How moving from manual to machine learning can ensure operational resiliency

 

By Christian Nentwich, CEO at Duco

Operational resiliency has – quite rightly – become one of the main concerns for financial institutions across the world.  With almost every industry facing major disruption over the past few months, the banking and financial sector has been forced to make significant operational changes – that would usually have taken months or even years to deploy – in the space of weeks or days.

The need for resilient, connected systems that can manage huge swathes of data from multiple, disparate sources quickly and accurately has never been greater. With the UK’s Financial Conduct Authority announcing it is conducting a survey into the post-Covid-19 resiliency of firms, how the sector responds to this crisis in the long term will be crucial.

Many firms will also be under pressure to ensure any new processes introduced at short notice – and potentially without the usual thorough due diligence – are fully compliant when it comes to filing annual audits at the end of the year.

With business continuity and regulatory compliance front of mind, many are asking how they can ensure their most critical processes are automated, and not subject to manual work or spreadsheet-based controls.

Reconciliation is one of these critical areas, helping to eliminate operational risk that can lead to fraud, fines, or in the worst case, the failure of a firm. And yet, even pre-pandemic, fully automating this essential function was proving elusive for many financial services organisations. Why?

Many firms are facing a situation where they have deployed a multitude of systems, processes, technology types and computing.  Within that, there are three key reasons that make automation difficult:

• A lack of standardisation – In many cases across the wider financial services sector, there are no strict data standards. For example, different counterparties provide trade and position data in different formats. Each one requires a bespoke reconciliation process or expensive data normalisation.
• Increased complexity – Cash or stock assets can be matched on a few basic fields, but for more complex products you need to take far more information into account. Most current systems are unable to deal with every asset type that crops up in a timely manner. And, that’s before we get to the range of data needed for regulatory reporting, and the associated reconciliations required.
• Poor data quality – The enemy of automation. Missing fields, inconsistent coding schemes and unavailability of common keys make automation difficult when using current solutions due to hardcoded assumptions within those systems.

So, in a world where the quantity and complexity of data that firms need to handle is set to increase exponentially, relying on manual systems and processes is no longer feasible – and makes little operational sense. So, how do firms deal with this influx of data in the most intelligent way, while ensuring long term resiliency?

The ‘Reconciliation Maturity Model’, is a new roadmap that will help financial firms improve the automation, efficiency and integrity of data across all reconciliation and data matching tasks.  The model guides reconciliation practitioners through five key stages of reconciliation maturity, from ‘manual’ through to ‘automated’ and eventually ‘self-optimising’ – where machine-learning technology automates nearly the entire process, and where intersystem reconciliations are all but eliminated.

Importantly, a more progressive approach to reconciliation automation will not only result in greater operational efficiency, it will also dramatically boost operational resilience, and put forward-thinking financial institutions in a better position to benefit from new technology and the added insight that intelligent systems bring.

The five stages of reconciliation maturity are:

1. Manual – By this we mean using Excel or some other form of spreadsheet, macros, home-grown applications or – in some instances we’ve come across – printing out sheets of paper and marking inconsistencies with a highlighter pen! However, as the organisation grows, and the data becomes more complex, the risk of error skyrockets. There’s no audit trail, no governance and it becomes increasingly expensive to scale. If in the 2020s you’re throwing an increasing number of bodies at a data matching exercise, you know something’s wrong.
2.  Hybrid – For the majority of organisations, this takes the form of a point solution, usually deployed to automate high volume, low complexity reconciliations such as cash or custody. These point solutions – by their very nature – tend to specialise in a certain type of reconciliation. Firms trading a wide range of assets, or those dealing with complex data, may need to use multiple point solutions to handle different reconciliation types. However, there will be many reconciliations that these point solutions are not able to handle elegantly. In these cases, firms tend to fall back on manual processes. The result is a patchwork quilt of different reconciliation approaches stitched together by manual work. The whole process is costly, difficult to keep track of, and difficult to scale.
3.  Automated – All reconciliations are consolidated onto automated systems, and small teams build and onboard reconciliations, and oversee exception investigation.The key to getting to this stage is using the right technology. To reach Stage 3, firms need to be able to onboard reconciliations in hours or days, not weeks or months. They need to be able to rely on agile, flexible technology that can deal with complexity without multi-week data transformation projects. Once this technology is in place, complexity and risk can be vastly reduced, while increasing efficiency and transparency across processes.
4. Improving – This enables greater efficiency and oversight of the reconciliation function as a whole. It also enables firms to normalise their data across the business and implement additional data quality checks across systems, highlighting areas of incomplete or incorrect data.  Organisations are then able to start consolidating systems and removing duplicate reconciliations which have already been handled upstream.  Processes become leaner, more efficient and more transparent.
5. Self-optimising – Full automation is deployed across the entire lifecycle of reconciliation, from onboarding to exception resolution. There is very little involvement from staff and continuous improvement is possible via a machine-learning enhanced system. Internal reconciliations are removed, leading to major reduction in cost and complexity.

While stage five is the ‘holy grail’ that all financial organisations should be aspiring to, many firms are at the ‘hybrid’ stage, and making the leap to ‘automated’ is the most challenging step.  However, once at stage three, firms are more able to move up the process to stage five – ‘self-optimising’.

At this point, with enough training data, machine learning – when implemented properly – will enable firms to spot and correct data errors, inconsistencies and poor quality at source, before issues are created in downstream systems. Internal reconciliations are often used as ‘after-the-fact’ control points in many financial institutions, but if data is fixed as it enters the organisation – by using machine learning technology that has trained on past data – these reconciliations will start to flag up fewer and fewer issues, and can be eventually be removed entirely.

So, while we know that moving from manual to machine learning is not an overnight process, it is a vital one if firms are to ensure operational resilience beyond Covid-19.  The Reconciliation Maturity Model provides a blueprint to getting there.