Business banking, rebuilt, then measured

A 2024–2026 platform redesign that became a configurable design system, and the marketing site as it turned from a manifesto into a product.

Role
Head of Design /
Founding Designer
Team
  • Andrey Schegolev
  • Alexandr Volynin
  • Elena Bakulina
  • Vladimir Belyakov
Surfaces
  • Responsive web
  • iOS
  • Android
Recognition
  • Banking Tech Awards USA 2026
  • Best User/Customer Experience Initiative for Business

Unlisted page, please keep it within the team. Figures are relative, not absolute.
The white-label partner is under NDA; the themes shown are neutral demonstration themes.

Site evolution: the bank's front door, step by step

What this is

Two very different ideas of what a bank's first screen should say.

The public site is the first thing a business sees before it opens an account — often before it has spoken to anyone at the bank. Over four years it was rebuilt several times, and not every rebuild was about looks. Some versions changed the style and kept the story; one changed the story itself and stopped describing a mission to show the product.

Laid side by side at the same width, the versions make the difference visible: when the site changed its style, and when it changed its job.

2022–24
6,153 px
2024 · Redesign edition
3,213 px
Early 2025
5,399 px
Late 2025 · live
8,050 px
Early 2026 · unreleased
6,250 px
2026 · direction, unreleased
5,131 px
Six versions of the main page, to scale — open any one full size; where tablet and mobile layouts exist, they open alongside. The page did not get shorter over time; it got longer once it started carrying product.

Line A: one manifesto, two styles

From 2022 to early 2025 the site told the same story in the same order: discover, banking is the mission, borderless is the vision, designed for the digital age, premium, secure by design, grow your global footprint, and a final call to open an account. The skeleton matches section for section. What changed was the voice: dark 3D art gave way to a light interface language. That was an argument as much as a taste — the heavy 3D slowed the pages down, and after a long back-and-forth with the front-end team we agreed to cut graphical weight for speed.

2022–2024Manifesto, dark 3D art3D and visual direction by Stan Zubov, then art director
2022–2024 hero: Discover, banking beyond borders, over dark 3D artwork

Redesign edition

By late 2024 the site had fallen behind the bank. The product had just been redesigned, and the dark 3D art made us look more like a crypto exchange than a serious bank. This edition kept the manifesto but swapped the art for product: a stricter grid, crisper type and lighter pages that load faster.

The work covered the main page, the business and individual product pages, a simplified pricing page, and early drafts of careers and FAQ.

2024 · Redesign editionMain and product pages

Take it wavy or easy?

Early 2025 tried the opposite move: keep the content, add motion. Shader backgrounds carried the mood the 3D art used to carry, at a fraction of the weight. It did go live, for about three months. Stakeholders found it too complex and too dark: the heavy, layered plates pressed on the content instead of framing it. That reaction set the brief for what came next — lighter, flatter, product first.

Early 2025Same manifesto, light interface language, shaders

Every shader in these cases is mine, made in Unicorn Studio —
and so is the motion across these cases, made in Jitter.
I'm an official expert in both tools. Some examples.

Late 2025Product in the hero

Break A:
the site becomes the product

In late 2025 the first screen stopped describing a mission and started showing the product: the app itself in the hero, the headline "Manage your money from one place", then stablecoins, pricing, use cases, the customer journey in five steps and the mobile app.

The page grew from about 5,400 to 8,050 pixels, and all of the growth is product content. That was a positioning decision as much as a design one, and it is also why the redesigned platform from the first case now sits in the site's first screen.

Break B:
one product, several art directions

With the content settled, we explored how the same product story should look. The directions share the same blocks — the headline, one card for fiat and crypto, stablecoin payments, the customer journey, licensing, the app — and differ only in treatment.

One of them, a dark striped hero with coloured glass textures, I took all the way to a working build myself — coded end to end with Claude Code, directed through Design Brain, the design harness I built — so the direction could be judged in a browser, not in a static frame.

Early 2026Dark striped hero, glass texturesunreleasedarival-site.vercel.appBuilt by me end to end with Claude Code, driven by Design Brain, my own design harness.
Early 2026 direction: dark striped hero with the app

Where it is now: three live entry points, A/B tested

While we were building the white-label system, we also decided to try more local versions of the site — pages tuned to a specific audience instead of one page for everyone. That work led to the current versions and the A/B tests around them.

The live site now runs three entry points: a root page that splits into "for all" and "for business", a business page and an individual page. It is an experiment, measured with A/B tests: a deliberately more mature, conservative presentation aimed at an older audience, to see whether it converts better than the lighter product pages.

In parallel, the business site is being rebuilt as a system — a local component library for the marketing surface and five breakpoints designed explicitly, so a new page is assembled from parts instead of drawn from scratch.

What else changed

In 2026 we also replaced the logo and settled the identity across the site, the apps and the back office. Among other reasons, the old mark looked a little too much like a certain London fintech. Given who's reading this, it felt polite to mention.

RootSplit hero: for all, for business; the handle drags between themarival.com
BusinessBanking built for borderless businessarival.com/promo
Business page hero: capitalised headline, open account button, card and the web app
IndividualManage your money from one placearival.com/individual-account
Individual page hero: blue background, headline and a person holding a phone

Who built the new versions

For the latest versions of the site Vladimir Belyakov joined my team, and the site was his entry project. He already knew the bank from the inside: as a designer at Arival he drew the very first version of the product interface and the zero version of the site, before the design function existed. The new versions he designed under my direction and side by side with me — I designed the additional options, built the UI kit the pages are assembled from, and worked through the responsive versions with him.

What I take from it

Style changes are cheap to see and easy to argue about. The change that mattered was structural: the moment the site stopped telling a story about the bank and started showing the bank's product.

Everything after that — the art directions, the live A/B test, the component library — is about making that product story fast to change and cheap to test.

If I had to keep one rule from all these versions, it's this: stop arguing about how the front door looks until you agree on what it's for. Every version that changed its job moved the business. The ones that only changed their style moved the discussion.

Pet projects

How I work now

Alongside the team, I run design through my own harness, Design Brain.

It sits on Claude: it turns a design decision into something a coding agent can build without guessing, and then checks what came back.

Review~640 rules

Every layout is reviewed against a library of design rules across twelve lenses — usability, visual, information architecture, motion, copy and others — each tied to its source.

SpecFrom the file

It writes implementation specs that coding agents build from, with values taken from the design file itself, not from memory or a description.

CheckSeen, then done

The agents' output is checked against screenshots and crops before anything counts as done.

git log --since=2026-03
3,323 contributions in the last year, all of them since March — most in private repos. Every tool, site and pipeline on this page is in there.
MarAprMayJunJulAugSep2026-03-01: 02026-03-02: 02026-03-03: 02026-03-04: 02026-03-05: 02026-03-06: 02026-03-07: 02026-03-08: 12026-03-09: 02026-03-10: 02026-03-11: 02026-03-12: 02026-03-13: 12026-03-14: 02026-03-15: 02026-03-16: 02026-03-17: 02026-03-18: 02026-03-19: 12026-03-20: 02026-03-21: 02026-03-22: 02026-03-23: 02026-03-24: 02026-03-25: 02026-03-26: 02026-03-27: 02026-03-28: 02026-03-29: 12026-03-30: 02026-03-31: 02026-04-01: 02026-04-02: 02026-04-03: 22026-04-04: 02026-04-05: 02026-04-06: 02026-04-07: 02026-04-08: 02026-04-09: 02026-04-10: 02026-04-11: 02026-04-12: 02026-04-13: 02026-04-14: 02026-04-15: 02026-04-16: 02026-04-17: 02026-04-18: 152026-04-19: 72026-04-20: 122026-04-21: 262026-04-22: 112026-04-23: 152026-04-24: 242026-04-25: 222026-04-26: 02026-04-27: 02026-04-28: 02026-04-29: 02026-04-30: 02026-05-01: 02026-05-02: 252026-05-03: 52026-05-04: 152026-05-05: 182026-05-06: 22026-05-07: 72026-05-08: 12026-05-09: 122026-05-10: 142026-05-11: 02026-05-12: 02026-05-13: 02026-05-14: 32026-05-15: 02026-05-16: 82026-05-17: 92026-05-18: 82026-05-19: 22026-05-20: 92026-05-21: 132026-05-22: 132026-05-23: 252026-05-24: 112026-05-25: 82026-05-26: 92026-05-27: 72026-05-28: 132026-05-29: 122026-05-30: 102026-05-31: 122026-06-01: 82026-06-02: 142026-06-03: 142026-06-04: 272026-06-05: 22026-06-06: 132026-06-07: 112026-06-08: 92026-06-09: 92026-06-10: 232026-06-11: 142026-06-12: 302026-06-13: 72026-06-14: 82026-06-15: 32026-06-16: 02026-06-17: 52026-06-18: 82026-06-19: 142026-06-20: 52026-06-21: 92026-06-22: 32026-06-23: 02026-06-24: 02026-06-25: 12026-06-26: 52026-06-27: 142026-06-28: 162026-06-29: 172026-06-30: 42026-07-01: 32026-07-02: 292026-07-03: 292026-07-04: 72026-07-05: 332026-07-06: 172026-07-07: 312026-07-08: 02026-07-09: 02026-07-10: 132026-07-11: 152026-07-12: 262026-07-13: 112026-07-14: 52026-07-15: 52026-07-16: 42026-07-17: 02026-07-18: 02026-07-19: 192026-07-20: 142026-07-21: 122026-07-22: 432026-07-23: 372026-07-24: 02026-07-25: 02026-07-26: 162026-07-27: 372026-07-28: 382026-07-29: 52026-07-30: 332026-07-31: 112026-08-01: 62026-08-02: 112026-08-03: 192026-08-04: 02026-08-05: 02026-08-06: 182026-08-07: 142026-08-08: 12026-08-09: 02026-08-10: 12026-08-11: 52026-08-12: 222026-08-13: 632026-08-14: 1112026-08-15: 402026-08-16: 412026-08-17: 542026-08-18: 832026-08-19: 292026-08-20: 82026-08-21: 02026-08-22: 02026-08-23: 222026-08-24: 242026-08-25: 52026-08-26: 12026-08-27: 92026-08-28: 62026-08-29: 242026-08-30: 732026-08-31: 1102026-09-01: 652026-09-02: 372026-09-03: 72026-09-04: 02026-09-05: 02026-09-06: 12026-09-07: 192026-09-08: 42026-09-09: 1192026-09-10: 482026-09-11: 02026-09-12: 02026-09-13: 502026-09-14: 1412026-09-15: 322026-09-16: 02026-09-17: 02026-09-18: 02026-09-19: 02026-09-20: 212026-09-21: 522026-09-22: 532026-09-23: 1142026-09-24: 2092026-09-25: 1662026-09-26: 02026-09-27: 1112026-09-28: 792026-09-29: 0

My own product, outside the bank

And a product I build for designers: Glitchy Brain.

A Telegram bot and a web app over one knowledge base I curate: about 2,000 searchable entries on tools and techniques, 1,900 visual references and 450 styles, linked by 7,000 connections across 133 tags. The bot is open to anyone; the web app is private for now. Next to it run glitchy.computer, a hub of 18 generative tools, and a daily design digest.

Glitchy Brain web app: a search for prototyping with the Figma dossier open
Web app · search with a tool dossier
Glitchy Brain web app on a phone: search results for Cavalry
Mobile · 273 results for “cavalry”
The Telegram bot answering lookups for Blender and Figma
Telegram bot
Bot: open to anyoneWeb app: privatedemo on request