Skip to content
Curious Context Labs

case study · Operations · Automation

Trade orders in the ERP within minutes, whatever format they arrive in

How PDFs, photos and email lines can become draft sales orders matched to your SKUs, confirmed by a person in one click.

business
Logistics and distribution
build
Automation
timeline
5–6 weeks
read
3 min

01 the business

Electrical wholesaler

75 staff, 1,100 trade accounts

Trade customers order by PDF, by photo of a handwritten sheet and by a line in an email. Two people retype every order into the ERP and translate customers' part names into the wholesaler's SKUs. Typos lead to wrong items shipped, and orders wait whenever one of them is off.

02 before

The problem

Trade customers send orders however suits them: a PDF from their system, a photo of a handwritten sheet, a line in an email. Someone in the office retypes each one into the ERP, matching their part names to your SKUs. Typos mean wrong items shipped, and orders wait in the inbox when that person is off.

  • Orders wait in the inbox until someone can key them.
  • Customers' own part names are matched from memory.
  • Pricing and credit checks happen after the order is entered, if at all.

03 the build

What the AI does

  1. Step 1: Watches the orders inbox in Outlook or Gmail and picks up every attachment and email body that looks like an order.
  2. Step 2: Reads the customer, delivery address, line items, quantities and requested dates, whatever the format.
  3. Step 3: Matches each line to the right SKU in NetSuite, SAP Business One, Cin7 or Unleashed, using the customer's past orders and part-number cross-references.
  4. Step 4: Checks prices against the customer's price list, stock on hand and credit status, and notes anything that does not line up.
  5. Step 5: Creates a draft sales order in the ERP with the original document attached and any questions highlighted.

paused: waiting for a person to approve

How it was put together

  • Watched the orders inbox and picked up attachments and email bodies that look like orders.
  • Read lines, quantities, dates and delivery addresses from any format.
  • Matched each line to a SKU using the customer's order history and cross-reference tables.
  • Checked price lists, stock and credit status, and created a draft order with the original attached.

04 human in the loop

Where a person stays in charge

A sales-ops or customer-service person reviews each draft order and confirms it with one click. Orders with unmatched items, price differences or credit holds are held until they decide what to do.

  • A person confirms every draft order before it is released.
  • Unmatched items, price differences and credit holds stop the order until someone decides.
  • The original document stays attached for audit and disputes.

05 rollout

From kickoff to live in 5–6 weeks

  1. Weeks 1–2Collect three months of real orders and build the cross-reference tables.
  2. Weeks 3–4Draft orders in parallel with the team and measure match accuracy.
  3. Weeks 5–6Switch to review-and-confirm for all incoming orders.

06 after

What changes

Orders are in the system soon after they arrive, without retyping. The team spends its time on the odd ones that need a phone call, and fewer wrong items go out the door.

worked example · your numbers will differ

Assume 2,400 orders a month with an average of 8 lines, at 45 seconds a line to key by hand and 1 minute an order to confirm a draft.

2,400 × 8 × 45 seconds = 240 hours a month of keying. 2,400 × 1 minute = 40 hours of confirming. About 200 hours a month saved.

What to measure

  • Time from order received to order entered.
  • Line-match accuracy before confirmation.
  • Wrong-item shipments.
  • Orders entered per person per day.

07 evidence

What the research says

  1. Gartner research put the average cost of poor data quality at $12.9 million a year per organisation.

    Gartner, Data quality: why it matters and how to achieve it (opens in a new tab) · 2020

  2. About 75% of generative AI's potential value sits in four areas: customer operations, marketing and sales, software engineering, and R&D.

    McKinsey & Company, The economic potential of generative AI: the next productivity frontier (opens in a new tab) · 2023

08 lessons

Mistakes to avoid

  • Skipping the cross-reference work. Matching is where accuracy comes from.
  • Auto-releasing orders. Keep the one-click confirmation.
  • Ignoring credit holds. The draft should surface them, not bury them.

09 questions

Common questions

Which ERPs does it work with?
NetSuite, SAP Business One, Cin7 and Unleashed are common, plus most systems with an API or import.
What about EDI customers?
EDI orders already arrive structured. This is for everyone else who orders by email.

keep reading

All case studies ->

Tell us how your business runs. We'll tell you where AI fits.

A free 30-minute call. You talk, we ask questions, and you leave with a few concrete ideas — or an honest answer that AI isn't the right fix yet.

Working with teams around the world.