"smart banking- real time driven at number26", christian rebernik, cto at number 26 gmbh
TRANSCRIPT
NUMBER26
1 bank account
1 MasterCard
1 app
Our experience is designed for real time mobile first experience
11
12
Account opening
Open your bank account online in less than 8 minutes
… every day from 8am - midnight
… including verification checks, risk scoring, whitelist/blacklist
14
Intelligent push notifications inform you about
… successful card & bank transactions
… fraud attempts
… insufficient funds or exceeded limits
Transparency
15
Block your card & report it as stolen
Set whether you want to pay abroad
… or online
… allow cash withdrawal or not
Real Time Control
NUMBER26
1 bank account
1 MasterCard
1 app
Our experience is designed for real time mobile first experience
18
19
Account opening
Open your bank account online in less than 8 minutes
… every day from 8am - midnight
… including verification checks, risk scoring, whitelist/blacklist
21
Intelligent push notifications inform you about
… successful card & bank transactions
… fraud attempts
… insufficient funds or exceeded limits
Transparency
22
Block your card & report it as stolen
Set whether you want to pay abroad
… or online
… allow cash withdrawal or not
Real Time Control
25
Identify transactions which should be linked
Clustering: Too many false positives
Vector Space Modelling allows better grouping into logical groups
- represent each transaction as feature vector (d1, d2, q)- Find the closest vector (lesser angle between two vectors) for
a query transaction vector (q) in terms of cosine similarity- Similarity score quantifies the likelihood of linking multiple
transactions together
27
Linking Transactions
Linking transactions helps you to keep track of your finance
E.g. you order 10 items and return 9 of them
How can you make sure you received a refund for all of them?
28
Simplifying the financial transaction overview
You’ll see all related transactions in just one screen
… making it easy to follow up
… and to understand your spendings and income
Linking transactions architecture
Async Inflow
functions for real time linking
grouped transactions
29
Event based trigger
Where did I spend all this money?
Person To Person(Money Beam)
Person To Merchant (Bank Transaction)
31
32
Simplifying the financial transaction overview
Grouping your expenses automatically into useful spending clusters
… making it easy to analyze
… and to understand your spendings and income
To PersonUse reference text
State of the art NLP techniques to parse, tokenize, lemmatise texts to extract sense.
Semantic sense leads to smart categorization
Based on a lexical database and translation engines, it is possible to find near exact sense from the texts.
Examples:“Yufka for 2” => Food“Nudeln” => Food“debit from a doctor” => Health“Babyoel” => Children“Taxifahrt” => Cars 33
To MerchantMerchants play a vital role. If you spend money at “Vapiano”, we know instantly.How can you teach a machine to categorized it as “Bar and Restaurant”?
Using the Machine Learning- Learn Model on card transactions- Train model- Test model on bank transactions
Tested methods to determine the transaction category on F-1 score
- Naive Bayes 0.57- Supported Vector Machines 0.80 - Multiclass logistic regression: 0.9
Best Result: multiclass logistic regression
34
Linking and categorizing transactions
Async Inflow
machine learning for categorizations
functions for real time linking
grouped and categorized transactions
35
Event based trigger
38
Our goals
Innovation leaderby user focus.
Building a Smart Fintech-hub
Expansion to become the firstpan-European bank
70 people - 20+ nationalities
39
Top 25 global payment companies -CB Insights Christian [email protected]
Top 100 hottest European startups 2015 -Wired
Top 25 hottest German startups -Gründerszene
Soundcloud, Bitcoin, Google, Rocket Internet, Zalando, Zanox, Bwin