Case Study

Middle Office AI


We found many asset managers were locked into rigid and aging middle office reconciliation systems. Because the majority of these systems are simple and don't learn, asset managers can't reduce labor costs.

Our work with several major, global asset managers presented several major challenges:

  • Rising interest rates have created urgency for more accurate predictions of cash and positions between IBOR and ABOR records.
  • Failed trades predictions. Most trades are predictably delayed, yet are treated as if they might be failed trades. In rarer cases, investment operations staff might assume a trade is delayed when it has actually failed.
  • Accurate cash predictions. Penalties for overdraft are now substantial given rising interest rates. It is vital that investment teams have a much more accurate prediction of their daily cash positions, given all other activities in flight.
  • Reducing ineffective reconciliations and labor. Many teams allow rules-based reconciliations to continue, often not understanding how they relate to one another. Some may be redundant, while others may be less effective at spotting root causes.
Diagram of a position passing through multiple books of record and the reconciliation process.
There is a transaction chain spanning multiple parties and books of record
Reconciliation systems generate too many false positives, keeping labor costs high in investment ops

Our middle office AO solution provides a platform to rapidly ingest any data from various electronic databases as well as from contracts and PDFs. As illustrated below, each asset class may flow through several machine learning "states." Our algorithm can track the position through each state - like a trip map - and help provide asset managers a more accurate views on their IBOR to ABOR.

We are developing algorithms trained to predict the root causes of different asset classes by following their transition states.
Middle office reconciliations must handle records from multiple systems and formats.
How It Works

We help setup a programmable pipeline of all data and document sources first. Then we assist in the development of more sophisticated reconciliations. These often start as incidetn detection checks, but can be integrated to feed more sophisticated predictive algorithms. Finally, we help rigorously test the algorithms. Once the algorithms and checks are operational, we compare them to current reconciliations to gauge the redundancies and help identify recs and labor costs that can be reduced.

Because our method accelerates the managing of different data sources and configuration of reconciliations, we are able to create very sophisticated algorithms such as incident detection checks which are setup to look for the same conditions that caused errors and delays in the past.


There are three major outcomes we have achieved in this financial reporting case.

  • Faster and more comprehensive data extraction and transformation. Our programmable pipeline can very rapidly connect to different data sources and documents.
  • More sophisticated reconciliations that are integrated with algorithms. We help design and configure incident detection checks, performance checks, and link them as inputs to more sophisticated algorithms.
  • Labor reduction resulting from checks and algorithms that lower exception rates. We have a significant track record of radically reducing false positive rates, which drive the volume of most exceptions. Exceptions (false positives) are strongly correlated with labor costs.

See a demo