Many companies still take orders the way customers like to send them: by email, in their own words. Someone reads each email, finds the customer, retypes the lines into the ERP, checks stock and writes back. It works, until volume grows or the person who knows every customer is on holiday.
Follow one order
Here is an example order, the kind a food distributor receives every day. The names and figures are illustrations.
Hello, could you send 240 boxes of gochujang 1 kg by Friday? Same price as our last order. Thanks, Jihoon
Done by hand, this one email means opening the CRM to find the customer and their last price, the ERP to check stock and create the order, and the mailbox again to confirm. Every retype is a chance to get a quantity or a price wrong.
What the automation does
- Reads the message. It recognises an order: one product, 240 boxes, delivery by Friday, "same price as last time".
- Finds the context. It matches the sender to the customer in the CRM, the product to the item in the ERP, and looks up the price of the last order.
- Checks it can be done. Stock, delivery date, credit limit: whatever your rules say must be true before an order is accepted.
- Acts. It creates the order in the ERP, updates the CRM and tells the warehouse.
- Confirms. It replies to the customer, in the same thread, with the order number and the delivery date.
If everything matches, nobody had to touch the order. If something does not, it goes to a person.
What waits for a person
- orders above a value you set, for example ₩3,000,000;
- a product or customer the system cannot match with confidence;
- a price different from the agreed one, or a discount request;
- not enough stock for the requested date;
- anything unusual: the automation stops and asks instead of guessing.
The person gets the email, the matched customer, the proposed order and the reason it was stopped, in one place. Approving takes seconds.
What you need before you start
- An ERP and a CRM with an API, or another way to read and write orders safely.
- Clean master data: customer and product codes that people actually use.
- Written rules: limits, who approves, what counts as an exception.
- A baseline: orders per day, minutes per order, errors per month, and how long customers wait for a confirmation.
Start small
Pick the customers who send the most routine orders, run the automation on them first, and compare the numbers with the baseline after a few weeks. Add customers and order types once the results hold.
Razum AI does exactly this kind of handoff, on top of the email, CRM and ERP you already use. See the order flow, or tell us about your workflow.
