Bank reconciliations are the most ubiquitous type of transactional matching reconciliations across all business types and typically require reconciliation between disparate systems (e.g. bank vs ERP). Small businesses that do all of their finances through one institution (bank or credit union) can accomplish matching through simple excel or embedded system tools.

However, mid-size and enterprise organizations have a much larger challenge with many account types, payment types, institutions, systems, time zones and payment complexities, plus vastly larger scale that make matching reconciliations far more difficult. This is especially true for companies that are growing rapidly through M&A, organic growth or international expansion.

If you are involved in the the bank reconciliation process, you've probably encountered one or more of the following:

• Duplicate entries - whether through human or system error.
• Different data formats between systems - date is one example where format varies widely across systems. Another is how credit card data is stored and presented.
• Date/Time discrepancies due to system time clocks, transaction timing, time zone differences, and local time cut-offs.
• Multiple transactions within a single invoice complicating reconciliation handling.
• Human generated data errors as innocuous as comma placement, inadvertent character spacing, abbreviations, or simple misspelling.
• Numerous people each using the same data fields slightly differently.

Traditional solutions to create more efficiencies in the bank reconciliation process include:

1. Data clean up at the point of entry or data cleansing during a downstream process.
2. Post-extract data transformation processes to force each data type to conform a rigid format.
3. Using Excel Macros or Visual Basic scripts to do some automation. While helpful for simple, stable, consistent data, it still requires significant human to run and cannot handle the introduction of any complexity.
4. More recently, Robotic Process Automation (RPA) can record and mimic specific actions normally accomplished by staff. While this can speed up an existing manual process, it does NOT improve the process or matching effectiveness as the process still relies on boolean rules (if-then statements, AND, OR, NOT, resulting in TRUE or FALSE outcomes.) that you currently use. It does what you are currently doing faster with or without your supervision which can helpful for very specific, rote tasks.
5. Some ERP systems contain a bank connector and reconciliation module, using Boolean business rules to find matches and separate reconciling items. While this process improvement can help efficiencies, these systems often only match 60-75% of transactions due to new/different/inconsistent data, still leaving the remainder to be matched manually by a human. Boolean-based systems are also time-consuming to implement and fragile to operate, necessitating extensive on-going administrative support and heavy expense for every new reconciliation use case.

Reconciliation Process Overview

The first step to automating any process is to clearly identify the steps and activities in the process in order to understand where steps can be omitted, improved or combined with other steps - whether that uses advance intelligence technologies or not. The following is a simplified version of the bank reconciliation process with areas of opportunity for automation by type of technology.

Best Use Technologies for Bank Reconciliation Automation

You don't have to use just one type of technology to solve the bank reconciliation challenge. In fact, using best of breed point solutions that fit into a workflow and solve the right problem in the right way is often a better approach than investing in one larger platform that claims to “do it all”.

If you are on an accounting or ERP system that has an effective process for capturing and extracting the right data sets for reconciliation, and it works for you, then you could continue to use it for that purpose. Robotic Process Automation might be a good addition to this process if there are numerous manual steps for managing the extraction and loading of the data on either side of the matching process. RPA is also helpful for tasks like routing documents for review and approvals.

Machine Learning is the right approach for large scale, complex data sets that require more than a simple matching of a couple of fields or if you are dealing with a large volume of transactions each month. ML is also very effective when you have many different bank reconciliations for matching or are growing rapidly and are adding new businesses, accounts and processes on a regular basis.

Why Machine Learning for Account Reconciliation?

The emergence of Artificial Intelligence - specifically Machine Learning - as an advanced analytical tool is the ideal approach to solve the matching reconciliation challenge because:

• Applying modern machine learning to account reconciliation, when done well, moves beyond rules to a sophisticated understanding of your data (including complex interactions across many fields, files, and time) and results in much higher accuracy and robustness. This allows for a greater degree of confidence in matching, creating fewer exceptions. However, if this is done as a typical ML project, it has significant up-front effort and cost, as well as ongoing support.
• An account reconciliation system with embedded machine learning can learn from feedback. It will observe what you match, what you deem is not a match and learn from those actions to use in the next round of reconciliations. It is a true automated learning platform vs the need to “fix” broken rules or call in a machine learning consultancy when performance degrades.
• An account reconciliation system with automated machine learning greatly simplifies setup and maintenance of the machine learning models. It eliminates most of the need to pre-process data, add classifications, or physically update the system every time the data changes. Implementations of new use cases can be done in a self service way and with significantly less time and effort.
• A cloud based, SaaS solution that is data source agnostic eliminates the need for IT to build an on-premise environment or integration into your ERP to configure the software and be successful.

Sigma IQ's platform has a state of the art matching engine that learns from feedback, is automated, and cloud based. As experts in account reconciliation, we have built in support for complex, multi-file, multi-source, many-to-1 or 1-to-many reconciliation scenarios. As a SaaS platform, the implementation process for new use cases can be done in less than a day without the need of IT. As experts in machine learning, our platform is data agnostic and can perform matching beyond bank reconciliations, including for hundreds of other use cases both in F&A and in operations.

Summary

The reduction in cost of cloud computing, the flexibility of SaaS solutions and the improvement in Artificial Intelligence approaches have created a new era opportunity for F&A staff. The matching reconciliation process is a complex, traditionally manual set of tasks that is perfect for a Machine Learning based solution.

If you deal with large scale, complex bank reconciliations and believe your staff's time is better spent on higher value tasks then reach out and let's discuss how Sigma IQ might be able to help. Click here to schedule a demo.