
The accounts payable department is responsible for paying suppliers. The function is responsible for validating that invoice information is accurate and for ensuring payments are made in time.

This process is highly time consuming for Finance teams. In addition to the time required for this process, accuracy is critical. Over and underpaying invoices both have significant consequences for businesses. For companies that deal with a large number of invoices, fraud can also be a concern.

With the power of AI, companies can consider automating large parts of the invoice process. This allows the accounts payable team to get through more invoices and focus on the ones that really matter. The invoice process can be broken down into a few steps:

## Extract information

Most invoices are received over email. Information about the invoice is provided in an attachment, usually a PDF. There's also important information that may be contained in the body of the email. The first step is extracting information from both these sources. Most tools use OCR, which has worked well to date. With the advent of large language models, this can be significantly improved.

## Validate information

Once data has been extracted from the invoice, it needs to be validated. Every business has its own rules for validating invoices and any automation tool should accomodate these rules.

The first step of validation is checking that all relevant information is present on the invoice. For example, if your business requires a VAT number, this should be validated on the data that was extracted. If the data is missing, the invoice can be flagged to a member of the accounts payable team or an email can be sent automatically to the vendor.

Once all fields have been validated, it's important to match the invoice against POs, reciepts or other internal records. This often called the 3 way matching system. It helps separate invoices into matched and unmatched invoices. By leveraging AI, this matching process can be automated so that your team spends time on invoices that are either very large, unmatched or both.

## Approval of invoices

Most businesses will have an approval chain for invoices.

Typically, invoices are routed to the right person using the supplier name, the amount or both. Once again, these rules differ from business to business. It's critical that the tool allows these rules to be customised. This process can break whenever someone leaves the organisation or moves to a different department. It is critical that the tool flags instances where the approval chain is broken.

## Entering information into your ERP or accounting software

One of the most time consuming parts of this process is inputting data into the ERP or accounting software that your business uses.

ERP systems require information in a specific format. On the other hand, every invoice you receive is generated by the vendor's ERP system and is therefore in a different format. This means your accounts payable department has to manually transform this data into the relevant format and then insert into the ERP. This process is very error prone.

The best invoice processing software will perform this task automatically. It will transform the information extracted from the invoice, transform it into the format you need and push it directly to your ERP system. With the power of LLMs and AI, this process can be highly automated.

## Measuring success

With any automation process, it's important to measure success.

In the case of invoice processing, success can be measured using a few metrics. Processing time measures the time between the receipt and payment of an invoice. Error rate measures the number of times there was a duplicate payment, data error or payment error. In addition to these core metrics, it's useful to look at the average approval time. This helps identify if the problem is with your team's approvals or with a different part of the process. Finally, it's always helpful to understand how satisfied suppliers are.

## Automating invoices with Kili

Kili provides an out of the box solution to automate invoices received over email. Our digital worker will automatically identify emails related to invoices, parse them, extract information and push them to your ERP. The process can be customised to your business' needs. Our builder allows you to define specific approval chains, automated checks and more.

Check out a demo of how Kili can do this for you [here](https://share.descript.com/view/aWp427l3IRX).


Automate invoice processing


Every job involves some amount of repetitive work. The account executive who needs to enter information into their CRM, or the HR executive who needs to answer the same question about paid time off.

If these tasks are repetitive, why hasn't software delivered on the promise of automation? Usually, there are two reasons: first, the task may be ever so slightly different when you choose to do it. Second, the effort of automating the task may have exceeded the value of doing so.

## Thinking on the fly

Large language models (LLMs) allow us to address both of these issues, unlocking automation like never before. It is now possible to build automations that are fluid. Consider the sales executive who needs to keep their CRM up to date. The information required to update the CRM lives in calls, emails and any notes they've taken. If they need to format this information for the CRM, they may as well just do it via the Salesforce (or whatever CRM they use) user interface.

On the other hand, if the update happened automatically by plugging into these resources, it would save the Account Executive precious time. Time that can be spent following up with leads and closing more deals. Large language models make this possible because they provide intelligence on the fly. The LLM can summarise the content and format it for the CRM. In the best case scenario, the AE needs to do nothing because information moves seamlessly into the CRM.

## Human in the loop

In some cases, we will find that AI needs to work in partnership with a human. The best example of this is the financial operations function. Every business needs to receive invoices, check them and pay vendors. This is typically handled by an accounts payable function.

Today, most businesses have software that helps with this but the automation is incomplete. This is because every business has it's unique process for handling invoices, and because every invoice looks different. Traditional automation tools like robotic process automation have low success rates because of this variance in the process.

LLMs address this shortcoming because they allow us to process information that may not always follow the same structure, like invoices. They significantly improve accuracy and reduce fraud. Even with LLMs, you require a human in the loop to avoid any chance of a mistake. The difference though is that it allows your team to handle more invoices whilst simultaneously improving reliability.

This human, AI collaboration also allows your team to work on the most pressing issues. For example, an invoice that looks fraudulent or resolving an issue that may have sent the funds to the wrong bank account.

## Employees managing digital workers

At Kili, we believe AI will enable every employee to 10x their output. We believe they will do so by managing digital workers that complete repetitive tasks for them. Instead of spending time getting context from different systems, employees will be able to spend time on what decision to make.

We believe in a world where AI quietly does work for your team in the background, and only reaches out to them when they need to make a decision. We're excited for this future and are building for it.

Join our closed beta if you're interested in leveraging this capability for your business.


Blog

Automate invoice processing

Delegate your repetitive tasks