Skip to content
Thnkers

Unblock

From PDF and DWG drawings to a priced draft: what a drawing assistant can and can't do

Software can read a specification and propose a bill of materials. It should never send a quote on its own.

Most quotes in make-to-order manufacturing start with a file, not a form. A PDF specification. A DWG single-line diagram. Sometimes a photo of a hand-marked drawing. Someone has to read it, turn it into a list of components and price that list.

Reading is the slow part. 83% of engineers already spend four or more hours a week on procurement tasks (Fictiv, 2026). Reading incoming drawings for quotes comes on top.

Software can now do a good share of that reading. Here is how we built it for a panel maker, what it does well, and where the line is.

How the draft is made

The flow in Quoted by hand. Then not. has five steps.

  1. Convert. DWG diagrams are converted to PDF so every file is read the same way.
  2. Read. The system reads the specification and drawings and identifies what each panel needs.
  3. Match. Each requirement is matched against the catalogue by specification and by analog.
  4. Price. The draft runs through the company's pricing formula: materials, assembly coefficient, cost, markup, VAT.
  5. Explain. Every proposed line carries a short rationale and a reference to the drawing sheet it came from.

Then the engineer reviews the draft line by line and confirms it. Nothing reaches the client unchecked.

The catalogue comes first

A drawing assistant is only as good as what it matches against. The client's price list was an 85 MB spreadsheet with 20 sheets and hundreds of images. No software can match reliably against that.

So the first two weeks went into structure. We imported the price list into a catalogue of 1,887 items organised into 223 analog groups by function, type, poles, rating and breaking capacity. Each category carries a risk flag: where a cheaper alternative is acceptable and where it isn't. The price difference between analogs is visible in the component card.

The drawing assistant arrived in week three. Building it first would have produced confident drafts against a messy list, which is worse than no drafts at all.

1,887

catalogue items in 223 analog groups, built before the drawing assistantThnkers portfolio

What it does well

  • The first read. It turns a stack of sheets into a structured list in minutes, so the engineer starts from a draft instead of a blank page.
  • Consistency. The same drawing produces the same draft and the same price every time.
  • Traceability. Each line points to the sheet it came from, so checking is fast.
  • Memory. The system records which components appear most often and surfaces that usage.

What it should not do

  • Decide. It proposes. The engineer confirms. That is a design rule, not a disclaimer.
  • Set prices. Prices come from the formula and the coefficients the manager sets. The assistant only feeds the formula.
  • Hide its reasoning. A line without a reason and a sheet reference should not exist.
  • Freeze. The instructions that guide the drawing analysis are editable by the director in the admin panel. When the business changes, it does not wait for a developer.

What the law asks for

Clients ask about it on the first call, and they are right to. The obligations depend on how the system is classified. We build so they are met by design: we classify the risk, log every proposal the system makes, document the data it uses and keep a person in charge of the outcome. The client receives the technical file with the system, so their legal team has answers before it asks.

What changes for the engineer

Before: the engineer read every drawing, searched the price list, chose analogs from memory and passed the list to the director for pricing.

After: the engineer opens a priced draft with reasons attached, checks the lines that matter, changes what's wrong and confirms. The director sees the assistant's usage statistics on the dashboard.

A priced quote now takes about two days instead of ten. The engineer reviews rather than reads.

Is this for you?

It is worth looking at if three things are true: requests arrive as files rather than forms, the same component families appear again and again, and one or two people do all the reading. If your requests are already structured forms, you need the formula and the catalogue, not the reader.

Related: Why your quotes take ten days · CPQ software or a quoting system built around your rules?

Questions

Can software read a DWG drawing and produce a quote?
It can produce a draft. In the system we built, DWG files are converted to PDF, read, matched against the catalogue and priced. An engineer confirms every line before the quote goes out.
What do we need before a drawing assistant is useful?
A structured catalogue. Without items grouped by function, rating and allowed analogs, there is nothing reliable to match the drawing against.
What does the EU AI Act require for this kind of system?
Obligations depend on the risk class. We build so they are met by design: risk classification, logging of every proposal, documentation of the data used, human oversight and a technical file delivered with the system.

From the portfolio

The Switchgear quoting app: a kanban order board with New, Sales, Engineering, Sign-off and Quote/Won columns.

Switchgear manufacturer (name under NDA) · Low-voltage electrical panels · Italy

2 days

to a priced quote, instead of about 10

Quoted by hand. Then not.

3 weeks

UnblockSee the business

Related goal

Unblock

Everything waits for one person.

One email a month. What we found, what we built, what it changed.

Tell us what's missing.

Thirty minutes. You describe the pain and the result you want. We ask questions. If we can help, you get one email with what we found, what we'd build and what it costs. If we can't, we'll say so.

or write to will@thnkers.com

Not ready to talk? See what the current way costs you