Experience Sharing by Global Outstanding Business Leaders on Data and AI Adoption

Start with the smallest unit of “data value” you can imagine. That's a decision. Someone is about to do something, and a little better information would make that choice a little better. That's the entire atomic value.

Every data platform, governance plan, and AI initiative is just a mechanism built to make this moment happen more frequently, more cheaply, and more reliably.

Directly listen to the experience sharing of leaders who build and scale data and AI capabilities in some of the world's most outstanding organizations. These organizations include Dasa, one of the largest diagnostic medicine and clinical analysis companies in Latin America; United Talent Agency, one of the most influential companies in the global entertainment industry; Kotak, one of the largest mutual fund companies in India; and even The White House.

Point 1: The quality of a decision depends on its context.

Before considering platforms or governance, ask yourself a question: What makes one decision better than another?

Nehhaa Purohit, the Chief Digital Officer of United Talent Agency, shared some interesting experiences because she paid a high price for it.

Although her dashboard showed that everything was normal, with an uptime of 99.99% and a latency of less than 50 milliseconds, in fact, the revenue per inference decreased by 2% every month for six consecutive months, and a loss of $14 million went unnoticed because the intelligent system was running normally, but the judgment was wrong.

Uptime measures whether the pipeline is clear, not whether the data transmitted in the pipeline is still valid. Purohit calls this gap “context debt”: the data is still technically valid, but semantically obsolete and gradually losing effectiveness, while all monitors still show normal.

A decision is “good” because at the moment someone takes action, the data still matches the reality: this is more difficult to guarantee than uptime, and almost no one measures it.

Point 2: A good decision does not have a cumulative effect, but repeatable decisions do.

A well - thought - out decision is great. But if you want to replicate this decision across the team or over a quarter, you'll encounter the real challenge that every leader is struggling with: the trade - off between speed and trust.

Animesh Kumar, the Chief Technology Officer of The Modern Data Company, said that the usual interpretation of this phenomenon is a false tug - of - war. Corporate governance is classified as “defense,” which makes it the first to be cut when the company needs funds for development.

This is completely upside down because unregulated data will actually slow down growth. No one will trust data that cannot be traced, so they will manually re - verify it, which adds an implicit cost to every subsequent decision. The solution is to make data governance a default attribute of the platform, rather than an additional paid project.

Mohamed Amin, the Vice President of Digital Transformation at Hosta, also faced the same problem, namely decision delay: the lag between insight and action. His solution is clarification, prioritization, and responsibility assignment: replace raw metrics with signals that can be directly applied to the business, replace detailed reports with a small number of key performance indicators (KPIs) that can drive decisions, and assign a responsible person to each indicator.

Different words, the same equation: repeatability means eliminating the resistance for the next decision, not just providing information for the current one.

Someone needed an answer faster than the official channel, so they developed their own solution. Kumar's approach is to break the problem down into structure, behavior, and incentive mechanisms, because if the official channel is slower than the workaround, the workaround will always win.

Point 3: A repeatable decision only makes sense when it affects the key metrics that the Chief Financial Officer cares about.

Even a well - thought - out and repeatable decision is not a strategy because senior management doesn't care about repeatability itself; they only care about profit and loss.

Dia Adams, the Chief Data and AI Officer of Datafolx AI and a former White House corporate data strategist, came to a clear conclusion:

Stop talking about technical metrics and start talking about business outcomes.

Instead of saying “We built a data lake,” say “This achieved $X in revenue through high - level personalization and reduced the customer churn rate by Y%.” She believes that operating income should be adjusted because earnings before interest and taxes (EBIT) includes both revenue growth and cost reduction, and the board already trusts this figure.

Jon Cooke, the CEO of Nebulyx AI, used the analogy of defense and offense to answer this question . Defense refers to the indispensable governance work in regulated industries: a versioned, reference - compliant knowledge base means that auditors only need to ask one question to know “Why did you make that decision in March?”

The offensive strategy is a combination of use cases, each associated with an intuitively changeable metric. These use cases will stack up because clean data can eliminate quality issues, thus accelerating the offense and preventing the progress of the next use case from slowing down.

Justin York, the head of Rubicon, questioned the entire premise. When asked which item on the income statement should be adjusted, his answer was: Take credit carefully, even if you can't trace it to a specific reason. If sales increase, is it the credit of the data strategy, or does someone still need to “make the sales”?

Usually, it's difficult to truly evaluate the return on investment in isolation because data is involved in all aspects, and the greater credibility risk lies in leaders over - hyping “solutions” that are vulnerable to actual workload and friction. Adams and Cooke built a confident financial story; York refused to exaggerate. Both approaches are correct, and the key lies in whether these numbers are attributable.

Point 4: Nothing can be scaled if no one uses it.

Even a well - argued, repeatable decision linked to profit and loss may fail in the last mile: when non - technical people have to take action based on the insights in front of them, the decision fails.

Gabriel Vernalha Ribeiro, who is in charge of data governance and AI at Dasa, believes that the problem should start from the data generation stage. His solution is process improvement: all projects must have a data collection plan and success metrics from the beginning, otherwise they cannot be approved.

Structured data from the start “is no longer a maintenance cost but an available asset at any time.” His last - mile principle is: Provide insights in the tools that people are already using, and replace charts that need interpretation with concise “what to do next” suggestions.

Cooke's “decision capsule,” independently developed by Purohit of UTA, goes a step further: display a single interface in the workflow, prompting the recommended action, confidence interval, and main drivers.

Purohit's version gracefully degrades when the context becomes obsolete. It hands the decision - making power back to humans instead of confidently giving the wrong answer, and this design choice avoids an estimated $8 million in wasted funds.

Ribeiro's concluding view ties everything together: Culture cannot be imposed; it must be encouraged and cultivated. Only when bad data causes obvious losses to the team and is linked to the team's own key performance indicators (KPIs) will the team truly take responsibility.

Point 5: The system can only be maintained when managers also think so.

The most surprising part of this series is not about data at all, but about who you hire to run the system.

Purohit's recruitment screening method is the clearest manifestation of first - principles thinking. She divides candidates into pattern matchers (good at stable states) and first - principles thinkers (good at decomposing highly ambiguous problems from scratch).

Her test: diagnose a system where the dashboard shows everything is normal, but the revenue per inference is quietly declining. Candidates who only focus on latency and error rates will be eliminated. They are looking for certainty in probability problems. Those who try to solve semantic biases will pass. One leader manages the system, and the other creates value.

Cooke described the same transformation through AI agents: the Chief Data Officer is no longer a librarian organizing known knowledge but an architect of bounded decision - making spaces. In these areas, agents act under intelligent constraints.

Adams calls it a “business architect,” replacing the “data plumber.” Different words, but the same intention: the job is no longer to keep the pipeline clear, but to maintain the accuracy of judgment at the moment of making a decision (whether it's a human or a system decision) based on existing data.

This article is from the WeChat official account “Data - Driven Intelligence” (ID: Data_0101), author: Xiaoxiao. It is published by 36Kr with authorization.