To provide a bit of context to this, Objectway, as a provider who is already striving to bring improved operational efficiency to its client base, has already been involved in introducing robotic process automation, as well as intelligent process design into its back-office systems, so it could be argued, through using these techniques that we have already made significant advances. However, the rapid improvement in the scope and accuracy of computer vision (advanced OCR) as well as NLP techniques to properly label, classify, and assimilate into business process more complex alpha numeric data sets within unstructured document types is now opening up opportunities to integrate lo code process building with these techniques.

Use Cases

When I consider how we can put these advances in accessible technologies to best use, the most important considerations for building use cases are often to consider situations where middle office and back office staff face challenges to maintain data integrity and consistency within a specific process, and thus often are placed in escalating exception handling scenarios. (We know that these are often the most stressful and time sensitive scenarios and thus the ones to be avoided).

To give an example of this, in the suitability process, accurate product governance, approved and visible to risk and compliance, is an increasingly important aspect of the way that different clients are able to receive and maintain personalized portfolios from the front office and access to the best solutions for their relationship managers. While all firms, including Objectway, like to think that data vendors who provide services to build and maintain securities master for back office and investment management applications, for example, would be able, through their own data management expertise, perform an accurate job in either a single or aggregated way, the reality is sometimes in key areas of product suitability, as they relate to exposure to complex instruments, the components of the total expense ratio, or a variety of corporate action events, the data for consumption to support a transactional or ledger process, can be stale, inaccurate or altogether absent.

In these situations, the ideal scenario for the back office operations is to ingest, reconcile and process additional information, but since this is often embedded into documents such as Factsheets, Pripps, Kiids, and other fund specific documents, consistent and scalable processing capability has not been available. This situation is now being rectified through the combination of more data vendors building and maintaining in partnership with manufacturers, document hubs, and back-office technology firms, deploying computer vision and NLP to read and extract data from them, creating, in the process, the opportunity to greatly improve product governance, giving both compliance as well as business, a win.

This same combination of technologies, it turns out, can also be trained and deployed to facilitate much more interoperability between systems, particularly when there is the need, within a business process for both data translation and data normalization to a specific target protocol. This means extending data management capabilities into more and more unstructured data scenarios without having to force different actors into using a particular, and often limiting, data template. Since this approach can limit access to building data insights, firms have often rely on a combination of manual data mapping and reconciliation, alongside ETL technologies for performing these tasks, but now, with be deploying with data science, operational and domain support, a combination of NLP, LLM, and Computer vision, there is now the real opportunity that intermediary solutions can be built that not only intelligently recognize, extract and transform automatically, but also continuously ingest and harmonize systems to documented protocol upgrades and extensions, regardless of the initial vs. target state. This could be particularly transformative in situations where either a marketplace lacks an established provider performing this task, or when the complex nature of the products and counterparties has foiled attempts at standardization procedures needed to leverage smart contracts and blockchain technology.

It is by the way now not too hard to imagine that the task of identifying and programming the change across different environments in its own right, as an added benefit for this type of AI system, for adoption in back office solutions. Since many of these applications were first imagined 20+ years ago, at their foundation, they are still based on legacy programming language. Open sourced and Commercial translators are becoming available as a means to some modernization, but the resources available that understand what to do are rapidly dwindling or disinterested. Given these circumstances, I am increasingly of the opinion that properly trained AI solutions using the aforementioned models, will make increasing sense as the way to tackle this particular set of problems.

The Role of LLMs

If one looks at the investment trend in AI, there has always been an underlying belief that complex back office, administrative and record keeping systems, such as those present in healthcare, resource planning, and financial services, will invest in all of these technologies separately, and in combination, so perhaps what I am pointing out is not new nor original. However, it is fair to say that the operational benefits that these bring do not always get a lot of focus, and thus are often, weaker points of investment than they should be. This might change however when the use of LLMs also becomes a reality in this part of the value chain.

The use cases emerging here relate, not so much to the creation of automated intelligent and evolutionary business process management within the middle and back office, but to the introduction of intelligent knowledge resources for individuals who sit in both operational and development roles in this part of the organization. Firms have spent, some forced, and some voluntary, on creating a much richer ecosystem of tools for communication, collaboration, retention, and storage, but very few have had the discipline in their overall lifecycle management process to turn them from an informational repository to a reusable knowledge asset. As a result, while the landscape is littered with many snippets of insight and value, the cohesion is lacking even when deploying elastic search.

The good news is that Generative AI, and its means of bringing new precision to natural language search has the capabilities with limited supervision to overcome this. Business leadership might wonder about the immediate ROI on this might be, but CIO shouldn’t worry, as LLM used in this way will preserve access to proper IP, reduce subpar reprogramming effort, and ensure that artefacts become more accessible, more quickly, within the scope of every project, and every sprint. This is particularly valuable for the most complex parts of the system, which inevitably sit in back office solutions trying to efficiently manage process, logic, and data in a resource optimal way.

In this blog I have tried to highlight that the back office is a rich realm for highly valuable and reward use cases for a combination of machine learning and artificial intelligence models. I have done my best to highlight that the benefits to be gained by deploying these systems to the middle and back office areas of risk, compliance, and operations, is not only one with internal efficiency benefits in their own right, but also to creating intelligent automation that will deliver better client outcomes while reducing firm risk of non-compliance. While there are some dev/ops AI firms being funded to try to work with larger organizations on tackling this, for those with the right competencies in middle and back office technologies, it looks like the increasing availability of licensable tools, will prove sufficiently robust in the right user hands, and product management support, to achieve results.