30 employees
approximately $6 million in annual revenue
roughly 30 years in business
This company has been operating successfully for roughly 30 years.
It has around 30 employees, approximately $6 million in annual revenue, and produces highly customized fabricated work for commercial clients.
Leadership wanted the business to become easier to run with less day-to-day dependence on the founders and a small number of key employees.
AI-assisted quoting looked like an obvious place to start.
It was not.
A typical job moves through:
The company had grown around experienced people who knew how to make that system work.
Leadership wanted the business to become easier to run with less day-to-day dependence on the founders and a small number of key employees.
That was the real goal.
AI-assisted quoting looked like an obvious place to start.
The company had years of job history, experienced people with strong intuition, and a quoting process that could potentially benefit from faster access to comparable work and past decisions.
But before building anything, we mapped how a job actually moved through the business.
That changed the plan.
The process on paper and the process people actually ran were not the same.
A large master spreadsheet carried too much of the operation.
Important information moved through conversations, inboxes, whiteboards, and individual memory.
Experienced employees held knowledge that the systems did not.
And jobs often changed after the original quote had been approved.
Those changes could affect materials, labor, production instructions, deadlines, shipping, customer expectations, and margin.
The problem was not that changes happened.
The problem was that the company did not have one reliable way to capture what changed, why it changed, who approved it, and what it affected downstream.
That made one question surprisingly difficult to answer:
What is true about this job right now?
An AI quoting system needs more than old quotes.
To learn from completed work, it needs to understand:
If those facts are scattered across spreadsheets, inboxes, conversations, and memory, better AI does not solve the underlying problem.
It sits on top of incomplete information.
So we made a deliberate decision:
Start by making the job history reliable enough for AI to become useful later.
We followed the work from customer request through delivery.
The goal was not to document the official process.
It was to understand how information, decisions, approvals, and ownership actually moved through the company.
That exposed job changes after release as a major point of friction.
The current focus is job change control.
For every meaningful change, the company needs to be able to answer:
The purpose is simple:
Make the current state of every live job clear and traceable.
Once changes are captured consistently, the company can start connecting:
What we quoted → what changed → what the work required → what it actually cost → what margin it produced
That creates a feedback loop the business does not have today.
It makes it easier to see where estimates were right, where reality diverged, what types of changes create cost, and what should be priced differently next time.
Some of the company’s most valuable knowledge still lives with experienced people.
They know which customer requests create trouble.
They know when a production plan looks wrong.
They know which exceptions deserve attention.
They know what to check before committing to a job.
The long-term goal is to make more of that judgment available to the company instead of leaving it only in individual memory.
Once the company has a reliable history of jobs, changes, outcomes, and decisions, AI becomes useful in ways it was not before.
An employee dealing with a new request can find comparable completed work, decisions, problems, and outcomes without searching folders or asking around.
Instead of managers checking every job, systems can surface the cases that actually deserve attention.
Future quotes can be informed by what similar jobs really required, where changes occurred, and what those jobs ultimately cost.
Routine comparison, information extraction, reconciliation, status checking, and follow-up preparation can happen inside the redesigned workflow instead of as separate manual tasks.
The point is not to add AI everywhere.
It is to introduce it where better information and a better process allow it to change the work.
This is an ongoing engagement, not a finished before-and-after story.
The first AI opportunity is not always the right first project.
This company may eventually use AI to support quoting, retrieve precedent, surface exceptions, and handle routine administrative work.
But those capabilities become far more useful once the company has reliable information about what actually happened.
That is the broader Cambriant approach:
Do not start by asking where AI can be added. Start by asking how the work should run now that AI exists.
Then build the conditions that make the answer possible.
START WITH THE WORK
You do not need an AI use case ready. In a discovery call, we learn how the business operates today, where information and decisions get stuck, and where redesign could create the most value.
Book a discovery call