Why we obsess over design in software built for the back office.
Most finance software is built to be tolerated, not used.
You know the type. Dense, gray, packed with fields and toggles nobody explained. You operate it the way you operate a terminal: carefully, defensively, always one wrong entry away from a broken run. Somewhere along the way the industry decided that back-office software does not need to be good, because the people stuck using it do not have a choice.
We think that bargain is broken. Not because ugly tools are unpleasant, though they are. Because of what they do to the person using them.
The old bargain
For a long time, effort and quality were the same signal. If something looked polished, someone had clearly labored over it, and you trusted it more. That held for a spreadsheet, a report, a piece of software. The more thought went in, the better it looked, and the better it looked, the more you paid attention.
This is not just intuition. It is one of the most replicated findings in human-computer interaction. In 1995, two researchers at Hitachi, Masaaki Kurosu and Kaori Kashimura, tested 26 versions of an ATM interface with 252 people. The interfaces users rated as more beautiful, they also judged easier to use — and that link to perceived usability was stronger than the link to actual usability. Nir Tractinsky replicated it across cultures in 2000 in a paper whose title became the shorthand for the whole effect: What Is Beautiful Is Usable. Strikingly, the beautiful-feels-usable bond held even after people had actually used the system. We read the care in a tool before we can articulate why we trust it.
The writer Tomas Pueyo has a sharp framing for what AI did to that link. He calls it shallow intent. AI can now produce work that looks finished without the thinking that used to sit behind every choice. The polish is real; the intent underneath is missing. And people can feel the gap. They see something that advertises depth, lean in, and find nothing there. They feel cheated.
His point was about images and text. It applies just as cleanly to software, and it is about to matter enormously in finance.
Building a tool used to be expensive, so shipping one at all was a signal of seriousness. That is over. Anyone can stand up an AI reconciliation tool this quarter. The demo will look clean. What separates the tools worth using from the ones that get abandoned in month two is exactly what Pueyo describes: whether someone cared about the hundred small interactions that never show up in a demo. The unmatched item that needs a second look. The moment the numbers do not tie out and you need to know why, fast, without fighting the interface. Shallow intent in a poster is a disappointment. Shallow intent in the tool your finance team runs the close on is a liability.
It is not about aesthetics. It is about cognition.
Here is the part most people miss. The argument for good design in finance software is not that it is nicer to look at. It is that the tool you work inside all day shapes how you think inside it.
This is where the research stops being about taste and starts being about the brain. In 2026, a study in the Journal of Environmental Psychology — “When science isn’t beautiful” — put people to work in two versions of the same lab, one bare and one aesthetically considered, and measured them. The pleasing space did not just feel better; it measurably changed reaction times and reactive cognitive control. The room people worked in altered how their attention performed.
Why would that happen? Because environment drives affect, and affect drives cognition. In a now-classic 1987 experiment, Alice Isen gave people a small, pleasant nudge — a few minutes of a comedy clip, a bag of candy — then handed them a genuinely hard creative problem, Duncker’s candle task. The positive-affect group solved it 58% of the time; the neutral group, 13%. Mild good feeling gave people broader access to remote associations — the exact cognitive move behind noticing what does not fit. Don Norman built his book Emotional Design on this bridge, and his line is the one worth keeping: attractive things work better, because feeling good widens how you think.
Now run that in reverse, which is what most finance software does. A tool that fights you keeps you in defense mode. You are checking, re-checking, bracing for the next error, holding half the process in your head because you do not trust the software to hold it for you. That takes up room. It uses the exact mental capacity you need for the actual work.
A tool that feels clear and calm gives that capacity back. It frees the part of your brain that does the finance rather than the data entry: spotting the anomaly, asking the sharper question, catching the thing the numbers are quietly hiding. The interface is not neutral. It is either taxing your attention or protecting it.
Cognition changes output
And this is the whole reason it matters commercially, not just aesthetically. A controller working in a tool they trust produces different work than one wrestling a spreadsheet from 2011.
The skeptic’s fair question is whether beauty actually changes output, or just how people feel about it. In 2026 a meta-analysis in the International Journal of Human-Computer Interaction — titled, fittingly, “Attractive Things Do Work Better” — pooled 31 studies, 234 effect sizes, and 18,794 participants. It found a small-to-medium positive effect of visual aesthetics on real task performance (g ≈ 0.29). Honest caveat: the effect varies a lot by task and context, so this is a real edge, not a magic wand. But the direction is settled. Better-designed tools do not just feel better. They help people do the work better.
The person who is not burning attention on the mechanics has attention left for judgment. They notice the reconciliation break that is not really a timing difference. They ask why a number moved before it becomes a problem. They close faster, with fewer errors, and they see more. Not because they are smarter that day, but because the tool stopped standing in their way.
That is the return on design in this category. Not satisfaction scores. Better work, from the same people, because you changed the conditions they think under.
First we found the soul. Then we built the site.
Here is the part that makes this personal, not theoretical.
Yes, we built Rexi’s new site with AI. Everyone does now, and pretending otherwise would be a lie. But look around: the web is filling up with vibe-coded pages that all feel the same. Competent, fast, forgettable. Shallow intent, shipped at scale. We did not want to add one more.
So we did it in the other order. Before we built the site, we went looking for its soul. That meant real, artisanal work, with an agency, by hand. We built 3D animation the old way to feel out every movement: how the image should breathe, what it should do to you in the first second. Only then did we define the Rexi crystal — its material, its light, the way it holds and refracts — and carried that through to the final render. There is craft sitting under every frame, and you can feel it. The site does not look like everyone else’s, and we are proud of that. Not because it is decorated. Because it has intent, all the way down.
That is also how we think about AI, everywhere — not just in pixels.
AI runs through the whole Rexi platform, and we lean on it hard exactly where it multiplies a person: making sense of a messy file, drafting the reconciliation summary, surfacing the item worth a second look. But we are just as deliberate about where it does not belong. Our reconciliation engine — the hard, precise core where the numbers have to tie out — has no AI doing the math. You cannot hallucinate a balance. In finance, a plausible-looking wrong number is worse than no number at all, so the one place precision is non-negotiable is the one place we keep AI out of the answer.
Design is the same discipline pointed at a different surface. You do not generate the soul with a prompt. You decide where the machine helps and where a human has to care, and that judgment is the whole game. Knowing where AI belongs, and where it does not, is not a limit on taste. It is taste.
Taste is a feature
So at Rexi we treat design as part of the product, not paint applied at the end. Taste is not decoration. In a tool people live inside for hours a day, how it feels is not separate from how well they work. It is one of the main things that determines how well they work.
Finance has been trained to expect the opposite: that serious software is supposed to be hard, and anything easy must be shallow. We think that is backwards, and AI is about to make it obvious. When execution is cheap and everyone can ship something that looks finished, the tools that win are the ones where someone clearly cared, all the way down. The intent shows. You can feel it in the first ten minutes.
We are building the tool we wished existed when we were the ones doing reconciliation at 11pm. Clear instead of dense. Calm instead of defensive. Built so the person using it can spend their attention on the work that actually needs a human, and hand the rest to something that will not make them brace for it.
Taste is a feature. We are treating it like one.
Sources
- Kurosu, M., & Kashimura, K. (1995). Apparent usability vs. inherent usability. CHI ’95 Conference Companion, 292–293.
- Tractinsky, N., Katz, A. S., & Ikar, D. (2000). What is beautiful is usable. Interacting with Computers, 13(2), 127–145.
- Isen, A. M., Daubman, K. A., & Nowicki, G. P. (1987). Positive affect facilitates creative problem solving. Journal of Personality and Social Psychology, 52(6), 1122–1131.
- Norman, D. A. (2004). Emotional Design: Why We Love (or Hate) Everyday Things. Basic Books.
- When science isn’t beautiful: Lab aesthetics impact reaction times and reactive cognitive control (2026). Journal of Environmental Psychology.
- Attractive Things Do Work Better: A Meta-Analysis on Visual Aesthetics and User Performance (2026). International Journal of Human-Computer Interaction.
A note on the evidence: each claim is tied to a peer-reviewed source or the original author’s own words; effect sizes are reported as published (meta-analysis g ≈ 0.29, with its heterogeneity caveat stated); popular framings (Pueyo’s “shallow intent,” Norman’s “attractive things work better”) are attributed by name rather than presented as empirical findings.
