At a time when data flows freely and technology has made collecting and processing information almost second nature, publishing that data in a meaningful way remains a challenge. It is, in many ways, an art. All too often, data is simply offloaded to clients, but not truly delivered. Effective data publishing is a form of storytelling, one that demands context, consistency, and clarity. It is what transforms raw numbers into insights that allow people to act on with confidence in business contexts.
Ravish Mishra, VP of engineering (data) at Angel One, addressed this critical aspect of data analytics during a session titled ‘The Art of Data Publishing: From Noise to Narrative’ at the DES 2025 summit. Speaking to a room full of data engineers and tech enthusiasts, Mishra navigated the history, challenges, and future of data architecture, encouraging companies to move beyond storage towards meaningful storytelling.
From Warehouses to Narratives
Mishra opened his talk by tracing the evolution of data management—starting with data warehouses, progressing through lakes and lake houses, and now advancing towards more AI-integrated architectures, including generative AI stacks.
“At Angel One and across India, we are actively building processes on top of GenAI and related AI technologies,” he said. He emphasised the classical ‘five Vs’ of big data: volume, variety, velocity, veracity, and value. While organisations have tackled the first three via modern architectures like data lakehouses, he pointed out that value and veracity continue to be challenging for many organisations.
“How many of you think your organisation is using data to its full potential?” he asked the audience. “If your platform enables storytelling that drives business decisions and enhances customer experience, then you’re on the right track.”
He raised concerns over the increasing volume of data, noting, “Over the past five years, global data volume has grown by 5%. As [the amount of] available information rises, organising it becomes critically important. Without effective management, we risk creating systems filled with hard-to-interpret, low-quality data [that is] unreliable for business decisions.”
The Noise in Data
A recurring theme in Mishra’s session was the concept of “noise” in data—inconsistencies and errors that dilute its usefulness. He highlighted several key types: quality noise, semantic noise, temporal noise, and structural noise.
“The real value of data is unlocked when it enables informed decision-making. Ideally, those decisions should be encoded within the data model itself. It’s not just about making decisions, but understanding why they are made,” he said.
According to Mishra, the true value of data doesn’t lie in the warehouse or lake; it’s in the narrative layer. He advocated for building dimensional models within the “gold layer” of data architecture, which supports easier querying and faster interpretation. However, even these models often rely on analysts to derive meaning manually.
“We should aim to encode decisions within data models,” he said, criticising the common anti-pattern of treating data as an afterthought.
“Businesses often fail to plan how they’ll measure a new feature’s success during its design. Instrumentation is added post-release, leading to fragmented data and missed insights.”
He also highlighted the inefficiency of duplicating business logic across both the source and analytics layers—a practice he strongly advised against.
Transform Data from Noise to Narrative
Mishra outlined a framework for building decision-driven data systems rooted in key principles: stability and immutability, fixed contracts, service-level agreements (SLAs), and semantic clarity. Drawing from his experience, he stressed the importance of immutability once data is published, ensuring that downstream systems don’t alter the integrity of events.
“A stable event is one that means the same thing now and five years later, i.e., the event is unambiguous and preserves its uniqueness through space and time,” he explained.
To address semantic noise, he called for a robust semantic layer that defines data meaning clearly and consistently. “Good data products must not only encode facts but also support narratives and explainable stories that reveal why decisions were made,” he said.
This includes enriching data models with diagnostic and prescriptive intelligence, building actionable interfaces, and keeping narratives updated as data sources change. “Fixed contracts and SLAs are crucial to ensure consistent analytics and reliable insights,” he added.
As the session concluded, Mishra reminded attendees that the ultimate goal of a data system isn’t just analytics; it’s narrative. “Decisions are encoded in data. Your models should not only answer questions but also tell stories,” he said. “That’s how we transform data from noise to narrative.”
