Hundreds of purchase orders, delivery notes and inspection forms landed every day in every format imaginable, from scanned PDFs to photos snapped in a tractor cab and sent over chat. The back office decoded and re-typed all of it into the ERP by hand, which held up deliveries and, with hazardous materials in the mix, risked the kind of recording error that draws a heavy penalty.
Farmers order the way that suits the field, not the way that suits a back office. Every day the company received hundreds of orders, delivery notes and regulatory inspection forms in completely inconsistent formats, from tidy scanned PDFs to a blurry photo of a handwritten form sent over WhatsApp.
Making that usable fell to people. The back-office team decoded each document and hand-keyed the details into the ERP, a slow, sisyphean load that formed a bottleneck at headquarters and pushed back the moment an order actually turned into a delivery.
And the stakes were higher than speed. These are fertilisers and pesticides, so a mistyped quantity or the wrong material against the wrong plot is not just an admin error, it is a safety and regulatory exposure that can bring a heavy penalty.
Put an AI agent in front of the ERP as a logic guardrail: it reads the document, validates it, and enters it, holding anything doubtful for a person.
The agent scans and reads whatever comes in, a clean PDF, a low-quality scan or a photo taken in the field, and pulls the meaning out of all of them, so the format the farmer happened to use no longer decides how much manual work an order creates.
From each document the agent extracts the critical details, the material and quantity, the plot numbers, the order specifics, turning a picture of a form into structured data the business systems can actually act on.
Before anything is written, the agent validates the extracted data against the agriculture ministry's safety rules, so a combination that breaches regulation on a hazardous material is caught at the door rather than discovered later in an audit.
Rather than replace the core systems, the agent sits as an intelligent layer in front of the existing ERP and enters the validated data automatically, so the company kept the system it relies on and simply stopped feeding it by hand.
When a document is unreadable or the data fails a safety check, the agent does not guess. It holds the item and routes it to a back-office person to resolve, so automation handles the clean majority and human judgement is reserved for the genuine exceptions.
With orders read and filed the moment they arrive instead of waiting in a typing queue, the whole supply chain moved faster, and the material reached the farmer in the field with less of the delay that manual processing used to add.
The typing disappeared, the compliance held, and the team moved to work that matters.
The time spent decoding and hand-keying documents dropped 82% straight away, clearing the headquarters bottleneck that had been holding up deliveries and freeing several back-office staff entirely from data entry to focus on procurement and on the relationships with the farmers they serve.
Because the agent checks every document against the safety rules before it is recorded, the company logged zero regulatory breaches from the day the system went live. The strict automated control turned the most dangerous part of the process, recording hazardous materials, into its safest.
Reading a document is the easy half; the value is refusing to write it until it passes the safety rules, and holding the doubtful cases for a person. That guardrail-in-front-of-the-ERP design is what let the company automate hazardous-material paperwork without taking on risk. It is the same legacy-integration discipline behind our workflow-automation work, aimed at an agricultural supply chain and its safety regulation rather than a property manager's back office.
A large Israeli property manager had 8 people re-keying data by hand. We automated it end to end around a 15-year-old ERP. Average handling time fell from 4 hours to 30 seconds.
Read the case → Credit scoring · AgricultureApproving farm credit meant sending a surveyor out for weeks. Computer vision on satellite imagery scored crop health and collateral, taking the company from dozens of applications a week to 25,000 farms a day and cutting underwriting cost over 70%.
Read the case →
Reef TRH