Back in 2013, I wrote a version of this post when "Big Data" was the phrase on everyone's lips. FitBit was the hot new thing. Farmers were starting to collect soil moisture data. The Walmart PopTarts-and-hurricanes story was doing the rounds.

Eleven years on, the fundamental hierarchy hasn't changed: data becomes information, information (sometimes) becomes insight, and insight (occasionally) drives action. But the landscape around it has transformed completely—and I think it's worth revisiting.

Data is still raw. That hasn't changed.

Data on its own has no meaning. It's the IP address in your server logs, the temperature reading from a sensor, the transaction record in your database. Individually, each data point tells you nothing useful.

What has changed is the sheer volume. In 2013, a few hundred data points per customer seemed impressive. Now we're measuring attention spans in milliseconds, tracking micro-interactions across dozens of touchpoints, and logging behavioural signals that would have seemed invasive a decade ago.

The old term "datafication"—turning previously unrecorded activities into stored data—has accelerated beyond what we imagined. Your watch measures your heart rate variability while you sleep. Your car knows your driving patterns better than you do. Your email provider understands your response times by sender category.

We collect everything now. The question has become what to do with it all.

Information is organised data—but organisation is the easy part

The traditional definition holds: information is data that's been structured and given context. The IP addresses become geographic distributions. The temperature readings become trend lines. The transaction records become customer profiles.

Here's what's different in 2025: this step is now essentially free. Aggregation, visualisation, basic pattern recognition—these are commoditised. Every SaaS product comes with a dashboard. Every database query tool can generate charts. The technology that seemed sophisticated in 2013 is now baked into every spreadsheet.

This means most organisations are drowning in information. They have dashboards for their dashboards. Monthly reports that nobody reads. Weekly metrics emails that get archived unread. The bottleneck has moved downstream.

The insight problem hasn't been solved—it's been transformed

In my original post, I described insight as "something novel and profound that may lead to competitive advantage." That definition still works, but the path to getting there has changed dramatically.

I wrote about two approaches back then: hypothesis-driven analysis (decide what you're looking for, then seek evidence) and correlation-based discovery (let the patterns emerge from the data). Both had limitations.

The hypothesis approach meant you only found what you were looking for. The correlation approach gave you relationships without causation—you might discover that umbrella sales correlate with ice cream sales, but that doesn't mean one causes the other. They're both responding to weather.

Machine learning and AI have changed this calculus, but not in the way the hype would suggest.

What AI actually offers—and where it falls short

Large language models and modern ML systems are genuinely good at pattern recognition across massive datasets. They can surface correlations that would take humans years to discover. They can identify anomalies that don't match historical patterns. They can cluster behaviours and segment populations with nuance that manual analysis couldn't achieve.

But here's what I've learned from working with these systems: they're excellent at generating candidates for insight, not insight itself.

The AI might tell you that customers who contact support within their first week are 40% less likely to churn—and that's valuable. But whether that's because early support contact builds confidence, or because the customers who bother to ask questions are more engaged to begin with, or because your onboarding is confusing... that interpretation still requires human judgment.

Insight remains the application of context, experience, and strategic thinking to information. AI has accelerated the upstream process enormously. The bottleneck has shifted from "finding the patterns" to "knowing which patterns matter."

The datafication reckoning

There's something else that's changed since 2013 that deserves attention: we've had a reckoning about what datafication actually costs.

GDPR arrived in 2018. Privacy has become a genuine concern for consumers, not just a theoretical one. "Surveillance capitalism" moved from academic criticism to mainstream vocabulary. Companies have faced real consequences for data practices that seemed unremarkable a decade ago.

This matters for the data-to-insight pipeline because it constrains what you can collect in the first place. The "datafify everything because you never know what will be useful" approach I described in 2013 is no longer viable—legally, reputationally, or often technically.

The better question now is: what data do you actually need to generate the insights that would change your decisions? Working backwards from decision to required insight to necessary information to essential data is more valuable than hoarding everything and hoping patterns emerge.

Where this leaves us

The hierarchy remains sound: data ? information ? insight ? action. But the practical reality has inverted.

In 2013, the hard part was capturing data and turning it into usable information. In 2025, the hard part is filtering signal from noise when information is abundant, and translating patterns into genuine strategic insight when AI can generate endless candidates.

The competitive advantage has shifted accordingly. It's no longer about having more data than your competitors, or more sophisticated dashboards. It's about the quality of questions you ask, the judgment you apply to AI-surfaced patterns, and the speed with which you can translate insight into changed behaviour.

The takeaway

If you're building data capabilities today, start from the decisions you need to make rather than the data you could collect. Ask what insight would actually change your actions. Work backwards from there.

AI tools can accelerate every step of the pipeline—but they can't tell you what matters. That's still the human part of the job. And increasingly, it's the only part that creates genuine competitive advantage.