Skip to main content

Fluid Artificial Intelligence (AI)

Fluid offers two AI-powered features that can be utilized independently or in combination:

  • Predictions for Suggested Amounts
  • A/B Testing

Any of these features can be enabled from the Settings page in the back office by toggling their respective switches.

Note: Before enabling any of the AI features historical data is required to perform Machine Learning (ML) and/or Data Analytics (DA).

Predictions for Suggested Amounts​

This feature predicts and displays tailored deposit amounts, including default and alternative suggestions, based on historical data and customer behavior.

The following parameters are necessary for generating predictions:

  • Country (ML and DA)
  • Currency (ML and DA)
  • Previous Deposit (ML)

In cases where parameters are unavailable or the data volume is insufficient, pre-set default amounts configured in the back office will be used.

AI Model Configuration​

When enabled, a dropdown menu for the AI model appears, providing three options:

  • Low: Suggested amounts align closely with predictions.
  • Medium: Suggested amounts are moderately higher than predictions.
  • High: Suggested amounts are significantly higher than predictions.

A/B Testing​

This feature facilitates testing of various AI models for Predictions for Suggested Amounts, enabling data-driven decisions on model selection.

Setup​

  • Divide the traffic split among the Low, Medium, High or No AI models to see which performs best.

  • To do this, specify the proportions of the desired split between the different models.

    • Assign a number for each AI model.
    • Use the same values if an equal distribution is desired.

  • Scenario 1 (using percentages - values add up to 100):

    • Low - 25
    • Medium - 25
    • High - 25
    • No AI - 25

    Screenshot of A/B Testing

  • Scenario 2 (not using percentages):

    • Low - 4
    • Medium - 6
    • High - 5
    • No AI - 5

Split Adjustment​

Whatever values are specified, Fluid will make the necessary adjustments to calculate the correct proportions to 100% of the traffic.

For instance, using scenario 2 as an example, our system would calculate the sum of the values as 20 and adjust the split for each model as follows:

  • Low - 20% (4/20)
  • Medium - 30% (6/20)
  • High - 25% (5/20)
  • No AI - 25% (5/20)

Recommendations​

  • Percentages should be split evenly between the selected AI models for balanced testing.
  • A/B Testing should remain active for at least four weeks without modifications to collect sufficient data for analysis.

Important: Enabling A/B Testing overrides any active Predictions for Suggested Amounts configuration.