labs@atheron:~/industries/warehousing/ai$ train agent --grounded

AI for 3PL service

Agents that propose replies from your own system.

Most of a 3PL customer service day is reading emails and retyping them. An agent can do the reading and the typing, and leave the decision to the person who knows the client. That is the narrow, useful job we build it for.

Start a project See the example projects

// what matters here

What is different in this industry.

Proposes, never commits

Every entry the agent prepares waits for a person to approve it, and the confirmation is logged, so nothing reaches the floor that a person did not approve.

Answers from the system, not from memory

Replies about stock or order status are built from the WMS at that moment. The model writes the sentence; the facts come from your data.

Measured on your own mail

An evaluation set of real, anonymised emails is run before every change, so a new model version is checked on your clients' wording.

The route that fits the data

Routine business email suits a frontier model through its API for a fast start. If clients forbid that, an open-source model in your own cloud does the same job privately.

// example projects

Priced examples for this service.

Each one is hypothetical, labelled as such, and priced live from our rate card. Open any of them in the estimator and make it yours.

Example projectWeb appAIFrontier model

Inbound inbox agent

Hypothetical. Not a client, not a result.

// The problem

The customer service team spends the morning reading client emails about inbound containers, rush orders and address changes, and retyping them into the WMS.

// What we would build

An agent that reads the shared inbox, recognises ASNs, order changes and appointment requests, fills them into the WMS for a person to approve, and proposes nothing it cannot trace to a source. It uses a frontier model through its API, the fastest start, since the emails are routine business correspondence.

// What is in it

Emails and attachments sorted into ASNs, order changes, appointments and questions
Proposed entries in the WMS, confirmed by a person with one click
Replies prepared from the order's real status, never invented
An evaluation set of real emails, run before every change
A frontier model through its API, with personal data kept to what the task needs

// Stack

Frontier model API
Python
Microsoft Graph or Gmail API
Next.js
PostgreSQL

// estimate

Build: ≈ US$81,700 to US$125,000, delivered within 19 weeksCAD 116,300 to 177,800
Hosting: ≈ US$558 a monthCAD 795 a month
Support: ≈ US$1,000 a monthCAD 1,425 a month

Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.

AI route: Frontier model
Model running cost: ≈ US$383 a monthCAD 545 a month

Timeline by milestone

Discovery: 2.9 to 3.4 weeks
Specification and evaluation plan: 0.4 weeks
Design approved: 2.4 to 3.4 weeks
Core features: 1.4 to 1.9 weeks
Full build: 0.9 to 1.4 weeks
Working pilot: 2.9 to 4.4 weeks
Testing and fixes: 0.9 to 1.4 weeks
Launch: 0.9 to 1.4 weeks
Production: 0.9 to 1.4 weeks

How it is paid

Deposit 20%: CAD 23,260 to 35,560
Discovery 2.5%: CAD 2,907.50 to 4,445.00
Specification and evaluation plan 9.4%: CAD 10,932.20 to 16,713.20
Design approved 2.5%: CAD 2,907.50 to 4,445.00
Core features 7.6%: CAD 8,838.80 to 13,512.80
Full build 5%: CAD 5,815 to 8,890
Working pilot 28.2%: CAD 32,796.60 to 50,139.60
Testing and fixes 2.5%: CAD 2,907.50 to 4,445.00
Launch 2.5%: CAD 2,907.50 to 4,445.00
Production 9.8%: CAD 11,397.40 to 17,424.40
Holdback, 30 days after launch (10%): CAD 11,630 to 17,780

Start from this example

// questions

Questions we get about this.

Will the agent send emails to our clients on its own?

Not unless you decide it should for a given type of message. We start with every reply waiting for a person, and loosen that only where the record shows it is safe.

Does it need access to our WMS?

Read access for status and write access for proposed entries, through an API or a database view we agree with you. It never sees more than the task needs.

What happens with an email it cannot classify?

It goes to the normal queue with a note on what the agent could not tell, so nothing is lost and the misses become new test cases.

Back to the industry: Warehousing and 3PL About the service: AI agents and models

// next

Not quite your project?

Tell us what you have in mind. We will come back to you with a range and the questions that would narrow it. Or book a call and talk it through.

Start a project Book a call