# Translating Business Goals into Product Decisions Richa Sachdev, JP Morgan Chase | MLOps Podcast | Episode 184 | 42:49 Hosted by Demetrios Brinkmann Source: https://www.youtube.com/watch?v=kZ2si-e0dPk Channel: MLOps Community, now AAIF Live (https://www.youtube.com/@AAIFLive-x1r). Summarised by MLOps Talks. Page: https://mlopstalks.com/talks/translating-business-goals-into-product-decisions Published: 2023-10-24 Tags: build-vs-buy, data-engineering, data-quality, product-strategy ## TL;DR - Richa Sachdev says product and technology decisions should start with business goals, user journeys, and measurable outcomes. - She separates operational, analytical, and machine learning data because each use case needs different storage, transformations, and access patterns. - Her teams balance legacy systems with modernization by delivering useful business results while building lighter replacement systems in parallel when a full migration is too risky. ## Summary Richa Sachdev describes how her career moved from software engineering into data engineering, machine learning, reporting, and process automation. At Vanguard, she helped establish data foundations before the organization expanded into descriptive, prescriptive, and predictive work. She explains how operational data supports immediate decisions, analytical data reveals trends, and machine learning data requires historical records, suitable storage, and a semantic layer. She is direct about the difficulty of joining structured and unstructured data across systems, especially when privacy, identifiers, security, and refresh schedules differ. In her current work at JP Morgan Chase, she starts with the card and rewards business, customer journeys, process controls, and measurable outcomes. Automation is assessed through reduced errors, lower latency, fewer calls, better quality checks, and closed audit findings. Her hardest decisions involve limited time, budget, legacy technology, and many stakeholders. She sometimes recommends running a lightweight replacement beside a legacy system before switching over. ## Key ideas ### Richa Sachdev built her data career on software engineering foundations [04:57](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=297s) Richa Sachdev began with computer science degrees and software development work that included production support, change management, and building applications from the ground up. At a healthcare company, she moved a legacy application to Java and Spring Boot, then sought out analytics teams working with electronic medical records, Hadoop, Hive, and Impala. She later joined Vanguard as its department's only data engineer, with a mandate to establish core data competency before adding data scientists and machine learning work. She learned machine learning on the job, while her engineering background gave her experience with code quality, version control, and change management. ### Prescriptive, descriptive, and predictive work answer different business questions [09:17](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=557s) Richa Sachdev defines prescriptive work as business-led analysis. The business has usable data and clear KPIs, then asks a reporting or operations team to cut the data into a decision-making format. Descriptive work begins with analysts and engineers examining distributions, graphs, and other patterns, then discussing with the business which findings deserve dashboards. Predictive work uses historical data, trends, and algorithms to estimate future outcomes. Her examples include predicting the next topic a customer may want to hear, the channel they may respond to, or the next advertisement they may see. ### Operational, analytical, and machine learning data need separate designs [11:59](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=719s) Richa Sachdev groups data into operational, analytical, and machine learning uses. Operational data covers immediate activity such as clicks, email responses, and offer sign-ups, often through reports and dashboards. Analytical data aggregates those events over monthly, quarterly, or yearly periods so teams can compare trends and KPIs. Machine learning uses larger historical collections for training, with attention to warehousing, low-latency access, and data availability. She says teams must decide where raw data, semantic layers, and analytical layers belong, while working with producers on formats, catalogs, schemas, and retention. ### Raw data and shared meaning matter when data feeds models [14:30](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=870s) For machine learning, Richa Sachdev prefers keeping as much raw data as practical in one place, then allowing individual teams to create transformations through a semantic layer. This helps preserve meaning that could be lost through an earlier transformation. She also describes the shift from centralized warehouses toward data mesh patterns, where data remains with departments and consumers handle access and joins. Unstructured sources such as calls, email, complaints, and social media make the architecture harder because teams must connect relational and non-relational data while handling privacy and access rules. ### Joining databases has no universal solution [17:52](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=1072s) Richa Sachdev says integrating databases involves more than choosing a storage product. Teams must consider architecture, security, refresh timing, consistent identifiers, and personally identifiable information. At one point, her team stored files in an S3 bucket and used programs to extract, parse, and condense the contents before algorithms consumed them. The derived information was stored so a customer could be viewed through a 360-degree frame across transactions, visited pages, calls, and email. Since the models did not refresh in real time, the team had time to improve the upstream data and optimize how insights were found downstream. ### Business context now shapes Richa's reporting and automation work [21:17](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=1277s) At JP Morgan Chase, Richa Sachdev works on data reporting and process automation in the card rewards space. She provides fast operational and analytical insights while learning the business processes behind them. For automation, she examines touchpoints, manual steps, errors, and controls. Possible approaches include workflow automation, robotic process automation, and machine learning that analyzes calls, customer journeys, and web or mobile dropouts. She is still a technical leader inside a business organization, so she has to keep business needs in view while deciding where technology can help. ### Business impact is measured through customer and process outcomes [24:24](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=1464s) Richa Sachdev says teams should connect their work to the rewards business and its customer journeys. Measures include earning and redeeming points, combining points, reducing fraud, making interactions easier, lowering calls and complaints, and improving the timing of awards or clawbacks. Process automation can reduce errors and latency, while metrics can track task duration, quality checks, and audit or control findings. She says the team is currently using simple heuristics, basic analysis, and human review while building the business case for more advanced machine learning. ### Modernization decisions balance immediate delivery against future architecture [32:27](https://www.youtube.com/watch?v=kZ2si-e0dPk&t=1947s) Richa Sachdev faces a recurring choice between pushing systems into the cloud and delivering reports or decisions to the business quickly. Time, staff, budget, ownership, and stakeholder coordination all constrain the choice. If a legacy system is too difficult to split apart, she may recommend building a lightweight system beside it and running both in parallel. The teams can then switch over after the replacement works. She also considers domain-driven architecture, microservices, data mesh, data-center migration, and technical debt, while keeping the direct effect on customers in mind. ## Notable quotes - Richa Sachdev: "The message is at the scene, the more senior you are, the more interplay you have to understand." (35:59) - Richa Sachdev: "If your foundational data is not correct, if you have a thought about the business needs and how you're going to be using the data, then all is lost." (25:25) - Richa Sachdev: "There is no silver bullet to it, and every company does it in a little bit customized way because the rules change, the access controls change." (19:06) - Richa Sachdev: "The tradeoff always is, should we rush to push everything into cloud and modernize it, or should we figure out first how do we get the reports and the important decisions out to the business as soon as possible?" (32:27) ## Tools & references mentioned - JP Morgan Chase - Vanguard - Java - Spring Boot - Hadoop - Hive - Impala - S3 - Snowflake - dbt - data mesh - machine learning ## Who should watch - You are building a data or ML platform and need to decide what belongs in operational, analytical, and training data systems. - Your team is proposing modernization while business users still need reports and decisions from legacy systems. - You need a practical way to connect automation work to customer outcomes, process timing, quality checks, and audit measures. ## Related talks - [How To Move From Barely Doing BI to Doing AI](https://mlopstalks.com/talks/how-to-move-from-barely-doing-bi-to-doing-ai) (Joe Reis, Ternary Data, 53:47) - [What Business Stakeholders Want to See from the ML Teams](https://mlopstalks.com/talks/what-business-stakeholders-want-to-see-from-the-ml-teams) (Peter Guagenti, Tabnine, 1:21:28) - [Data Engineering + ML + Software Engineering](https://mlopstalks.com/talks/data-engineering-ml-software-engineering) (Satish Chandra Gupta, Slang Labs, 57:05) - [A Journey in Scaling ML](https://mlopstalks.com/talks/a-journey-in-scaling-ml) (Gabriel Straub, Ocado Technology, 52:41) - [Path to Production and Monetizing Machine Learning](https://mlopstalks.com/talks/path-to-production-and-monetizing-machine-learning) (Vin Vashishta, 56:35)