Other parts of this series:
Neural networks, the next step-change in financial services artificial intelligence (AI) adoption, present considerable benefits to the industry by being able to synthesize vast amounts of complex data to make extremely precise decisions. But they present drawbacks too, for exactly the same reason. In financial services, explainability is just as important as accuracy. Neural network decision-making can be very difficult to explain.
As I mentioned in my previous blog post, firms that are too quick to adopt neural networks without accounting for the pitfalls could find themselves experiencing more of the drawbacks of this technology and less of the benefits.
Luckily, by taking a step back and focusing on process over technology, applying ethics as a matter of policy and being willing to trade (to some degree) accuracy for explainability, firms can avoid these pitfalls. The proper approach can help you decide if, when and where neural networks make sense for your business.
Six steps to effectively implement neural networks
The effective use of neural networks depends on establishing a solid foundation of appropriate IT infrastructure, sufficient processing power, and robust frameworks, controls, tools and ethical principles. Once these are in place, the following six steps can set you on your neural network AI journey:
- Have your training data ready. Neural networks require vast amounts of data. Data quality and volume issues should be addressed across the organization, with the chief data office (CDO) showing leadership in the proper curation and governance of a data lake that can provide deeper insights and comply with data protection regulations. Diversity and fairness should govern data sampling to help reduce bias, provide enhanced coverage and improve accuracy.
- Change your culture. As data can be used to create biases, it is key for humans to assume responsibility for the data. Data scientists should adopt and maintain a highly ethical and aware culture to protect against bias and discrimination.
- Keep experimenting. Finding the proper algorithm for the task is often a process of ongoing trial and error. Sometimes neural networks are not the appropriate tool for the job.
- Test and validate. Standard testing as well as cross-validation are critical to avoiding discrimination and having a model that does not discriminate and is returning accurate and meaningful results.
- Keep on top of documentation. Each step of the design process should be thoroughly documented, updated throughout the model’s lifecycle and accessible to all stakeholders.
- Create a center of excellence for AI. Ethical and compliant use of neural networks is essential to making sure they are used responsibly and do not create derogatory issues for the firm and its customers. Establishing a center of excellence provides sustainability in the ethical and productive use of AI, including neural networks.
Use this powerful tool with care
Neural networks are powerful tools for improving efficiency, enhancing decision-making accuracy, boosting revenue and improving the customer experience. But they have the power to be destructive too, if not approached with due care and consideration. By following the principles outlined here, firms should be in a much better position to apply neural networks to the appropriate processes in the appropriate manner with the appropriate safeguards. This is the path to realizing the full benefits from an exciting new technology.
For detailed information on neural networks and how to apply them in financial services, please see Accenture’s report: Neural Networks: The Next Step for Artificial Intelligence in Financial Services