Question 1

What is Revolut?

Accepted Answer

Revolut is a global financial super-app offering multi-currency accounts, payments, cards, savings, stock and crypto trading, and business banking. Founded in London in 2015, it serves tens of millions of customers across more than 100 countries and has grown into one of the world's most valuable fintechs.

Question 2

What is Revolut's PRAGMA model?

Accepted Answer

PRAGMA is Revolut's in-house family of transformer-based foundation models trained on roughly 24 billion banking events from 26 million anonymized user records spanning over 100 countries. Rather than licensing external AI, Revolut built PRAGMA to learn directly from its own transaction and behavioral data.

Question 3

Why did Revolut build its own AI foundation model?

Accepted Answer

Banking events are a unique data modality that general-purpose LLMs were never trained on. By building PRAGMA on its own raw data, Revolut gets a single underlying model that can be adapted to credit scoring, fraud detection, and personalization — with reported performance gains far beyond off-the-shelf approaches.

Question 4

What products does Revolut offer?

Accepted Answer

Revolut spans personal banking (accounts, debit cards, payments, currency exchange), wealth (stocks, commodities, crypto), credit products in select markets, and Revolut Business for companies. Premium subscription tiers add perks like travel insurance, higher limits, and metal cards.

Question 5

How significant is PRAGMA in the banking industry?

Accepted Answer

PRAGMA is notable as the only published foundation model of its scale built by a consumer neobank rather than a payments network, and the largest published encoder backbone for consumer banking event sequences. It signals a shift where financial institutions train proprietary models on transaction data instead of relying solely on external AI vendors.

Question 6

How does PRAGMA work?

Accepted Answer

PRAGMA is a family of encoder-style transformer models scaling from 10 million to 1 billion parameters. It fuses multiple event streams — card transactions, app navigation, trading activity, and push notification interactions — into a single user-level embedding that downstream models adapt for specific tasks.

Question 7

What results has Revolut reported from PRAGMA?

Accepted Answer

According to Revolut's published material, PRAGMA-based systems improved credit scoring performance by around 130% and lifted fraud recall by about 65% versus prior approaches. A single foundation model backbone serves multiple tasks including risk, fraud, and product recommendations.

Question 8

How does PRAGMA improve fraud detection?

Accepted Answer

Instead of scoring transactions in isolation, PRAGMA interprets each event in the context of a user's full behavioral history. A midnight payment reads very differently if it follows a burst of rapid transactions on an unfamiliar device in a new location — contextual patterns that sequence models capture and rule-based systems miss.

Question 9

How does Revolut's multi-currency account work?

Accepted Answer

A Revolut account holds and exchanges dozens of currencies at interbank-linked rates, with a single card that pays in the local currency wherever you spend. Transfers between Revolut users are instant and free, and international transfers are typically cheaper than traditional bank wires.

Question 10

How was PRAGMA trained while protecting user privacy?

Accepted Answer

Revolut states PRAGMA was trained on anonymized records from 26 million users across 111 countries, covering 24 billion events. The training data consists of behavioral event sequences rather than personally identifying content, in line with financial data protection requirements.

Revolut

Frequently Asked Questions

Where AI Leaders Stay Informed

Is this your company?

Where AI Leaders Stay Informed