Predict the Future: How Top CEOs Use Data to Stay Ahead

Bill Schmarzo, Dell Technologies' data strategy leader and former Yahoo vice president of advertiser analytics, joined Colin Iles for the Standard Bank OneHub Series to argue that most companies still get data wrong at the root: they treat it as a bookkeeping expense instead of an economic asset. Drawing on Adam Smith's 1776 distinction between accounting value and economic value, Bill explains why the biggest constraint on data programs is cultural, not technical, and why incumbents, not digital natives, may hold the real advantage.

Data as an economic asset: why the balance sheet gets it backwards

Bill's starting argument is that data has no inherent value until it is used. He traces this to Adam Smith's Wealth of Nations, which separates accounting value (what someone would pay for an asset) from economic value (what that asset produces in use). His example: two people each buy an identical $40,000 car, one an ordinary driver, one an Uber driver. The accounting value is identical, but the Uber driver's car generates far more economic value because it is used to create income. Applied to data, Bill says possession is actually a cost, since storing, backing up, and protecting data all cost money, and a data set only generates value once it is put to work, from hospitals reducing hospital-acquired infections and unplanned readmissions to manufacturers cutting unplanned downtime through predictive maintenance.

The use case trap: why companies fail from too many ideas, not too few

Organizations rarely struggle to find use cases for data, Bill says, they struggle to choose between too many of them. His fix is a workshop methodology he calls the Art of Thinking Like a Data Scientist: identify, validate, value, and prioritize use cases with the actual business stakeholders in the room, plotting each on a value versus feasibility chart and picking the ones with both high value and a high chance of success. He argues against building a full data lake with 35 data sets before doing anything: a single use case might only need three data sets, so build only what that use case requires, then reuse the same governed data across the next use case at zero marginal cost. He calls this the data economic multiplier effect, the reason data behaves unlike any traditional asset.

Hub and spoke: the org model Bill recommends over a centralized data team

Bill's recommended structure has a central hub of data scientists and data engineers who productize models into repeatable, reusable applications, and spokes of data scientists embedded inside business units working directly with stakeholders on live problems. Ideas and hardened models move fluidly between the two: the periphery surfaces opportunities close to customers and operations, and the hub productizes and redistributes what works. He argues incumbents with legacy products, from theme parks to CAT scan manufacturers, have an underrated advantage over digital native companies, because they already have the customers, channels, and products to layer intelligence onto.

Why big-bang transformation always fails: the friendly-first playbook

Bill is blunt that trying to mandate a top-down cultural transformation in one move is "impossible, it'll fail, guaranteed." His alternative is to find one friendly business unit willing to try a different way of working, prove a use case within nine to twelve months, and let visible success from that first win pull other business units in voluntarily. He says incumbents can realistically hit meaningful impact in nine to twelve months if they prioritize one problem rather than trying to boil the ocean, and that subsequent use cases speed up from there as data, tooling, and cultural trust compound. Big-bang mandates imposed from the top get killed by what he calls organizational passive-aggressive behavior, not by any technology limitation.

Decisions, not data or questions, are where value actually gets created

Asked how to create value from data in a specific setting like upstream oil and gas, Bill said the same answer applies everywhere: focus on the decisions people are trying to make, not the data itself. He traces this insight to his time as vice president of advertiser analytics at Yahoo, where he realized the value creation process was never about data or better questions, it was about better decisions, such as who to target with which ads and how much to pay for them. Decisions, he argues, are universally understood by stakeholders, can be tied directly to measurable value, are actionable in a way questions are not, and are the natural interface between business teams and data science teams.

Why does Gen AI worry Bill less than autonomous analytics?

Bill argues Gen AI, despite the hype, is essentially a more advanced knowledge management system: it finds correlations between words and presents them fluently, but does not possess real knowledge, so users still have to think for themselves. He is more excited by what he calls autonomous analytics, reinforcement learning systems that continuously learn and adapt with minimal human intervention, as seen in self-driving vehicles. He believes the same approach could extend beyond physical products like compressors and refrigerators into internal processes such as hiring and admissions, reducing confirmation bias over time.

Is Bill worried about AI's impact on jobs?

Not in terms of net job losses. Bill's stated concern is a widening gap between people who have AI and data literacy and those who do not, which is why he wrote his book on AI and data literacy for citizens of data science. He frames the real risk as a divide in job quality and opportunity rather than headcount, arguing the fix is making sure everyone in an organization understands where and how to apply data and analytics to their own work.