Senior Product Designer · Fintech & AI

I make complex fintech and AI products feel effortless.

Ten years in fintech, banking and telecom across the Netherlands, Spain and the UAE. I work on heavy, high-stakes journeys: I find where the real complexity lives, cut through it fast, and carry stakeholders with me. Most recently: payments, transaction intelligence and AI assistants in MENA fintech.

Focus

Fintech · AI products

Based in

Dubai, UAE

Open to

New roles & advisory

Selected work

Three case studies

About

I take financial products that are genuinely complex underneath and make them feel effortless on the surface.

A decade across four markets and most of the hard corners of fintech: banking, payments, B2B self-service, telecom billing, and lately AI-native products. I work end to end, but my strength is upstream: framing the real problem, finding where complexity actually lives, and aligning stakeholders around a direction before a single screen is drawn. Then I execute, fast and to a high bar.

berlat.design

Independent UX practice

  • ·
    UX audits: deep teardowns of fintech products, the kind that get shared
  • ·
    Advisory: design partnership for founders building in MENA
  • ·
    AI-native UX, including Arabic-first interfaces

Writing

LinkedIn · Medium

Let's talk

Building something complex?
Let's make it feel simple.

olgacherepennikovaux@gmail.com

Etisalat Business · B2B self-service

Quick Pay & Recharge

A single-page redesign of a working but slow payment flow, built to make a "quick" payment actually feel quick.

Role

UX Lead, end to end

Context

Etisalat Business (B2B self-service)

Platform

Web + adaptive

Timeline

6 weeks · October 2025

Status

Concept, not yet shipped

Where I started

Quick Pay exists so business customers can pay and recharge in seconds without logging in or calling anyone. Every extra step works against the only reason the page exists: keeping payment volume fast, self-served, and off the call center.

Etisalat Business already had a Quick Pay & Recharge flow in production, covering bill payment and recharge, with batching built in. The trouble was simple: for a flow sold as a fast, no-login guest payment, it wasn't quick.

See the existing flow live ↗

THE EXISTING FLOW · LOG IN, THEN A SCREEN PER STEP Log in Choose payment type Bill Payment · Recharge Choose method Account · Bulk upload · Status ↓ then, for the selected path Accounts Confirm Contact Terms Payment Done
The existing journey: a login, then bill payment or recharge, each split across a separate screen per step.

The brief

The mandate from my manager was specific: audit the UX of the existing flow and update its components. Quick Pay & Recharge was a special case, with no dedicated components of its own, so a real part of the job was adapting our design system to fit it.

That was the ask. In parallel, I explored whether the flow itself could be revamped to genuinely save customers time, not just refreshed.

How I worked

Six steps, from understanding the live flow to iterating with the team. Tap any step to jump to it.

01

Audit

I mapped every screen and path, bill payment and recharge, each by single account, bulk upload or status check, and marked where the friction sat. One finding shaped everything after: Quick Pay had no components of its own, it was stitched together from generic ones, so any redesign also meant giving it a proper home in the design system.

02

Benchmark

Before redesigning, I tore down the no-login Quick Pay flow of seven telecom operators: e& as the baseline and six competitors, read against the same five dimensions, entry, pay-versus-recharge model, amount, payment methods and confirmation. The short version: e&'s batch payment is genuinely ahead, but the market wins on two things e& does not do, a true one-field start and auto-fetching the amount due. Both target the same moment, where a payer hesitates, mistypes the amount, or gives up and calls instead. That is where payments are won or lost.

03

Ideation

With the two gaps named, I explored on paper: how few fields a payment really needs, where the amount could be fetched instead of typed, and how much of the journey could live in one place. The aim was never a prettier screen, it was fewer moments to stall, mistype, or give up and call.

04

Two versions

I deliberately designed two directions rather than one, so the product owner had a real choice instead of a single take-it-or-leave-it:

  • Version one: same logic, new skin. The existing multi-step structure kept intact, but rebuilt on updated components and brought fully back in line with the brand. Low risk, close to the original mandate.
  • Version two: the one-page concept. The whole journey collapsed onto a single page, type, method, accounts, contact, terms and payment in one place, each step confirming inline, a persistent order summary in view, and the batch-payment strength carried over. Higher ambition, bigger time saving.

Final design

The shipped flow, end to end: the existing logic rebuilt on refreshed components and realigned to the e& brand.

Version two, the one-page concept. The walkthrough below puts the whole journey on a single screen. The more ambitious option, and the one that was not taken.

VERSION TWO · ONE PAGE One page: type, method, accounts, live total, contact and payment, all inline Confirmation
Version two: the same journey collapsed into one continuous screen that confirms each step inline.
Version two: the one-page concept, walked through end to end.
05

Present findings to the PO

I brought the whole story to the product owner in one readout: where our flow loses people, what the market does better, and the two versions side by side. Anchored on competitors and completed payments rather than taste, the conversation stayed concrete. The PO backed the one-page direction and asked me to take it further.

06

Iteration loop

From there it was a loop, not a handoff: I iterated against the PO and management on scope, and engineering on what was buildable. Their read was decisive: the full single-page model was more than the platform could deliver in six weeks. Better to surface that inside the loop than after committing.

Where it stands

The multi-page flow stays in production for now, and the one-page version is the direction I would push to ship next.

If I carried it forward, I would sequence the ambition rather than drop it: prove the consolidated confirm-and-pay step first, then fold in the rest.

The bet is straightforward: fewer steps and no forced login should raise the share of payments completed in self-service, which is exactly what a guest payment page is measured on.

A few decisions behind it

This work was done at Etisalat Business. The screens under "Where I started" are the existing production flow; account numbers and figures are anonymized.

CBD · Transaction intelligence, with Lune

Making transactions make sense

Turning a raw bank feed into something people can read at a glance, and actually find when they need it.

Role

Product Designer, end to end

Context

CBD app, enriched with Lune

Platform

In-app experience

Timeline

Jan 2026 – May 2026

Status

Live

Where I started

A raw transaction is noise. Instead of a name and a logo, a purchase shows up as a machine code like IPP2526512598937: no merchant, no category, nothing to recognise. People scroll a wall of these and give up trying to see where their money went, or to find a single charge.

Lune turns that feed into something legible and searchable. I owned two sides of it: making each transaction instantly recognisable, and making the whole list easy to search and filter.

Recent transactions before and after Lune
Before Lune, a wall of cryptic strings and generic arrows. After, real merchants, logos, and a clear status on every row.

Making transactions legible

Enrichment gives every transaction a real merchant name, a logo, and a category. The design question was how to surface that so it aids recognition without turning each row into clutter.

Transaction detail before and after Lune enrichment
The detail view: a raw description like IPP2526512598937 resolves to Careem, with a category and sub-category attached.

This could not stop at one happy-path screen. I mapped enrichment across every transaction type, credit card, debit, transfers, ATM, bill payments, refunds, donations, and financial products, and every state each can be in: completed, pending, rejected, refunded, in both the list row and the full detail view.

The enrichment systemEvery cell designed, in both list row and full detail, before and after
Transaction typeCompletedPendingRejectedRefunded
Credit card
Debit
Domestic transfer
International transfer
Own account
ATM
Bill payments
Refunds
Donations
Financial products
10 transaction types4 states eachList row + full detailBefore & after= 160+ states designed

Finding a transaction

The other half was search. The original filter was a bottom sheet that put date first. I rebuilt it as a full-screen filter page: date, spending category, cards, direction, status, and amount in one place, with proper empty and loading states.

A/B filter variants, bottom sheet versus full-screen page
The two variants: Filter A, the original date bottom sheet, against Filter B, the full-screen filter page I designed.

Rather than ship on taste, I ran an A/B test on the date interaction: the original bottom sheet against the new full-screen page. The other filters were disabled to isolate it.

A/B test · date filter
Bottom sheet vs. full-screen page
12 participants, directional. Task: "You paid your phone bill last week, find the exact amount in your history."
MetricFilter 1 · bottom sheetFilter 2 · full page
Sample57
"Easy to complete"2.8 / 54.4 / 5
"Confident in result"3.8 / 54.9 / 5
Rated 4 or higher40%86%
Rated 2 or lower40%0%
Tapped the disabled filters60%29%
01
Filter 2 won on every metric. Filter 1 was bimodal: two people loved it, two rated it near zero. It does not fail gracefully.
02
People reach for category, not date, to find a known merchant. Most still tried the disabled category filter. "Find a phone bill" activates a merchant model, not a date one. That is a product signal, not just a UI one.
03
One prototype bug surfaced: the custom date picker needed several taps to open. A fix to make, not a pattern to judge.
Recommendation: adopt the full-screen filter page for the date filter.

Where it landed

[Outcome: what changed once it shipped, e.g. adoption of categories, filter usage, support tickets, retention. Add a real, anonymized number if you have one.]

A few decisions behind it

Draft: image slots and the bracketed lines are placeholders. Send me the screens (before/after enrichment, filter before/after + A/B variants) and the A/B details, and I will finalize the copy and drop the visuals in.

AI Assistant · CBD, built on Dyna

It started with a banner

I was asked to add one entry-point banner to a transactions screen. The variants landed so well that CBD handed me the redesign of the whole AI Assistant, built on Dyna.

Role

Product Designer, end to end

Context

CBD AI Assistant, built on Dyna

Platform

iOS app, conversational

Timeline

Feb 2026 – Jul 2026

Status

Live

The ask: one banner

CBD already had an AI Assistant. My brief was small: add an entry point on the transactions screen so people could summon it straight from their spending. I designed several variants, from a quiet AI-powered insights chip to a personal ask it to summarize your spending.

Five AI Assistant entry-point banner variants
Five entry-point variants, from a subtle insights chip to a personal ask.

From a banner to a rebuild

The variants landed. CBD liked the direction enough to widen the brief: not just the way in, but the assistant itself. What existed was a wireframe, the logic sketched but not resolved. I took it apart, unpacked every state and flow, and rebuilt it into a real experience.

The existing AI Assistant before the redesign
The starting point: a wireframe of the assistant, before the logic was unpacked.

A new AI Assistant

I rebuilt the experience end to end: a new UI, a new logo and identity, and motion. Two ways in, a floating AI button on the home screen and a hard swipe-up gesture. The assistant opens over the app with quick prompts and a chat that keeps every answer plain and checkable against your CBD records.

Redesigned AI Assistant: home floating button, spending assistant, chat
The floating entry point on home, the spending assistant, and a chat that answers in plain language.

One of the entry points is a hidden gesture, so I designed onboarding that teaches it: swipe up any time, and what the assistant is for. Bank-grade and encrypted, available any hour, in plain English or Arabic.

AI Assistant onboarding teaching the swipe-up gesture
Onboarding teaches the hidden swipe and sets expectations: encrypted, always on, English or Arabic.

How the assistant behaves

A wireframe shows screens. It does not say how the thing behaves. So I mapped the whole assistant: the ways in, the loop it runs every time, and how one pattern grows into new domains.

Ways in
Floating AI buttonSwipe-up gestureTransactions banner
The loop, every time
Prompt or tap a suggestionPlain answerFollow-up suggestionsClarify: "what do you mean?"Verify against CBD records
Grows into
Spending · liveInvest · nextOne pattern, more domains

Track your spending, prompt by prompt

I unpacked each starting prompt into what it returns, where the conversation can go next, and how it fails safely.

PromptWhat it returnsFollow-upsEdge & error states
Summarise my spendingTotal for the period, then the top 5 categories with amounts.Break down by sub-category · Detailed breakdown by top merchants · Clarify a termNo spend in the period · Ongoing month is flagged as partial · Unknown term routes to a plain-language explainer
Top categories I spend onCategories ranked by amount, with each one's share.Open a category · Compare to last month · See the merchants insideFewer than five categories · The uncategorised bucket stays visible, not hidden
Top merchants I shop withMerchants ranked by spend, with logos.See this merchant's transactions · Spending trend · Set an alertUnknown merchant kept visible with a fallback name
Show recurring paymentsRecurring charges with amount and next date.Flag one to review · Remind me before the next chargeNone detected yet · Needs more history to be sure
Food & grocery spendCategory total plus the trend against last month.Break down by merchant · Set a budgetNo matching spend · Data still syncing

Every answer runs the same states

Prompts shownStreaming · Stop availableAnswerFollow-up or type your own
Branches: Stop keeps the partial answer · Error shows "could not load, retry" · No data shows an empty state with a nudge · Off-topic is redirected to what the assistant can do

Built to scale: a second generation

The system was made to hold more than spending. The next release extends the same pattern to Invest: portfolio performance, top and weak performers, crypto, with the same conversational shape and the same guardrail, verify against your own records.

Second generation of the AI Assistant for investing
The second generation: the same assistant, now for investing.

A few decisions behind it

Screens are from the CBD AI Assistant, built on Dyna. The persona and figures shown are anonymized.