Microsoft Copilot – all you need, or should you Bankfeed?

Payment reconciliation is crucial for day-to-day operations, accurate financial reporting, cash flow management, and identifying discrepancies. The payment-invoice matching task is typically a multi-step process, starting with getting bank statements out of your banks and into an accounting system, matching transactions to ledger entries, and posting new amounts to reflect any residual transactions that weren’t previously known.

Luckily, the days when payment management was a tedious manual process of putting the puzzle pieces together are gone. As the list of automated solutions expands, we are here to explore the evolution of payment reconciliation methods available in Microsoft Dynamics 365 Business Central: the standard ruleset, generic AI (artificial intelligence), and trained AI.

As a standard feature, the Bank Acc. Reconciliation page in Business Central offers automated rules that match part of the bank transactions with ledger entries. Pre-defined rules can be customized to fit specific business needs, enhancing the matching accuracy. However, those rules are static, often outdated, and thus result in many unmatched transactions requiring manual inspection. It becomes time-consuming, especially for businesses with high transaction volumes. Not to mention that manual processes are susceptible to human errors, potentially leading to inaccuracies.

Bankfeed uses a similar, yet more advanced, method for algorithm-based payment reconciliation. Bankfeed first recognizes each transaction using a complex list of pre-defined rules to match payments with invoices. The accrued domain-specific experience of various industries and countries allowed us to incrementally grow a set of rules and logic that successfully matches bank transactions with an 80-95% accuracy, having the option of custom tailoring the system to each customer’s specific outliers.

Bankfeed takes all the available data to match the account and the actual entry. The customer or supplier can be recognized by their name, IBAN or registration number, while the document to match with is identified using the “Message to Recipient” or the “endToEndID” fields. Additionally, text-to-account mapping is used to identify bank fees. All of it adds up to an automatic payment-invoice matching that saves countless hours of otherwise labour-intensive, repetitive work.

Sounds great but can it be even better?

Here comes Microsoft Copilot. Saying that it’s an AI based on a large language model won’t say much – it’s easier to see it as your personal assistant ready to help you in the ways you need the most. Step by step, Microsoft is introducing it across their product line, with Business Central being no exception.

As a proof of concept, Microsoft presents us with additional help in the reconciliation process using the “Reconcile with Copilot” feature. It inspects the remaining transactions after the rule-based matching and identifies more matches based on the dates, amounts, and descriptions. For example, if a customer paid multiple invoices as one lump sum, Copilot reconciles the single bank statement line with the multiple invoice ledger entries.

In addition, Copilot suggests the most likely G/L account to post any residual transactions to and offers the opportunity to remember a specific transaction description for the next reconciliation. For example, Copilot might suggest that the transaction with the description “Fuel Stop24” should be posted to the “Transportation” account.

However, being just a proof-of-concept, it has its limitations. As a generic AI, Copilot may not cater perfectly to industry or business-specific nuances without extensive training. It also performs best when G/L account names, ledger entry descriptions, and bank transaction descriptions are all in the same language, preferably – English. Mixed languages often result in fewer matches and suggestions – at least for now.

It’s a powerful tool, though. With smart usage, Copilot can generate new matching rules, help detect or explain discrepancies and take care of some rare but specific scenarios. And that’s also exactly what you should expect from the upcoming releases of Bankfeed. We’ll give you a sneak peek into that.

While the generic AI of Copilot is a step forward, it still may not cater perfectly to the specific industry or business without extensive training. That is where machine learning and trained AI come in. A data model from Bankfeed uses advanced machine-learning techniques to analyze and match the bank transactions with the posted documents in the system. It keeps training on financial data to make accurate and reliable matches. The solution continuously learns from user actions, improving its performance over time. It might very well feel like magic.

This comes with its own set of challenges though – volume of data, performance, responsibility, and availability. Having all that in mind, Bankfeed utilizes all the available tools incrementally – starting with an old but good human-made ruleset, continuing with Copilot and reaching for the skies (100% payment-invoice matching rate) with the trained AI data model, where applicable. And though this sounds quite complicated, you shouldn’t be worried – it’s our goal to make this seamless to the end user.

The journey from traditional, manual payment reconciliation to the sophisticated use of AI technologies like Microsoft Copilot and Bankfeed represents a remarkable evolution in financial management. This transition signifies a leap towards efficiency and accuracy and opens the door to future innovations in the field. While Copilot offers a glimpse into the potential of AI assistance in automating reconciliation, we believe that our approach of combining human-made rulesets with AI advancements, and machine learning promises a future where payment-invoice matching can achieve unparalleled accuracy.