Client Case Study: Cutting down PO processing from hours to minutes

When we first spoke with Jamie Shah, President of [Chem-Impex](https://www.chemimpex.com/) a manufacturer and distributor of high purity research chemicals, she mentioned a task that occupies 2 - 4 hours of her employees’ day: manually processing Purchase Orders received via email.

These emails are sent by all sorts of different buyers: universities, labs, pharma firms, distributors. And they’re all in the client’s format. Clearly, not an ideal situation, but a necessary process to bring in sales.

In Comes Koka Labs!

For us, this was the perfect opportunity for automation. Here was a defined, repeated process, that from a human point of view, didn’t add too much value. And with the power of AI, we could retrieve the information from these emails and PDFs and parse them in the right format to provide visibility and upload to the ERP system via API.

The first thing we did was understand exactly how the process works. When, where, and from whom are these emails received? What do they look like? And from there, what was the step by step process from understanding the contents, getting the relevant information, and entering it into the backend system.

Once we understood all the pieces, we got to work.

How We Built the Automation

We used a visual workflow builder called n8n. It allows us to lay out each step of the process while allowing us to insert code “nodes” and call APIs as needed.

We created nodes for each step, while setting up a dummy email and output sheet so we can test it out. The good folks at Chem-Impex provided us with a variety of old purchase orders to get started. We used different software components - dedicated nodes, code blocks, and of course, LLMs to create the workflow:

Check inbox for new emails

Check whether these emails had purchase orders

If so, extract the contents of these purchase orders via an LLM

Organize the contents of the purchase order

Match customer ID and product information with internal tables

Output all relevant sheets for the client to view

Create a JSON to upload via API to ERP system

Label the emails in the inbox as processed

Finally, we stress tested this with 100 POs and made sure the system handled all “edge cases.”

Outcome

We were very excited to hand it over to Jamie and her team. We’re expecting overall processing time to be cut from multiple hours for multiple people to just a few minutes!