Use cases

Use Upgini for Risk Management

Credit scoring
by Phone number, IP and Email

Extract powerful LLM features with GPT for your credit scoring

Personal Loan
Pre-Approval

Credit scoring

Credit scoring

Decrease NPL
Increase Sales
Proven commercial effect
Why does it matter?
Upgini has over 200 connected data sources of behavioral data. This data could be used to improve your credit risk scoring models. Especially in thing credit history segment. The average uplift to models with Credit Bureau data is from +3 to +7 percentage points of Gini.
How could Upgini help you?
Upgini could enrich your DWH with a rich customer profile by Phone number, Email, IP address. Use this data to improve your Credit scoring.
How to improve credit scoring with Upgini
  1. Obtain a train dataset with customer phone number, postal_code, IP-address, email and target label.
  2. Search for new relevant features for your task: launch FIT method of Upgini and get a list of relevant features from connected data sources.
  3. Make a test enrichment: launch TRANSFORM method of Upgini and enrich any dataset with only new relevant features that was found on FIT step.
  4. Use Upgini in production: Upgini will transfer relevant features for the all Phone numbers, Emails, Postal_codes and IP-addresses to your DWH. You may use this data directly in your DWH.
Extract powerful LLM features with GPT for your credit scoring

Extract powerful LLM features with GPT for your credit scoring

+20% better accuracy than other NLP libraries
+5% of Gini to your credit risk models
Use case description
It is customary in fintech risk management to use external data pertaining to an individual or their household, which is typically requested via SSN, passport number, last name, first name, and telephone number. However, during the loan application process, a wealth of additional data is generated, including the names of employers, the positions held by the client, the purpose of the loan, and other text fields. These text fields can also be utilized to search for external data. Upgini employs innovative technology to search for external data in text fields from the application, supplemented with external contextual data based on large language models such as GPT. This approach can enhance credit scoring models by up to 5 percentage points of Gini.
How to use Upgini for this task?
  1. Obtain a train dataset with customer ID, text fields, and target label. The customer IDs that Upgini supports for this task are phone number or email. Use manually labeled events such as "NPL 90@12" as target labels.
  2. Launch GPT feature generation in Upgini for your task: launch FIT method with parameter generate_features and you'll get a list of relevant features from connected data sources and relevant features extracted from conversation texts.
  3. Make a test enrichment: launch TRANSFORM method of Upgini and enrich your dataset with only new relevant features.
  4. Use Upgini in production: Upgini provides relevant GPT features through online API
Try Upgini to improve your credit risk model with GPT features
Just copy the code and send it to your analyst to repeat!

Upgini will quickly generate a quality report.Ask your analyst to send it to you to check the result.

Copied
%pip install -Uq upgini

# read labeled data
df = pd.read_csv("train_dataset.csv")

from upgini import FeaturesEnricher, SearchKey

# map data
enricher = FeaturesEnricher(
    search_keys={'Date': SearchKey.DATE,
                'phone_number': SearchKey.IP},
	generate_features=['text1', 'text2']
)

# launch fit step
enricher.fit(df[['Date','phone_number', 'text1', 'text2']], 
             df['npl_90_12'])

# launch enrichment step
df_enriched = enricher.transform(df)
Personal LoanPre-Approval

Personal Loan
Pre-Approval

Increase sales
Decrease NPL
Proven commercial effect
Why does it matter?
Banks often offer pre-approved loans to customers with a good credit score and regular repayment history. These offers can be an effective way for banks to boost their business, as they can prompt customers to take out loans they might not otherwise have considered. The ease and speed of access to funds, coupled with attractive interest rates and benefits, make these offers appealing to customers
How could Upgini help you?
Upgini can enrich your DWH with a comprehensive customer profile using phone numbers, emails, and IP addresses, which could help you calculate the probability of approval more accurately.
How to use Upgini for this task?
  1. Obtain a train dataset with customer phone number, postal_code, IP-address, email and target label (whether the application was approved or not).
  2. Search for new relevant features for your task: launch FIT method of Upgini and get a list of relevant features from connected data sources.
  3. Make a test enrichment: launch TRANSFORM method of Upgini and enrich any dataset with only new relevant features that was found on FIT step.
  4. Use Upgini in production: Upgini will transfer relevant features for the all Phone numbers, Emails, Postal_codes and IP-addresses to your DWH. You may use this data directly in your DWH.

Make ML&AI models happier
with more relevant data

No credit card required. No time limit on Free plan

Get started
Use cases
Fraud preventionMarketing and CRMDemand forecastProduct managementPersonalizationRisk managementLead scoring
Resources
DocsTutorials & Publications
PricingSell your data
for AI & ML applications
aicpa soc
SOC 2
© 2022-2024 Upgini  ·  Designed in UAE, build worldwide