Mr. Satyanarayana delivering a lecture at Duke University
As enterprise artificial intelligence (AI) evolves rapidly, the challenge of establishing trust in these systems is becoming increasingly critical. While many organizations can implement AI assistants for tasks such as document summarization and inquiry response, far fewer can leverage AI-driven insights for decision-making, particularly in high-stakes environments where the cost of errors is significant. This disparity between AI’s capabilities and enterprises’ willingness to fully utilize them highlights a pressing concern in business technology today.
“Enterprises do not scale plausibility. They scale evidence,” said Praveen Koushik Satyanarayana, Senior Director of Customer and Data Strategy at Tredence. He emphasized that while AI models can generate seemingly correct answers, effective production systems require adherence to principles of repeatability and defensibility. Questions such as the appropriateness of definitions, scope precision regarding time, geography, product, and cohort, and the ability to reproduce results consistently are paramount for enterprises.
With over a decade of experience building analytics systems across sectors including retail, consumer packaged goods, travel, banking, and healthcare, Mr. Satyanarayana leads a team of more than 100 at Tredence. His previous initiative, Customer Cosmos, an AI-powered platform for customer analytics, reportedly generated over $1 billion in client value. This experience has informed his understanding of the operational realities that emerge when AI is integrated into business workflows.
Mr. Satyanarayana’s current project, Milky Way, aims to redefine agentic analytics by embedding reliability into its design rather than leaving it to chance. This is achieved through three foundational elements: rubrics that explicitly define quality criteria for various problem types, evaluation sets that serve as established ground truth, and decision traces that verify the process and outcomes, thereby fostering trust.
One common misstep, according to Mr. Satyanarayana, is viewing an agent as merely a conversational interface rather than as a critical decision-making workflow. In the context of enterprise operations, such workflows necessitate controls, including explicit evaluation gates and traceability mechanisms. “’Helpful’ is not a specification,” he stated. “Enterprises need criteria that can fail loudly. A rubric turns quality into measurable checks. If you cannot score behavior, you cannot govern it, you cannot improve it reliably, and you cannot earn permission for autonomy.”
Milky Way categorizes incoming requests based on task type—ranging from descriptive to prescriptive and governance tasks. It selects the appropriate methodologies and evaluates its outputs against the established rubrics prior to releasing any results. These rubrics act as quality gates, assessing whether the system adhered to the correct scope, applied identified business definitions, followed permissible query patterns, and validated its calculations. In cases where a rubric fails, the system either seeks clarification or flags the output for human review.
“Ground truth is not one table of truth,” Mr. Satyanarayana explained, underscoring the necessity for a diverse reference library that aligns with various question types and definitions within an enterprise. Milky Way maintains three tiers of reference artifacts: validated definitions and calculation rules, approved query patterns, and curated test scenarios akin to regression tests. This library expands as teams encounter new question types and identify failure points within the system.
Every request processed also generates a structured trace detailing the data accessed, logic applied, and validations conducted. Business stakeholders receive clear narratives, analysts access detailed scope and logic paths, and compliance teams can review immutable records of data access and intent. “Speed without proof creates distrust,” Mr. Satyanarayana remarked, noting that these traces facilitate debugging, validation, and auditing, ultimately reducing disputes as stakeholders can examine the methods used instead of relying on differing assumptions.
In Milky Way, autonomy is not an all-or-nothing proposition; rather, it is stage-gated based on measured reliability. Workflows can operate in three stages: human in the loop (system proposes, human approves), human on the loop (low-risk workflows execute automatically, with exceptions escalated), and selective autonomy (narrow workflows execute when performance metrics meet stability). The extent of autonomy is directly correlated with the evidence that supports it.
Tredence has garnered recognition for its contributions to generative AI, being named an Emerging Visionary in Generative AI Consulting and Implementation Services by Gartner and a Leader in the ISG Provider Lens for Generative AI Services 2024. Mr. Satyanarayana has also contributed thought leadership to publications such as The Fast Mode and CMS Wire.
While the methodology employed by Milky Way may not be suitable for every enterprise due to the substantial overhead associated with maintaining rubrics, reference libraries, and trace infrastructures, organizations facing significant consequences from errors may find the investment worthwhile. “The path to autonomy runs through proof,” Mr. Satyanarayana concluded, leaving open the question of whether organizations will adopt this mindset.
See also
IISc Research Reveals Shape-Shifting Molecules for Future AI Hardware Integration
HP Reveals 2026 CES Lineup: New EliteBooks, OmniBooks, and OMEN AI PCs Unveiled
Chinese LightGen AI Chip Surpasses Nvidia A100 with 100x Speed and Efficiency
Software Engineering Jobs Surge to 105K in 2026, Driven by AI and Cloud Demand
SoftBank Acquires DigitalBridge for $4 Billion to Enhance Global AI Infrastructure
