1. Home
  2. Services
  3. Quote automation

AI quote automation for manufacturers

Quote automation for manufacturers that price parts from a customer's drawing. An AI agent reads the request for a quote (RFQ), the drawings and the CAD files. It matches each part to your materials and past jobs. Then it drafts a priced quote in your ERP or CRM, the software that runs your orders and customers. A person on your team checks the draft and approves it before it goes out. Where a part is bought in, the agent drafts an RFQ to your suppliers too.

Drawings and CAD files readParts and materials matchedPriced draft in your ERPA person approves every quote
Agent run, RFQ with 3 drawings and 1 STEP file
Quote QTN-0418
Draft
bracket_rev-C.pdf, title block and notes Read
housing.step, volume worked out Measured
Material 6061 aluminum, matched to stock item Matched
Anodize note, RFQ drafted to 2 suppliers Waiting
Saved as a draft for the estimator to approve

Illustration with sample data.

What quote automation means on this page

The quote is drafted where your team already works. We build the agents inside your ERP, CRM and email, so there is no new app to log in to. An AI agent is a program that uses an AI model to do one set job, such as reading a drawing. CAD stands for computer-aided design. A CAD file is the design itself, as a 3D model or a 2D drawing.

Our team built a multi-agent RFQ system that reads drawings and CAD files. It runs in ERPNext, an open-source ERP, and Frappe CRM. Customer quotes came first, then RFQs to suppliers. Separate agents handle intake, matching, pricing, supplier RFQs and reading supplier replies. A person approves every quote before it goes out. Read the case study: a multi-agent RFQ system in ERPNext.

What the agent reads on a drawing

A drawing carries far more than a shape. These are the fields an agent pulls out for a quote, in the usual design.

The title block

The part number, revision, material and finish usually sit in the box in one corner. The revision matters. A quote for rev B is wrong for rev C.

Dimensions and tolerances

The sizes, and how far each may vary. Tight limits add machine time and inspection. ASME calls its Y14.5 standard the guideline for this drawing language, known as geometric dimensioning and tolerancing (GD&T).

The notes

Heat treat, plating, coating, deburring and inspection. Each note can mean an outside process, with its own supplier quote.

The RFQ itself

Quantities, the price breaks asked for, the date needed and the delivery terms. These come from the customer's email or form, not from the drawing.

Anthropic says its model's answers about where things sit in an image are approximate. So the agent takes each dimension from the printed number. A number it cannot read cleanly goes to a person. It is never scaled off the picture.

CAD file types, and how an agent reads each one

Customers send what their design software exports. Some formats go straight to the AI model. Others need code to open them first.

Scroll the table sideways to see every column.

FormatWhat it isHow the agent reads itSource, read 17 Nov 2026
PDF drawingThe 2D drawing, saved as a PDFThe AI model gets each page as an image plus its text. Anthropic allows up to 600 pages in one request, or 100 when the request's context window is under 1 million tokens.Anthropic, PDF support
JPEG, PNG, GIF or WebPA scan or a photo of a drawingThe AI model reads it as it is. Each image can be up to 8,000 by 8,000 pixels, or less when one request holds more than 20 images.Anthropic, Vision
STEP (.step or .stp)A 3D model in the ISO 10303-21 format, which is plain textCode opens the model with a library such as Open CASCADE, which reads STEP files made to AP 214 and AP 203. It can then work out the part's volume. With the material's density, the volume gives the weight to buy.Library of Congress; Open CASCADE
DXFAutoCAD's exchange format for drawings. Plain text, and openly documented by Autodesk.Code reads the file entry by entry. The Library of Congress notes that most entries carry number codes, not names, so the reader must know the codes.Library of Congress
DWGAutoCAD's own binary formatAutodesk has published no specification for it. Ask the customer for a PDF, STEP or DXF export, or license a reader built for DWG.Library of Congress

Which formats a build reads is agreed per project. A large RFQ package is split by part, and each drawing is read on its own. Anthropic notes that dense pages can fill the model's context window before the page limit is reached.

Quote automation on the quote desk, from RFQ to approved quote

Seven steps. Our build used separate agents for intake, matching, pricing, supplier RFQs and reading replies. The steps show how a system like it is usually laid out. What yours covers is agreed per project.

  1. The RFQ arrives. A customer emails an RFQ with drawings and CAD files, or uploads them to a portal. The intake agent saves every file, finds the customer in your CRM or ERP by the sender's address, and opens a job.
  2. Each drawing is read. The agent pulls the title block, the dimensions, the tolerances and the notes from each drawing. Code measures any 3D model. Fields it cannot read cleanly are marked.
  3. Parts and materials are matched. The matching agent looks for the part number in your ERP, then for the customer's own part number, then for past quotes on the same or similar parts. It ties the material on the drawing to a raw material item.
  4. The price is built. The pricing agent works out material, machine time, setup and outside processes from your own rates. It applies only the price breaks and margins your team signed off.
  5. Bought-in parts go out for quotes. Where a part or a process is bought in, an RFQ is drafted to your approved suppliers. A buyer approves it before it goes. Another agent reads the replies.
  6. The draft quote is saved. The quote lands in your ERP or CRM as a draft. Each flag sits on it, with the source of each number.
  7. A person approves. Your estimator checks the draft, clears each flag and approves it. The quote then goes to the customer from your system, the way quotes go out today.

Where each number on the quote comes from

A price for a made part is built from your own data. The agent finds the numbers, and your estimator checks them. The ERP column uses ERPNext, from its guide read on 17 November 2026.

Scroll the table sideways to see every column.

Line on the quoteWhere the number comes fromWhere it lives in ERPNext
MaterialThe material named on the drawing, the stock size to buy, and your price for itA BOM's raw material rate can come from the valuation rate, the last purchase rate or a price list.
Machine timeThe operations a similar past job used, and the minutes each one tookEach BOM operation has a workstation, an hourly rate, a time in minutes and a batch size.
Shop costsWhat it costs to run each machine or work centerA workstation can hold operating costs such as electricity, rent and consumables.
Outside processesPlating, heat treat or coating, priced by your suppliersA Request for Quotation goes to suppliers. Each reply becomes a Supplier Quotation.
Price breaks and marginYour rules by quantity and by customerA Pricing Rule can set a rate, a discount or a margin by item, customer, quantity or date.
The quoteLines, quantities, prices, taxes, validity and termsA Quotation to a Lead or a Customer. It becomes a Sales Order once the customer accepts.

ERPNext's guide says BOM costing gives “an approximate cost” of making an item. That is one reason the estimator approves the price, not the agent.

CPQ software or quote automation from drawings

CPQ software (configure, price, quote) and an agent that reads drawings both end in a quote. They start from different inputs. Vendor facts below were read on 17 November 2026.

Scroll the table sideways to see every column.

QuestionCPQ softwareAn agent that reads drawings
What does it start from?A product catalog and price rules. Salesforce lists “account data, product catalogs, pricing rules” among the inputs.The customer's RFQ, drawings and CAD files.
Who builds the quote?A sales rep picks the options, and the software applies the rules.The agent drafts the lines, and the estimator checks them.
Which products fit?Options on products you already sell. SAP says its CPQ can use existing variant configuration models, plus make, build and engineer to order rules.Parts made to a customer's drawing, often for the first time.
Where is the quote saved?In the CPQ tool, or in the CRM it is part of.As a draft quote in your ERP or CRM.
What does it cover after the quote?Oracle says its CPQ covers the opportunity-to-quote-to-order process, approval workflows included.It stops at the approved quote, plus RFQs to suppliers for bought-in parts.
How is it paid for?A fee set by the vendor.A one-time build, then hosting and AI usage for each document.

When CPQ software is the better buy

Four signs to buy a tool, not build an agent.

  • Your products are configured, not drawn. A buyer picks a model, a size and the options. That is the job CPQ was built for.
  • Requests come in through your own web form. The fields are already structured, so there is no drawing to read.
  • Reps quote live, in front of the customer. Salesforce lists guided selling and bundling among CPQ's common features.
  • Your ERP ships an agent that fits. Business Central's Sales Order Agent drafts sales quotes from emailed PDF, PNG and JPG files, for items already in Business Central.

When an agent that reads drawings fits

Four signs that a tool will leave the hardest step to your estimators.

  • Customers send drawings, not part numbers. The first job on every RFQ is reading the drawing.
  • Most parts are new to you. There is no catalog entry to pick, so someone has to work out the material and the operations.
  • Your rates and past jobs live in your ERP. Copying them into a second tool means keeping two lists in step.
  • You also buy in parts or processes. The same system can send RFQs to suppliers and read their replies.

Tell us what your customers send

Name your ERP, the file types your customers send and the parts you quote most. A person on our team replies by email. The first call is free.

Tell us what customers send

Where the quote is saved, in the systems our team has taken live

Our team has taken nine ERPs live, from SAP to QuickBooks, so the system is your choice. The agent writes to the quote record your team already uses. Each row was read in the vendor's own documents on 17 November 2026.

Scroll the table sideways to see every column.

SystemQuote recordWhat we found
ERPNextQuotation, to a Lead or a CustomerIt stays a Draft until someone submits it. Workflows can add approval steps. Frappe's REST API supports token logins, so the agent can have a login of its own.
Frappe CRMLeads and dealsOpen source, built for teams “with unlimited users.” Its README lists an ERPNext integration that adds invoicing and accounting.
Oracle NetSuiteEstimateAn estimate has no accounting impact until it is converted. Once the customer accepts, it can become a sales order, an invoice or a cash sale.
SAP Business OneQuotationSAP calls it “an offer or proposal” to a customer or a lead. Its Service Layer API has approval calls for quotations.
Odoo 20QuotationOnce the customer confirms it, the quotation turns into a sales order.
Business CentralSales quoteWhen the customer accepts, it is converted to a sales order or a sales invoice.
Sage IntacctQuoteSage's developer guide lists quotes among its Order Entry transactions, with orders and invoices.
QuickBooks OnlineEstimateIntuit's MCP server, a connector that lets AI tools use QuickBooks, can create, read, update and search estimates. It covers 29 record types.
XeroQuoteXero's API specification lists six quote statuses, from DRAFT and SENT to ACCEPTED and INVOICED.
SalesforceQuote, from an opportunitySalesforce says its CPQ and CRM sit on one platform, so a seller moves from opportunity to quote in one place.
HubSpotQuote, tied to a dealThe quotes API can save a quote as an editable draft before it is published to the buyer.
Custom ERPsYour own quote tablesNo vendor API is needed. The agent can work through your database, on a login of its own.

The table covers the products whose documents we read for this page. Our team has also taken SAP S/4HANA and ECC, Dynamics 365 Finance, and Sage X3, 100, 200 and 50 live. Pages with a link below are live.

Where a person approves

Four points where the usual design stops for a person. In our build, a person approved every quote before it went out.

Before a quote reaches a customer

Every draft waits for your estimator, who can change any line before approving it.

When a drawing is unclear

A field the agent cannot read, or a revision that does not match the RFQ, goes to a person with the field marked.

When a part does not match

The line stays unpriced until someone picks the material or the operations.

Before an RFQ reaches a supplier

A buyer checks the items, the quantities and the suppliers to ask.

Where it can, the approval sits in your ERP. ERPNext's workflow guide shows a quotation that goes to a sales manager, then to a regional manager, before it is approved.

What we found when we checked

Notes from our own search data.

Pulled 16 November 2026

Searches for AI quoting software rose from 40 to 18,100, then fell

US Google searches for “ai quoting software” ran at 40 or 50 a month from December 2025 to June 2026. They hit 4,400 in July, 18,100 in August and 12,100 in September, then fell to 2,900 in October. “Quote automation software” jumped from 10 in September to 1,600 in October. The plainer term, “quote automation”, moved between 50 and 390 a month across the same year.

See CPQ against an agent
Saved 14 November 2026

2 of Google's top 10 for AI quoting software were about parts

We saved Google's top 10 US results for “ai quoting software”. A phone app for making quotations ranked first. By their titles and snippets, two results were about quoting parts: a tool for aviation parts at 8, and an instant quote engine for 3D printing at 10. Neither snippet mentioned an ERP.

See where quotes are saved

Questions about AI quote automation

Can an AI agent read our customers' drawings and quote from them?

Yes, as a draft for your estimator. Our team built a multi-agent RFQ system that reads drawings and CAD files and drafts quotes in ERPNext. A person approves every quote. A PDF drawing is read page by page: Claude, one of the AI models we build with, gets each page as an image plus its text. The agent pulls out the part number, revision, material, finish, quantity and notes. It takes each dimension from the number printed on the drawing. It does not measure it off the picture.

Which file types can it take? We get PDF, STEP, DXF and sometimes DWG.

PDF, STEP and DXF can all be read. DWG is the hard one. PDFs and images (JPEG, PNG, GIF or WebP) go to the AI model as they are. STEP and DXF files are plain text, by the Library of Congress's format notes, so code can open them. An open-source library such as Open CASCADE reads STEP files and can work out a part's volume. DWG is AutoCAD's own binary format, and Autodesk has published no specification for it. So for DWG, ask the customer for a PDF, STEP or DXF export, or license a DWG reader. Which formats a build reads is agreed per project.

We already pay for CPQ software. Why would we need this?

You may not. CPQ software (configure, price, quote) applies your product options and price rules to build a quote. Salesforce lists its inputs as structured data, such as product catalogs and pricing rules. That fits a rep who picks options on a product you already sell. A part made to a customer's drawing is often new, with no catalog entry to pick. The agent reads the drawing and drafts the lines. If you keep your CPQ tool, the agent can pass those lines to it through its API, the door one program opens for another.

How does it price a part we have never made?

With your own costing, the way your estimator prices it now. In the usual design, the agent finds similar past jobs and proposes the material, the operations and the minutes for each. Your ERP holds the rates. In ERPNext, a bill of materials (BOM) can list operations, each with a workstation, an hourly rate and a time in minutes. ERPNext works out the operating cost from those and adds the material cost. Plating or heat treat is priced from supplier quotes. Your estimator checks every line before the price is set.

Will the agent send quotes to customers by itself?

No. In our build, a person approved every quote before it went out. The agent saves each quote as a draft in your ERP or CRM. ERPNext's workflows can add approval steps. Its guide gives the example of a quotation of $100,000 or more that must go to a set role to approve. Microsoft built Business Central's Sales Order Agent the same way: designated users “review and approve all outgoing messages.” Where the approval sits is agreed per project.

Will this replace our estimators?

No. The estimator still sets and approves every price. What changes is the typing and the lookups: reading the drawing, finding past jobs and keying in lines. In the usual design, any line the agent cannot match or read cleanly is flagged, and the estimator decides it. Salesforce makes the same point about AI in CPQ. It calls quoting a rules-heavy process, so AI “should not operate on its own.”

We run Business Central. Doesn't Microsoft's Sales Order Agent already do this?

For catalog items, it may, so try it first. Microsoft says the agent reads order details from PDF, PNG and JPG attachments and drafts a sales quote. It finds items by item number, vendor item number, GTIN or cross-reference, then by attributes such as size or material, then by a wider search. Microsoft adds that the quality of your item data affects how well it works. A part made to a new drawing has no item record yet, so there is nothing for it to find. That gap is where a built agent fits.

Which ERPs and CRMs can it write quotes into?

The ones our team has taken live. Our team has taken nine ERPs live, from SAP to QuickBooks, so the system is your choice: SAP, Oracle NetSuite, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage and custom ERPs. On the CRM side, we have set up Salesforce, HubSpot and Frappe CRM. Each one saves a quote its own way. NetSuite and QuickBooks call it an Estimate. SAP Business One and ERPNext call it a Quotation. The table on this page lists each one.

Our drawings are confidential. Will they be used to train the AI?

Not by Anthropic's stated default. A drawing is sent to the AI model's provider to be read. Anthropic says that by default it does not use inputs or outputs from its commercial products, the Anthropic API among them, to train its models (Privacy Center article of 18 August 2026). That is the vendor's policy, not a promise of ours, so ask the same of any provider a build uses. We sign your NDA before you share any data, if you ask.

Some drawings are old scans, turned sideways or marked up by hand. Can it cope?

Partly, and it should flag them. Anthropic warns that its model can make mistakes on images that are low in quality, rotated, or smaller than 200 pixels. It also says the model's answers about where things sit in an image are approximate. So a poor scan goes to a person with the unclear fields marked. In the usual design, a dimension the agent cannot read as printed text is left blank. It is never guessed from the picture.

Can it also send RFQs to our suppliers for bought-in parts or plating?

Yes. That was the second half of our build. Customer quotes came first, then RFQs to suppliers, with separate agents for sending the RFQs and reading the replies. A buyer approves each RFQ before it goes out. When replies land, the prices are read into your ERP for the estimator to use. Our RFQ and RFP automation page covers the supplier side in full.

How long does it take to build?

It depends on your systems and your files. Layer3 Labs, an AI agency, gives two to four weeks for one workflow with one or two integrations. It gives six to twelve weeks for a project of three to six connected automations (guide updated 2 July 2026). Drawing to quote is one workflow. Adding supplier RFQs makes it several. Testing on your past quotes adds time. The timeline for each build is agreed per project.

What does AI quote automation cost?

Four things drive the build cost: the file types, the systems the agent writes into, how your costing data is kept, and how many part families it must quote. Running costs come on top. Anthropic's own example puts a 1,000 by 1,000 pixel image at about $1.30 per thousand images on Claude Haiku 4.5 (read 17 November 2026). Other firms' published build prices are side by side in what an AI agent costs to build.

Can you also put AI on the shop floor, like machine monitoring or camera checks?

Not as a first job for us. Plant-floor AI runs on machine sensor data and camera feeds, which needs different hardware and a different build. Our team's past work sits in the ERP and the office, from RFQs to bookkeeping. If the shop floor is your main need, look for a firm that has shipped it. If quoting is the bottleneck, start with the quote desk.

What do you need from us to start?

Past RFQs with their drawings, and the quotes your team sent for them. Add your item list, BOMs, routings and rate tables, and a login to a test copy of your ERP. In the usual build, those past pairs become the test. The agent's drafts are compared with what your team sent before anything goes live. Tell us the system you run and the parts you quote most. The first call is free.

Sources

Every source behind a figure or a vendor fact on this page, with the day we read it.