Explainable AI

Explainable AI has become a popular topic due to an explosion in AI applications and an increased focus from regulators requiring some sort of interpretation around AI results, which have historically largely been black boxes. Explainable AI is particularly important for financial services and other highly regulated industries or for applications that need interpretation of inputs/features. Explainable AI is one of the areas we specalise in at Advantage Data.

What is an explanation?

An explanation offers insights into individual predictions made by a complex machine learning system. The explanation enables a decision maker and other stakeholders to understand, scrutinise, and presumably trust an otherwise un-explainable or ‘black box’ model. These methods can provide insight into the question “What were the make or break factors for the model to make that prediction?”

Why explain?

  • Produce granular insights and feature importances relevant to individual predictions in real time, for any model.
  • Establish trust with stakeholders, especially front-line decision makers.
  • Implement highly accurate and complex models that are interpretable. Disrupt the traditional accuracy vs interpretability tradeoff that have held back complex and accurate models from use in production.
  • Enable automated storytelling of machine learning predictions which can give analytical and business context to predictions.
  • Compare and validate internal models and perform due diligence on external models (e.g. proprietary vendor models)
  • Identify “algorithmic bias” that may be otherwise be undetected.

What we offer

  • Build a real time ‘explainer’ to assist stakeholders with model interpretation
  • Analyse existing or proposed models for explanations consistent with legal and business rules
  • Stress-test high-risk or consequential automated decisions retrospectively or in real-time

Use cases

Sales forecasts

A retailer uses a model to produce sales forecasts for its chain of stores for the next six months. XAI can provide insight beyond the forecasting confidence interval and indicate the reasons behind the forecasted sales levels unique to each store in its franchise.


An insurance provider predicts a set of customers will probably churn in the next month. XAI can provide insight into the reasons why the model predicts certain customers are likely to churn. Different reasons could provide the insurance with an opportunity to segment and target customers based on the explanations and personalise retention offers to maximise the likelihood of retaining the customer relationship.