Order processing · Production

Choco AI automates food distributor order intake with LLMs, achieving over 95% prediction accuracy

The problem

Food and beverage distributors received orders through multiple unstructured channels (SMS, WhatsApp, voicemail, email, fax) and employees had to manually interpret and enter each order into their ERP system, often switching between devices and printouts.

First attempt

Initial outsourcing of human labeling to an external agency led to unreliable results due to lack of domain expertise.

Workflow diagram · grounded in source
1
Multi-channel order intake
Trigger
Orders arrive via email, voicemail, SMS, WhatsApp, and similar channels and are collected into the system.
▾ source quote
“collecting unstructured orders (via email, voicemail, SMS, WhatsApp, etc.) and integrating them directly into distributors' ERP systems”
2
LLM text extraction
Ai action
LLMs extract text from diverse order formats and identify relevant details.
▾ source quote
“LLMs extract text from diverse order formats and identify relevant details”
3
ML product catalog matching
Ai action
Custom ML models match extracted information to the correct products in the distributor's catalog.
▾ source quote
“Our custom ML models then match this information to the correct products in the distributor's catalog”
4
Distributor review and correction
Human review
On a single screen distributors review predicted order details and can edit mispredicted entries before accepting.
▾ source quote
“On a single screen, incoming orders are displayed on the left, while our system's predicted order details are displayed on the right. Distributors can quickly review and edit mispredicted entries before clicking "Accept order"”
5
ERP sync
Integration
Accepted orders are seamlessly synced to the distributor's ERP system.
▾ source quote
“clicking "Accept order" to seamlessly sync orders to their ERP system”
6
Correction-driven learning
Feedback loop
Corrections made in the review UI are automatically learned from so the system improves on future orders.
▾ source quote
“If our system predicts the wrong tomato, the customer can make that correction in the UI, which Choco AI will automatically learn from so that it is more likely to get it right the next time around”
Reported outcome

In one distributor case Choco AI reduced manual order entry time by 60% and enabled processing of 50% more orders daily without additional staffing, while achieving over 95% correct predictions system-wide and scaling to hundreds of new customers.

Reported metrics
Manual order entry time reduction (one distributor case)60%
Orders processed daily increase (one distributor case)50% more orders daily
Prediction accuracyover 95%
Customer scalehundreds of new customers
Show all 5 reported metrics
manual order entry time reduction (one distributor case)60%
orders processed daily increase (one distributor case)50% more orders daily
prediction accuracyover 95%
customer scalehundreds of new customers
labeled training examples accumulatedtens of thousands of examples
Reported stack
ChatGPTGPT-4oLLMsWhatsAppERP systems
◆ Does this fit your context?

Compare to your context

Tell us your scale, team, and constraints. We'll show what changes at your size, what fails at your scale, and whether this case is a fit, needs adaptation, or won't scale to you. Free demo, no signup.

Compare to your context →
~30 seconds · free
Source
https://choco.com/us/stories/life-at-choco/scaling-ai-applications-with-llms
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

In one distributor case Choco AI reduced manual order entry time by 60% and enabled processing of 50% more orders daily without additional staffing, while achieving over 95% correct predictions system-wide and scaling…

What tools did this team use?

ChatGPT, GPT-4o, LLMs, WhatsApp, ERP systems.

What results were reported?

Manual order entry time reduction (one distributor case): 60%; Orders processed daily increase (one distributor case): 50% more orders daily; Prediction accuracy: over 95%; Customer scale: hundreds of new customers (source-reported, not independently verified).

What failed first in this deployment?

Initial outsourcing of human labeling to an external agency led to unreliable results due to lack of domain expertise.

How is this order processing AI workflow structured?

Multi-channel order intake → LLM text extraction → ML product catalog matching → Distributor review and correction → ERP sync → Correction-driven learning.

Related order processing cases

WHAT TO DO WITH THIS

Now compare it to your context

This case is one data point. Whether its pattern fits you depends on your volumes, your stack, and your exception load — that comparison is the step no case study can do for you.

Browse the Order processing cases →