Balancing Innovation and Ethics Across the Analytics Lifecycle
The increased use of advanced data analytics techniques in organisations looking to innovate and gain a competitive advantage brings an increased risk of unintended negative impacts on people.
In order to balance innovation with ethical outcomes, organisations should ensure robust governance across the analytics lifecycle, including regular assessments of the project purpose, data, processes, models, people skills, and potential negative impacts.
An iterative, agile approach to analytics provides many opportunities to review the ethical validity of insights and outcomes at each phase of the project. The key to an ethical analytics project is to embed consideration of organisational purpose and values in each phase of the analytics lifecycle.
A multi-stage assessment of data ethics has the benefit of not having to identify every conceivable risk at the start of the project, allowing for more open discovery and testing that enables innovation. An evidence based approach to risk management supports mitigation of adverse impacts before implementation.
A Phased Approach
The first opportunity to embed ethics occurs during the scoping phase. Organisations need to clearly define the purpose, outcomes and benefits of the project, as this provides a frame of reference for assessing risk. The project purpose should be continuously re-evaluated to ensure it remains aligned with organisational values, is achievable and relevant. Scoping should identify potential negative impacts on people, especially if they are vulnerable, there is an imbalance of power, or lack of data ownership and control.
During the data acquisition and analysis phase, the key focus should be to assess whether data is fit for purpose to achieve the project benefits. This involves identification of gaps and limitations in the data, understanding data context, quality, lineage, and any data use restrictions. It is important to identify people or communities that are misrepresented, under-represented, or not represented in the data to mitigate the risks of deriving invalid insights from that data.
During the modelling phase, analytics teams should have the appropriate training, experience, and diversity to analyse and interpret the data, and to design and train models to mitigate potential adverse effects. This involves not only technical skills, but also an awareness of ethical issues, including the ability to recognise and address potential biases or unfairness. Extensive testing, feedback from diverse stakeholders, and iterative model development provides an opportunity to challenge assumptions and outputs, and understand the limitations of the model. Models should be documented so that they are transparent, explainable, reproducible, and auditable.
In the implementation phase there should be an examination of project outputs to ensure alignment with the outcomes and benefits. Additionally, once implemented an appropriate process should be established for the ongoing monitoring of the model’s operation, including relevant human oversight and governance over the impact of business decisions based upon model outputs.
Organisations can promote ethical outcomes by embedding regular checks across the analytics lifecycle. This ensures that analytics projects do not veer off course and result in unintended negative impacts. It starts with encouraging analytics teams to continuously challenge their assumptions, ask difficult questions and be attuned to the ethical issues across the analytics lifecycle.