What part of the reconciliation process are you targeting?

A bank statement lists the activity in the bank account during the recent month as well as the balance in the bank account.

A bank reconciliation statement is a summary of banking and business activity that reconciles an entity's bank account with its nancial records. The statement outlines the deposits, withdrawals and other activity aecting a bank account for a specic period. A bank reconciliation statement is a useful nancial internal control tool used to thwart fraud. The benet of reconciling the bank statement is knowing that the amount of Cash reported by the company (company's books) is consistent with the amount of cash shown in the bank's records.

The reconciliation process that I'm targeting is to ensure that the money leaving an account matches the actual money spent.

Sakayo,

Let me put it this way:

if we start with the steps in a reconciliation (lets take as example the steps given in https://corporatefinanceinstitute.com/resources/knowledge/accounting/bank-reconciliation/):

reproduced here for convenience:

Each of these can be further decomposed into further substeps. Which one of these steps are you targetting?

Regards

Step7

These are steps that I have to follow in my project:

1.Get bank and business records :You need a list of transactions from the bank, and your ledger of income

and outgoings.

2.Find your starting point : Find the last time the balance on your business books was the same as the

balance in your bank account. Start the reconciliation from there.

3.Compare bank and business records : Compare opening balances of both the bank column of the cash

book as well as bank statement ; these could be different due to uncredited or unpresented cheques from a previous period. Now, compare credit side of the bank statement with debit side of the bank column of

cash book and debit side of the bank statement with the credit side of the bank column of the cash book.

Place a tick against all the items appearing in both the records.

4.Analysis: Analyze the entries both in the bank column of the cash book as well as pass book and look

for entries which have been missed to be posted in the bank column of the cash book. Make a list of such

entries and make the necessary adjustments in the cash book.

5.Start reconciliation: Correct if any mistakes or errors appear in cash book. Calculate the corrected and

revised balance of cash book’s bank column. Then, start bank reconciliation statement with updated cash

book balance.

6.Check the expenses on your books : Add the un-presented cheques (cheques which are issued by the

business firm to its creditors or suppliers but not presented for payment-Expense) and deduct uncredited

cheques (Cheques paid into the bank but not yet collected-Income). Make all the necessary adjustments

for the bank errors. In case the bank reconciliation statement begins with the debit balance as per bank

column of the cash book, add all the amounts erroneously credited by the bank and deduct all the amounts

erroneously credited by the bank. Do vice-versa in case its start with the credit balance.

7.End balance : After you’ve checked all the deposits and withdrawals, your business bank balance should

match the totals in your business accounts. This will be the starting point for your next reconciliation.

Then I'm targetting the End balance, Step7 but following all the steps

Sakayo,

Machine learning can be implemented in this type of transaction can be described as a supporting role.

Example 1

Let say that you have an application that automates, by using rules the steps you outlined at step 7 you find that quantities do not match. Then based on historical data that is labelled then you can find that the reason that it does not match was with 90% confidence data entry error

Example 2 (not directly related to reconciliation but can be fitted within the flow)

In step 3 that Compare bank and business records machine learning can detect outlier patterns and flag them as possible deviations in customer behavior.

Example 3

Given that you have an application that allows users to do reconciliation you can train an ML to see if the user left steps within the reconciliation flow.

Machine learning is good in situations where patterns of data point to a general behavior, but the particulars of reconciliation are best suited for other techniques.

Regards

But according to you, which kind of other techniques that you I can apply to do this reconciliation?

Because the aim of my project is to build at the and a system that companies can use to do their reconciliation.

Is it possible in this case to implement ML algorithm (like train and do classification using random forest for instance) and then build to system with the ML algorithm inside?

Regards

Sakayo,

The problem with building a system fully on ML is that of accuracy based on training data, especially for reconciliation. While it can in principle be done via rule learning like random forest the question for you is:

Would you be willing to risk the liability if the system fails to properly reconcile transactions on X case that the data may not have?

That is why ML tools are at this stage for support purposes only

Regards

Dear Arturo!

You're right, now I don't know what kind of methods I can apply to solve the problem, I'm really confuse. It's a master topic thesis.

It's difficult to get articles on this, if I can changed this topic but no. If I can partially apply ML on it and then use another methode It's good but I don't know a mathematical model that I can combined with ML to build the system. If you can suggest me something

Regards.

It seems also that there exist already the software, system or technologie in this topic according to this:

https://medium.com/@lawrenceklamecki/using-machine-learning-to-solve-data-reconciliation-challenges-in-financial-services-b2f1a2dbb954

https://www.consileon.com/news/machine-learning-in-financial-services-and-reconciliation/

Regards

Sakayo,

I already was aware of the systems in the links you mentioned and they solve similar punctual problems such as the ones posed in my post.

It is even funny that in https://www.consileon.com/news/machine-learning-in-financial-services-and-reconciliation/

they start as machine learning and then switch to AI in the narrative. My opinion seeing this is that they are merely using ML in small peripheral tasks and use mainly hard coded rules to carry out the main reconciliation process but use the ML Buzzword to drive interest.

In https://medium.com/@lawrenceklamecki/using-machine-learning-to-solve-data-reconciliation-challenges-in-financial-services-b2f1a2dbb954

You can also see that the problems that they point as using ML are mere support roles

In most implementations such as these you will need to implement hard coded rules that can pass regulations that can be audited(lessens the liability).

Regards

Arturo

Now I really don't know can I proceed to solve this problem because I'm not a software developer. If I know a way to do it I can try but now I'm loss. I really want to apply just a bit ML and then solve the problem with a simple method.

Regards