some important overall distinctions between these roles. Analyst: the traditional Analyst role primarily engages in standard reporting and ad-hoc analysis to probe deeper into business performance. They certainly support decision-making processes…