Enterprise AI

Organizations have run AI projects for several years now. But they are mostly isolated examples in the areas of marketing and risk management. These projects were typically limited by data and compute constraints and are nowhere near to the large, sophisticated AI engines that power the fintechs and bigtechs today. Banks can no longer afford to remain at the level of piecemeal projects, at least if they hope to stay relevant and compete against the digital natives in the coming years.

It is essential that banks today invest in enterprise AI with the following clear capabilities to support a wide range of use cases across products and services, customer engagement and operations:

1. An enterprise AI platform that can support some of the latest in AI techniques, all the way to deep learning. More importantly, the platform should have the ability to incorporate the latest advances that are coming out of academia, open-sourced from FAANGs, etc.
2. AI is compute-resource hungry, and nothing can be more frustrating than a constraint on resources. To deliver on scale and performance, it is inevitable that the enterprise AI platform should be cloud based.
3. It is always worth remembering that AI is not for AI’s sake – the goal is to deliver business outcomes. And for that to happen, there needs to be a clear way of taking AI recommendations and using them to drive business processes. This requires the enterprise AI platform to:
   • Operationalize the AI models – starting with the data pipelines, training, deployment of models, and ongoing monitoring of the models.
   • Integrate the model outputs with operational systems to integrate seamlessly into the overall decision-making process.
4. There needs to be recognition of the fact that AI has its roots in the open-source industry, from shared datasets to shared algorithms. A robust AI ecosystem in the enterprise must be able to replicate this democratization across functional silos. That is critical to enable a culture of reuse and sharing of best practices; and perhaps, more altruistically, to build and foster a learning ecosystem with the enterprise data science community.
5. And last but not least, the need for strong AI governance cannot be understated. One of the biggest challenges of AI relates to its ethics. Enterprise AI needs to take ownership of the following ethical dimensions:
   • Data privacy and security: Access to data must be balanced with the enterprise security guidelines, especially when it comes to customer-sensitive data.
   • Algorithm bias: Consumer banking is especially sensitive to this – particularly when it has material implications for specific population segments. There needs to be constant monitoring of algorithms to guard against bias.