AI solutions are nothing without quality data

November 30, 2020

Artificial Intelligence solutions are nothing without data, and banks certainly have a lot of it. The challenge is around quality rather than quantity, and an inability to identify and really harness the types of data needed to design, build and develop an effective AI solution.

There are several capabilities financial institutions can build that will serve the organisation well when executing AI projects, and a number of steps that can be taken within the project itself.

Firstly, banks need a centralised data hub. This should include a data model that rationalises data points across the organisation – transactional, personal, risk, financial, digital usage, product and external data sets. The hub can be built upon a cloud-based architecture that allows data to be identified, accessed and manipulated as a particular business area sees fit. We fully appreciate that this is no small task so it should be an ongoing exercise that lays the foundation not just for AI projects, but for nearly all change initiatives across the bank.

A hub for production data will allow the business to analyse how an AI solution can be applied and will provide historic data that will be used to train AI models. But crucially it is unlikely to be suitable for the early phases of an AI project due to data controls that need to be adhered to in development environments.

During development, the organisation will want high quality, robust test data that can be used without restrictions in sandbox and test environments. The capability to produce and harvest high quality test data can be purchased through a supplier or developed as a function to speed up the rate at which the business can quickly test their hypotheses in a low risk manner.

With the ability to easily source the right quality and quantity of both test and production data, each individual project then needs to consider how it will prepare and use that data for its specific use case.

Up front analysis of the problem at hand, the end goal and the available data is essential to developing and implementing an effective AI solution. Ultimately the data determines what the algorithm does.

To do this, project teams should analyse, prepare and constantly monitor the data for their AI solution.

Analyse: You need to fully scrutinise which data points will be required to prove the hypothesis and then build an accurate, scalable AI model. In a world of data proliferation, you also need to challenge yourself to identify the right data points. Consider the potential impact that your own inherent biases, or biases inherent within the data, will have on your selection and work to mitigate against that. Consider the bounds of data privacy and whether the data you select will be used in an ethical manner. Full analysis of these considerations at the start of the project will ensure that the end product – whether it is a new feature, process, service or other – will most closely meet its intended aim.

Prepare: Once the right data points have been selected you need to ensure that it is of the right quality. Review the data set in detail, identify the key features that will be used for the model and work to resolve any data quality issues at the source. Due to legalities around GDPR and data controls put in place within the organisation, consider the classification of your data and the masking techniques required to ensure no sensitive information is unnecessarily shared – particularly where a vendor solution is being used.

Constantly Monitor: During each phase and crucially once the model has been implemented, ensure its performance is constantly monitored – it cannot be assumed that once the model is in place it will run without issues. Over time the quality of data received by the model may reduce or changes to the data set could adversely affect the model’s output. These problems would most likely reside under the surface so the business owner of the solution should regularly review the data inputs and the model outputs to ensure it continues to meet the intended aim.

Read our whitepaper “Navigating AI Financial Services” for more information and for the Woodhurst Blueprint© for building Artificial Intelligence capabilities

Download this free insight to find out: