How an AI agent prepares a quote
It reads the request, drafts the quote from your price lists and earlier quotes, and marks what it can't match. Your estimator checks the draft and decides what goes out.
- Item 1 · Steel bracket, 200 pcsPrice list, line 14
- Item 2 · Powder coatingYour March quote, item 3
- Item 3 · TransportYour transport rule
- Open: delivery datemissing in request
Your estimator decides
Twelve steps by hand become one agent loop.
An example: a quote in a firm that makes parts to order. Your steps will look different, and the trial run shows which of them an agent can take on.
Example: a quote by hand today
- 01Open the request email
- 02Find the customer in the ERP
- 03Read the drawing and the item list
- 04Look up each item
- 05Check the supplier price lists
- 06Search for a similar quote
- 07Work out the price
- 08Ask production about the date
- 09Write the quote
- 10Check it against the request
- 11Send it
- 12File it and note it in the CRM
- 01
Reads
The request, your price lists and earlier quotes.
- 02
Drafts
A quote with a source on every line.
- 03
Marks
Whatever it can't match with certainty.
- 04
Your estimator decides
Changes what's needed and approves.
- 05
Corrections go back in
Each correction goes into our monthly check.
One loop, with your rules at every step. Every hand-off is logged.
At first it only suggests.
Nothing is sent, booked or filed until one of your people approves it. It writes to the ERP only through the route your IT partner signs off.
Scheduled jobs, such as loading a new price list overnight, start at a set time, and the result is waiting for you in the morning.
One place to ask for approvals, in Teams or Slack.
Today your people chase sign-offs in chats and emails. An agent sorts each request, adds the context and sends it to the person who decides.
Request in
In Teams or Slack, where your people already ask.
The agent sorts it
It sums up the request and passes it on with what the person needs to decide.
A person decides
Within an agreed time, and the decision is logged.
A dashboard for each role on your team.
Each view is read-only and takes its figures straight from your systems.
Examples. We build each view with your team.
When a draft is wrong
Your estimator corrects it before approving. We see the correction, find the cause, an old price list for example, and fix the rule behind it.
When the model changes
We keep the job and the data separate from the model, so the model can change. Before we switch, we check the drafts on fixed examples, and your documents, rules and workflow stay as they are.
We look after your agents every month: new price lists, changed rules and a check against fixed examples.
Where they runData and securityOur approachTrial run on your own data
Questions about how agents work
How is this different from a chat with ChatGPT or Copilot?
In a chat, one person asks one question at a time and copies the answer into the systems by hand. An agent does a fixed job between your systems, by your rules, and we look after it every month.
Can we see what the agent did?
Yes. Every step is logged, so you can trace each draft back to its sources and see who approved it.
Which model does the agent use?
The one that suits the job and how sensitive the files are. We keep the job separate from the model, so the model can change later, and before it does we check the drafts on fixed examples.
Does the agent write into our ERP?
Only through the route your IT partner signs off, and only after one of your people has approved the result. At first it only reads.
See it on a job of your own.
In the call we work out which job suits a trial run on your own documents.
30 minutes with Lukas. You leave with at least one recommendation you can use, whether or not we ever work together.
