Gandalf as a Scoring Engine

Scoring Tables can help you to prevent fraud, score lending applications, calculate insurance risks
and in every other system, that requires quantitative representation for a transaction risk.

How it works?

1
Create a Scoring Table

Intuitive interface works just as you expect. Simply add some fields to your data model, and define rules that rely on it.

Our concept is simple:
- table have columns, each column is a field in your data model;
- each row in this table is a rule;
- each cell is a validator.

Whenever all validators are passed in a rule - we will add it’s Score to a final Decision, that is returned to a API consumer.

2
Generate sample request with Debugger

After creating table go to built-in debugger to try your results. Enter sample data in it’s form, submit and get your first Score!

3
Integrate your API consumers

Generate credentials on Setting page, call Gandalf from an external system by copy-pasting sample request from debugger to your CLI.

                        curl -u 17c13e411af461aa55b9801aaedfbea3863f6404:e15615d7e2fbd57206ba6a66abd0b3dadfd75c69 \
     -H 'X-Application: 576bf5f9ce3c0c02ee2d314d' \
     -d '{"ip":"183.183.210.01","merchant_distance":100,"country_code":"GB","card_brand":"visa","card_bin":"545700","card_issuer":"Barclays","turnover":1000,"turnover_month":1000,"payments_count_day":1,"turnover_ip":1000,"payment_amount":10}' \
     https://api.gndf.io/api/v1/tables/5745ce96f70466a2098b457c/decisions

                    

Whats next?

  • Analyze Scoring Tables

    Analytic tool will help you to see what rules are triggered more often than others. Find your balance between precision and recall.

  • Review Decisions History

    Gandalf will save all data on each decision it makes. Dive deeper in individual cases whenever you need.

  • Start a Split Test

    Use scientific method to improve your scoring engine. Create a hypothesis, add a variant that implements it and allocate some traffic to test it.

  • Try Machine Learning

    Consider investigation more modern Machine Learning approaches. Get familiar with Regression, Clustering and Anomaly Detection. Read more about Deep Learning, Neural Networks, Random Forests and Regression Analysis. TensorFlow is a good tool to look at.

Story behind Gandalf

We are product-oriented fintech team. Almost all projects we created needed a decision engine that reduces business risks. And all solutions we could find is hard to use and maintain, or very expensive. So we decided to create free open-source alternative, that is scalable, reliable and flexible enough to cover 95% of cases.
We decided to use open-source database MongoDB that have good scaling capabilities to make it suitable for Big Data.
Also we believe that vendor-lock is a bad thing, so we published all source code under a GPL license, feel free to change it as you wish.