What will doing data analysis actually look like?

Will we still sit in front of a grid? Talk to a model while walking through a warehouse? Put on a headset and wander around a forecast someone else—or some AI—built? I have guesses. I would be suspicious of anyone selling certainty.

The spreadsheet may age more like a book

I have a hard time believing the spreadsheet disappears. A grid is an awfully good way to arrange information. Put related things next to each other, compare rows, change an assumption, see what happens. There is a reason people keep using it, including people who have been told repeatedly that they should have moved on by now.

In that sense, the spreadsheet reminds me of a book. The basic form works. Books survived radio, television, and the internet, although their place in our lives changed. A book no longer has to carry every kind of information. I would much rather search a train schedule than consult a bound volume of departures. I still want a book for other things.

That is roughly how I expect spreadsheets to persist: as a persistent, almost timeless user interface, with a changing share of the work. We might spend less time constructing every formula ourselves and more time inspecting a model, challenging assumptions, or returning to the grid for a precise comparison. Less time in a spreadsheet does not necessarily mean less value from one.

It also does not follow that every future interface needs to look like Excel. A book can be a good form without every website pretending to be a page-turning book.

Could we actually walk through a model?

I know. Headsets… again.

The idea of wearing AR or VR equipment to do ordinary office work leaves a lot of people cold, and I understand the reaction. I am not especially eager to turn answering email into an immersive experience. Comfort, cost, distraction, and looking ridiculous all matter. A technology has to earn its place.

But I can imagine a more specific use: helping us understand a complicated model.

Excel already has ways to trace relationships between formulas and cells. Suppose those relationships became a space you could explore. Inputs here. Assumptions over there. A cluster of calculations feeding an output. Change one assumption and watch the consequences travel through the model.

Imagine asking an AI to build a forecast, then being able to walk through what it created with a colleague. You follow the path from source data to revenue, notice that returns have been left out, and open the underlying calculation. The point would be to inspect the logic, not admire a giant floating spreadsheet.

That is a possibility I would like to see tested, especially when AI makes it easier to produce models than to understand them. It is not a claim that today’s headsets can reliably audit your workbook. A three-dimensional mess is still a mess. For plenty of tasks, the flat grid will be clearer and faster.

UCSB is already exploring data you can see and hear

There is a real research setting behind this line of thinking: the AlloSphere at UC Santa Barbara. Directed by JoAnn Kuchera-Morin, it is a shared immersive facility where researchers can explore data through surrounding visuals, spatial audio, and interaction. This is an instrument for research across science, engineering, and the arts. It is not a prototype Excel office.

The work includes sonification: representing data through sound. UCSB researchers have explored visual and auditory representations of quantum models. A different example, Sketches of Sensorium, brought climate and ocean-health data into an immersive audiovisual installation. UCSB’s account describes contributions from the Ocean Health Index at its Halpern Lab and ocean-climate data from NASA scientists working with University of Maryland researchers.

That interests me because a chart already asks us to translate symbols into something we can perceive. A higher bar represents a larger quantity. A different color represents a category. Sound gives us another set of choices: pitch, rhythm, loudness, location.

You do not have to visit a research facility to encounter the idea. NASA’s data sonifications translate astronomical data into sound, and its accessibility work discusses how this can help blind and low-vision audiences experience the data. These are designed translations of data; you should not assume you are hearing a microphone recording of the object in an image.

My speculative business example would be a repeating sound for a normal process, with a change in that pattern drawing attention to an anomaly. Whether that helps would need testing. What does each sound mean? Can people distinguish the signals? Does it reveal something a chart misses, or just create a noisy office? We would still need a clear mapping, access to the values, and other ways to inspect the data.

No, I do not think keyboards are going away

The prediction that we will stop typing entirely strikes me as silly. Voice is useful. It does not follow that every other way of putting a thought into a computer has become obsolete.

The accessibility benefits are substantial. Microsoft’s Windows voice access, for example, supports controlling a PC and authoring text by voice, including for people with mobility disabilities. More ways to participate in the work are a good thing. Voice can also be a convenient way to get a rough thought down or ask a question when your hands are occupied.

But I may not want to announce a client’s numbers to everyone at the airport. I may want to revise a sentence before anyone hears it. I may need to change one cell reference, check punctuation, compare two versions of a formula, or work quietly beside other people. Typing lets me do those things. Precision and privacy are not temporary inconveniences waiting for a better microphone.

There is also a difference between getting words out and working out what you mean. Sometimes talking helps me think. Sometimes I need to look at a sentence and change it. I do not want an interface that assumes those are the same activity.

And the choice is broader than voice versus keyboard. Handwriting, sketching, and paper can remain part of the workflow. Microsoft documents both converting digital handwriting to text in OneNote and extracting text from pictures with OCR. These are distinct capabilities, and recognition still needs checking. But the general direction appeals to me: use a useful way to think, then bring the result into the analysis.

I expect people to switch among these methods, sometimes within the same task. Talk through a question. Sketch the process. Type the exact condition. Review the result in a table.

The future I want involves getting out of the building

There is another possibility here that I find more appealing than an infinite virtual office: analysts spending more time with the things they are analyzing.

Data are symbols. A row might stand for a shipment, a patient visit, a maintenance job, or an hour of somebody’s work. The row is useful because it leaves things out. It is also dangerous when we forget how much it leaves out.

Consider a warehouse report showing that picking time has increased. You can slice it by shift, product, and employee. You can ask an AI for explanations. You can build a beautiful model. Walking the route might reveal that the popular items were moved, a scanner is unreliable, or the recorded start time no longer means what you thought it meant.

That visit does not replace the data. It helps you understand the data and decide what to investigate next. One observation can mislead too, so you still need to check whether it holds across the process.

If computing becomes easier to carry, speak to, or consult in context, perhaps more analysis can happen near the work itself. Ask a question while looking at the equipment. Capture a note while the person who understands the process is standing there. Return to the desk with a better question.

I would like AI to give analysts more time for that. Otherwise we risk getting much faster at producing explanations for a world we have not bothered to look at.

Different tools, some very familiar responsibilities

I expect the places and interfaces of analysis to keep changing. I also expect some unfashionably familiar responsibilities to remain: define the question, understand where the data came from, check the assumptions, look for errors, and explain what the result does and does not support.

Those responsibilities may become more demanding when producing an analysis is easy. We will have more outputs to judge, including convincing ones we did not build ourselves.

So I would keep learning the tools without treating every new interface as a referendum on whether the old ones must die. Keep the spreadsheet when the grid helps. Use voice when speaking helps. Try a spatial or auditory view when there is a real question it might help answer. And occasionally close the laptop and go see what the numbers are about.