Applying machine learning to reconciliation in accounting will prepare firms for a machine-first future.
Success in financial services primarily relies on the speed and accuracy of data normalization and reconciliation. Banks and their custodians, brokers, and portfolio managers, along with market utilities such as the end users of products like insurance, loans, etc., are all fundamentally data-driven with regard to strategic decision-making. As such, this requires their employees to input, match, compute, and report data spread across various business units daily. This is where the need for machine learning (ML) arises, as more the manual work, the higher the chances of operational and oversight errors.
Banking, financial services, and insurance (BFSI) organizations can now experience higher levels of accuracy and quality delivery with ML.
We explore why automation and ML are fundamental to reconciliation, a key accounting process. We also outline a reconciliation maturity model (RMM) as a solution to address the complexities involved in this essential process.
Why is automation necessary?
Reconciliation is a mission-critical process that eliminates operational risk, identifies potential fraudulent activities, and reduces errors that lead to fines and penalties.
To rationalize their reconciliation functions, a reconciliation maturity model (RMM) allows banks to automate data import and export, transaction-level manual matching, report generation, and more. This translates to negligible user intervention. Before delving into the functions of RMM, let us understand the various complexities of the reconciliation process.
Why is reconciliation so complex?
Multiple pinpoint solutions in reconciliation often makes it challenging.
As a result, diverse processes are sewn together through manual sheets or spreadsheets that are highly inefficient and prone to miscalculations and errors. Furthermore, a lack of standardization, increasing complexity, poor data quality, and data inconsistency (see Figure 1) magnify the complications in the reconciliation process.
Figure 1: The complexities of reconciliation
An increase in the operational costs and the headcount or volume of transactions goes hand in hand with the level of complexity of processes. Because of inadequate reconciliation, businesses are left with no way of tracking payments, detecting fraud, or identifying potential disruptions to their cashflow. Further, inadequate reconciliation invites regulatory fines, financial and reputational loss, and in the worst case, failure of the entire business.
Firms must overcome these shortcomings and switch from manual processes to automation and optimize the complete reconciliation function to derive additional value from the data being scrutinized. Once the data is streamlined and systematized into a single system, machine learning (ML) techniques are used to revamp every step involved in the process.
With adequate training data, ML can identify errors and irregularities in daily functions, thus reducing the number of reconciliations required. All financial organizations must strive to make this mechanism harmonious so that the inter-system reconciliations become irrelevant. RMM addresses these irregularities.
Reconciliation maturity - Enabling efficiency
At its core, RMM enables financial firms to consolidate, automate, and drive efficiency across reconciliations.
It operates in five different stages (see Figure 2).
Figure 2: The stages of RMM
Stage 1 – Manual reconciliation: Reconciling data manually through spreadsheets, macros, or in-house applications results in inconsistencies on the physical printout sheets. Despite being more affordable and quicker to set up, manual process lead to more oversight errors, which increases as the data becomes more complex. With negligible audit trail and governance, it is progressively expensive to carry out manual operations at scale. The higher the data complexity, the more the dependence on manual labor, as macros become opaque and obsolete.
Stage 2 – Blended reconciliation: As organizations reach a certain size or start handling massive amounts of data, automated reconciliation becomes a necessity. This takes the form of blended or point solutions that are mainly deployed to handle higher volume and reduce the complexity of reconciliations.
Manual mechanisms and spreadsheet calculations can result in poor coordination among users involved in processing and accounting operations. Moreover, there is an increased probability of duplication, as the data is often divided into fragments. When point solutions prove inept in handling certain reconciliation tasks smoothly, organizations tend to resort to conventional manual reconciliation. The result is a mix of various reconciliation approaches manually weaved together, and the whole process becomes expensive and difficult to track.
Given the expensive and complex nature of data tracking, most organizations centralize their functions in low-cost locations, onboarding more people at a reduced cost. Large enterprises utilizing this model will require hiring fewer people to pick up manual work.
Stage 3 – System-driven reconciliation: As organizations continue to grow, they tend to progress naturally through stages 1 and 2. However, reaching Stage 3 is not an ordinary leap and requires automation of all the reconciliations. If the fundamental rethinking and restructuring of an organization’s reconciliation operations are executed accurately, the advancement to stages 4 and 5 is not only smooth but also inevitable.
Using the right technology, instead of exclusive point solutions, is the key to reaching Stage 3. To advance to this stage, organizations are required to get their reconciliations operational within a few hours or days. Agile and adaptable technologies handle all complexities efficiently while enabling faster process transformation. Once the technology becomes operational, the risk and complexity are significantly reduced, whereas transparency and efficiency across the process tend to increase.
At Stage 3, enterprises use a mix of legacy technologies, along with flexible reconciliation solutions to handle complex data. This involves a web of automated systems where operators are divided into teams with separate individuals performing the reconciliations and control processes. Thus, there is higher efficiency with a low risk of errors and data duplication.
Stage 4 – Improvised reconciliations: At this stage, all reconciliations are automated into a single intelligent solution, making the processes more efficient, leaner, and simpler. The solution also empowers organizations to normalize data across different business functions and implements supplementary quality checks for highlighting anomalies and incomplete data. Firms continue to consolidate different systems and eradicate dual reconciliations that have already been taken care of in the upstream processes. Because of the higher degree of automation, reconciliation data is compiled in one place to reduce the time spent on manual tasks and improve team productivity. Teams find it easy to run analytics, identify pain areas, and provide feedback to the business.
The ease of switching from previous stages to Stage 4 relies mainly on legacy solutions. For new organizations, it will be easier to advance to Stage 4 just by migrating their reconciliations to an agile system. This stage employs end-to-end advanced data integrity algorithms across all stages of work. The biggest hindrance of such a system is the performance or interpretation of such monitored processes, leading to discrepancy and function failure.
Stage 5 – Fine-tuned reconciliations: Given the rapid evolution of ML, self-optimization in reconciliations seems to be a tangible possibility. This stage involves automating the entire reconciliation process. It covers all operational functions such as matching, onboarding, exception management, configuration, and management reporting. The increase in the daily transactional volume facilitates improvement and optimization of the resources. As new data enters an organization, the fully automated system identifies inconsistencies before they can impact the downstream systems.
At Stage 5, the count of intersystem reconciliations reduces. The iterations are also eliminated entirely when the reconciliation solution flags and fixes any pending or unmatched item, providing complete penetration into the data passed through an organization. The reconciliation team becomes strictly confined to users who substantiate the propositions (the available options that clear or match the outstanding items) of the fully automated system. The team investigates anomalies in case a system is unable to identify and fix those by itself.
Stage 5 is fully automated with no requirement for maintenance personnel. ML mechanisms adapt to avoid anomalies and mature to maintain objectivity. Here, the primary focus is on eliminating errors with a significant reduction in the cost and complexity of daily business operations. When applied sensibly, ML techniques provide an opportunity to rectify data errors and also identify deviations at an early stage.
The future of data is now
As we enter a new decade, the RMM model will support firms in reaching Stage 5: fully automated account reconciliation
Firms have already reached this superlative state. They have implemented ML within their contemporary systems, which have brought advantages for users and it will only get better with time. Also, as more people are keen on utilizing these systems, the ML algorithms have become more refined, as they are fed more data.
We believe organizations must gear up to take advantage of ML and embrace a future where manual reconciliation is curtailed. An ideal way to do this is by positioning ML algorithms at stage 4 or 5. This way, firms can reap instant benefits both financially and operationally. This will prepare them for newer capabilities when they arrive.