Landscape of data

I spent several rounds of research sitting with management consultants.

The instinct is to give consultants more data, or better data. In fairness, both are good things. Better data is rarely the wrong investment. But data, however good, only pays off when the person using it can move between altitudes.

A consultant diagnosing a business has to see it at several layers. Zoomed in, they need revenue, products, the tech stack. A step out, they need competitors, market position, growth relative to the field. Further out, they need the full landscape: the players in the market, the businesses upstream and downstream that shape the conditions their client is operating in. Each layer surfaces different causes, different risks, different leverage points. Consultants who work from one altitude solve for what they can see. The good ones solve for what's actually happening.

We already know how to do this. Maps have worked this way for centuries on paper, and for two decades on screens.

Zoom in on London and you see the London Eye, the path along the river, the pier it sits on. Pull back and the Eye becomes one landmark among many: Westminster, the bridges, the South Bank. Pull back further and London itself becomes one node in a country, surrounded by rivers, motorways and green belts that shape how everything moves. The place doesn't change. Your understanding of it does.

The same logic applies to business data. Zoomed in: revenue, products, the tech stack. A step out: competitors, market position, who's growing faster than who. Further out: the full market, the upstream suppliers, the downstream buyers, the broader forces acting on the industry. The business doesn't change at each altitude. What you understand about it does.

The workflow I built tries to do two things. It reads intent from how a consultant searches, and starts them at the altitude that matches it. From there, the interface stays out of the way: progressive disclosure and a careful layout let them drop into detail or pull back to context without losing the line of inquiry they came in on.

One detail kept getting called out in testing. Instead of the page growing endlessly from the bottom, the way ChatGPT and most deep research tools do, consultants could branch off at the point of interest. They could open up a thread of inquiry without losing the one they came in on. Consultants who saw it described it as a frustration they hadn't known they had. They'd lived with the alternative for years, and only recognised it as a problem the moment a different version existed.

If you're building anything that delivers information to someone making a decision, the question to ask isn't "how much data should I show?" It's "how easily can they change altitude on it?" Dashboards, research interfaces, AI assistants, internal analytics products: they all serve someone trying to make a decision, and they all default to dumping data at a single altitude.

Data tells you what is. Altitude tells you what it means. The tools that figure out how to give people both are the ones I'd bet on.