The impact of Artificial Intelligence on Enterprise Risk Management
More and more businesses are starting to adopt Artificial Intelligence based solutions. Besides considering the apparent project risk that implementing such techniques introduce, enterprises need to consider what the overall impact will be on their Risk Profile after the launch. Many might have to change their risk oversight and mitigation approaches before implementing Artificial Intelligence techniques.
The terms Artificial Intelligence and Enterprise Risk Management are both used within various different contexts and frameworks, hence for the purpose of this article we’re going to assume the following meanings:
- Enterprise Risk Management (ERM) is the holistic approach to managing both the upside and downside of qualitative and quantitative risk in a manner that is appropriate to the organisation’s objectives.
- Artificial Intelligence (AI) is an umbrella term for new technologies such as process automation, natural language generators, voice and image recognition as well as new types of data science models. As far as actuaries are concerned, we will be referring to AI as any model that can at least recalibrate autonomously.
A traditional type model will give the same results when fed the same data. An AI model will use new data to reassess the underlying relationships and may recalibrate the relationship between the target variable and the potential explanatory variables. Depending on the level of autonomy, the AI model might even be able to adjust parameter values, assumptions and the functions used. Therefore, feeding the same data again, might result in a different output.
The purpose of ERM is to maximise risk-adjusted returns by taking into account all risks and their dependencies. It is focused on the upside and opportunities available as well as the risks. One way in which ERM can achieve this goal is by incorporating models into decision-making processes. The purpose of incorporating AI-type models is fully consistent with this goal and seeks to make decision-making processes even more efficient.
If we consider an ERM framework, we see that AI techniques (and technology in general) can assist in many of the underlying framework components. It can help identify new and hidden risks. It can more accurately measure risks. It can offer new ways of managing risks, and it can provide more holistic oversight when it comes to monitoring risk. In effect, adopting AI tools has the ability to increase the risk management efficiency of the business. However, the adoption of AI tools can also change the Risk Profile of the business, especially around Stakeholder, Model and Business Risk.
Incorporating sophisticated AI techniques into an organisation can change how a company interacts with its workforce. There is an important consideration around any AI tool about the target operating model (TOM) in which it is going to operate. Will the outputs of the AI based tool make redundant certain activities performed by humans? Alternatively, will the outputs advise an analyst about a course of action? An interesting case would be where the AI operates autonomously within certain parameters and refers cases that deviate from the norm to an analyst for consideration. The above considerations about the TOM are, strictly speaking, not about modelling but they must be considered as part of any AI based tool development. If they are not considered, stakeholder risks could increase disproportionately and have wider consequences in the roll out of AI based tools. Even if they are considered, the potential substitution of humans (depending on the TOM) means that in practice businesses need to prepare themselves for a degree of resistance from humans involved in the value chain. Staff may be reluctant to embrace these tools in fear for job losses; or customers may not engage with ‘technologies’ that substitute humans, like chatbots.
Besides introducing resistance to adopting these techniques, AI can also introduce model risk. Models have been wrong in the past, and AI models are not exempt from this risk. This risk may be intrinsically higher because AI model’s ability to recalibrate much more often than traditional models. The increased efficiency of the model (compared to traditional models) also requires more intense monitoring. This should be considered as part of the roll out of the AI model. While it is easy to think about model risk as a ‘techy’ one, it is important to bear in mind that it can affect business risks more widely. For example, model risk might mean that the business is taking a different type of customers with wider consequences for the risk profile of the business and, perhaps, not in line with the business strategy.In addition, while the AI model would have been trained (calibrated) before its roll out into the business, model risk would be high right after implementation as the values for the parameters are fine-tuned with real data. The performance of the AI model would need to be monitored closely.
Risk Profile and time
Let’s put the above in a time frame. Before AI, the company will have an average level of stakeholder risk and an average level of model risk. This can increase when an AI model is implemented. Stakeholder and Model risk and the wider need to monitor the performance of the model can also show important dependencies as a result of the roll out of AI tools. It is therefore important that that stakeholder and the underlying TOM considerations are an integral part of the AI development and not an after-thought. An effective communicating with employees and providing them with effective training, if there is likely to be displacement, would go a long way to mitigate these risks. These actions are crucial to ensure that in the long term, AI reduces the risk profile of a business.
Every competitive business should consider how it can embrace AI based models to power its business. AI has the potential of increasing efficiencies in the delivery of existing services and opening new opportunities that might not have tapped because of the limitations of traditional models or data. It is vital for Boards to ensure that the impact on risks of AI based tools are understood and managed appropriately.
Isaac Alfon (PhD), Managing Director, Crescendo Advisors (firstname.lastname@example.org)Michael Jordan (FASSA/CERA), Dupro Advisory