---
title: "AI quoting software: ready-made or custom, costs and limits · Volcanic Minds"
description: "What AI quoting software does, which errors it removes and which it does not, when a ready-made product is enough, when to build one, and the costs."
url: "https://volcanicminds.com/en/guides/ai-quoting-software"
lang: "en"
type: "second_level_page"
updated: "2026-10-30"
alternate: "https://volcanicminds.com/guides/ai-quoting-software"
summary: "AI quoting software reads the request for quotation, with emails, drawings and bills of materials, finds similar past jobs and prepares a draft quote with prices worked out by your rules, which the estimator checks and signs. It cuts the time per quote and the copying and calculation errors, but does not remove them. For machined or sheet metal parts there are ready-made products, paid plans from 59 to 799 € a month; custom software is worth it when you work to order with bills of materials, assemblies and rules of your own, and starts from 10,000 € for the first process."
---

# AI quoting software what it does, what it costs, where it goes wrong

What AI adds to quoting software, which errors it removes and which it does not, when a ready-made product is enough and when it has to be built, with the products' price lists and an example on a machine shop's requests for quotation.

## What does AI quoting software do?

It reads what arrives in free form, searches your history and prepares a draft quote; your rules work out the price, and a person signs it. Traditional quoting software, even a good Excel sheet, does the sum once someone has given it the data: material, quantity, times, operations. AI works before that, where today an estimator opens an email, reads a PDF drawing, copies dimensions and material, and tries to remember a similar part made two years ago.

| Step | Who does it | What it needs |
| --- | --- | --- |
| read the request: email, attachments, drawing, bill of materials | the language model | a readable drawing: a skewed scan reads worse |
| extract the data: material, dimensions, quantity, tolerances, finishes, revision | the model, which flags uncertain fields | the list of data a quote must have |
| find similar jobs, with the times and costs they really had | a search of the history, by code and by description | a history with actual costs, not only with quotes |
| work out cost and price: material, setup, machine time, outsourced operations, margin | your rules, in tables you can read | rates, price lists and markups kept up to date, in one place |
| check the draft, correct it, sign it | the estimator | a draft that shows where every figure comes from |
| once the job is closed, compare the quote with the actual cost | the management system, then whoever prepares quotes | the hours and costs recorded on the job |

The step that decides the quality of quoting software is the fourth. Ask a language model ([LLM](/glossary#llm)) to estimate a price and it gives you a plausible figure, but it cannot say where the figure comes from, and the next time you send the same request it may give you another. A calculation with your rates always gives the same figure, and when it is wrong you find the line that got it wrong. **In the quoting software we build, the model reads and proposes, the price comes from rules you can read, and a person signs the draft.**

In the guide [AI agents for SMEs](/guides/ai-agents-for-smes) the request for quotation becomes a file for whoever sets the price: part data, similar jobs, questions to ask the customer. Quoting software goes one step further and puts a draft price in the file too.

## Can AI quoting software produce quotes without errors?

No. It removes almost all copying and calculation errors, brings new ones, reading errors, and leaves estimating and judgement errors where they are.

Ask an AI assistant and the answer changes from one to the next. On 30 October 2026, to the question of a 25-person engineering company that works to order, Google's AI Overview answered that “human error is reduced to zero” and AI Mode that errors are “almost totally eliminated”, both in Italian; ChatGPT and Perplexity answered that they are not eliminated. The second pair is right, and the difference lies in the type of error.

| Error | Example | With quoting software |
| --- | --- | --- |
| copying | a quantity copied wrong, two codes swapped | almost disappears: the data goes from the document to the calculation without being retyped |
| calculation | an Excel formula that skips the last row, a markup applied twice | disappears if the rules sit in one place and are tested |
| reading | a 3 read as an 8 in a scan, the wrong revision of the drawing, a tolerance written in a note | is new: it is contained by flagging uncertain fields and comparing with the history |
| estimate | optimistic cycle times, a forgotten setup, a material that went up since the last price list | stays: only the comparison with the actual cost finds it |
| judgement | a customer worth taking at a low margin, a machine that is fully booked next month | stays with the estimator: the software does not know your factory |

Reading errors are the new ones, and you have to look for them on purpose. They are contained by checks the calculation runs on its own: a quantity out of scale for that customer, a material you do not machine, a weight that does not match the dimensions. The draft shows them at the top, and the estimator looks at those first. **Quoting software that does not say which data it read with low confidence is more dangerous than an Excel sheet, because it produces drafts that look right and are not.**

## Why does quoting software need actual costs?

Because software that learns from past quotes also learns their mistakes. If for years you quoted 40 minutes for a part that takes 55 in the shop, the history says 40, and the draft will say 40 with confidence. Only the comparison with the hours and costs recorded on the job tells you which of the two numbers is right.

It is the point you read least about: among the 157 pages offered by Google and the AI assistants that we read on 30 October 2026, only one gives a section to comparing quotes with actual costs. Yet the largest gain is there: more than quoting quickly, no longer quoting below cost.

**If today you do not have actual costs, start there: without them, you know the software prices quickly, but not whether it prices right.** How to record the hours and costs of a job, and how often to look at them, is explained in the guide [Job costing in the shop](/guides/machine-shop-job-costing).

## Ready-made product or custom quoting software?

It depends on what you sell. If you make parts to drawing, machined or cut and bent from sheet metal, ready-made products read the drawing or the 3D model, estimate the times and work out the cost with your rates. If you work to order, with machines or plants made of multi-level bills of materials, assemblies, catalogue components and variants, a product designed for the single part is not enough.

The AI assistants' answers we collected on 30 October 2026 all advise a 25-person company to start from a ready-made product, and the products they name almost all start from the drawing or the model of a machined part. For those who build machines or plants to order, that advice only goes halfway.

| What you sell | What arrives | The route |
| --- | --- | --- |
| parts to drawing, machined or from sheet metal | PDF drawing or 3D model, material, quantity | a ready-made product, tested on your drawings |
| parts to drawing, with a management system the order must go through | the same, plus customer and job records in the management system | a ready-made product, plus a connection built as a project |
| machines, plants or assemblies to order | specifications, multi-level bills of materials, variants, components from several suppliers | custom, on top of your price lists and your management system |
| catalogue products with options | requests with codes and configurations | a configurator (CPQ) or the feature of the management system you already have |

Choose a product by how well it follows the way you set prices. Before buying it, test it on 20-30 real requests that are already closed, comparing its draft with the estimator's price and with the actual cost, and ask the supplier:

- whether you can load your own rates, machine by machine, and your setup times;
- how it handles outsourced operations, treatments and assemblies;
- whether it shows where every figure in the draft comes from;
- how an accepted quote reaches your management system;
- where your customers' drawings end up, since they are often covered by a non-disclosure agreement.

Custom software is worth it when no product passes the test: the case of machines and plants built to order, or of pricing rules that are your edge over competitors. Custom does not mean a model trained from scratch: the [models that already exist](/frequently-asked-questions#modelli-ai) do the reading, and around them we build your rules, the search in your history and the connection to the management system. How to choose between a product and custom software in general is explained in the guide [AI in business: where to start](/guides/start-with-ai-in-business). **If a ready-made product can do your quotes, we tell you, even if it means a smaller project.**

## What do ready-made products cost?

Those with a price list in euros cost from 59 to 799 € a month, excluding VAT, and you pay by users, by quotes or by archived parts. Paperless Parts, the one AI assistants name most, does not publish its price. The table collects the products named in the answers of 30 October 2026: it is a sample, and the order is not a ranking.

| Product | What it reads | How you pay | Price list on 30 October 2026 |
| --- | --- | --- | --- |
| RapidMech, Italian | the technical drawing, PDF included; finds similar parts in the archive | by users | 199 € a month for one user, 299 € for up to 3, 549 € for up to 10; the connection to the management system is a project |
| Fabriqer, Italian | the 3D model (STEP, SolidWorks, CATIA) | by archived parts | free up to 5 parts; from 59 € a month for up to 100 parts to 449 € for up to 100,000, billed annually |
| QuoteBuddy | the technical drawing | by quotes per month | 149 € for 100 quotes, 399 € for 500, 799 € for 1,500 |
| Toolpath | the 3D model, for machining | by seat | $40 a month per seat on the Core plan, $125 on Pro, billed annually |
| CloudNC Quote Agent | the request by email, 2D drawings and 3D models | by quoted part | $2.50 per part, with volume discounts |
| Paperless Parts | the CAD model, with custom pricing logic | on request | not published |

Be careful with the prices the assistants quote: Perplexity gave Toolpath $400 and $1,500 a month, which are the plans for those using its [APIs](/glossary#api), and Google's AI Overview gave QuoteBuddy a price per user, while it is paid by quotes. And the subscription is not the whole cost. Calibrating rates and times on your machines, loading the history and connecting the management system is work for you or for whoever helps you, and the connection, where there is one, is built as a project.

## What does custom quoting software cost?

From 10,000 € for the first process, plus model usage, after a feasibility study that costs up to 2,500 €, deducted from the pilot if you sign. These are Volcanic Minds' entry thresholds: the price of your project is written by the study, with a range and a maximum. For the same requirements, the price grows with these items.

| Item | Costs less when | Costs more when |
| --- | --- | --- |
| what arrives | native PDFs, always from the same customers | scans, photos, a different format for each customer |
| pricing rules | already written in a sheet everyone uses | in the heads of one or two people |
| history | jobs with actual costs in a management system | quotes scattered in folders, without real costs |
| product | single parts or similar families | multi-level bills of materials, variants, assemblies |
| connections | a management system with APIs | a closed management system, several systems to keep aligned |
| where the drawings live | cloud services that exclude training on your data | a model on your own servers |

Ask an AI assistant for the figure and you get anything: on 30 October 2026 the answers ranged from 3,000 € to over $150,000, and Google's AI Mode, asked the same question, first wrote 15,000-50,000 € and more, then 5,000-10,000 € for an AI module. None cited a primary source.

Model usage, on the other hand, costs little. An estimate, with our own assumptions: a request with an email and a two-page PDF drawing, plus the instructions and the similar jobs found in the history, comes to about 20,000 input [tokens](/glossary#token) and 2,000 output tokens. Anthropic estimates 1,500 to 3,000 text tokens per PDF page, plus the tokens of the page image. With Claude Sonnet 5.5, at 2 dollars per million input tokens and 10 per million output tokens at 30 October 2026 prices, that is 6 cents of a dollar per pass. With three passes per request and the thirty requests a week of the example further on, about 130 a month, it is less than 25 dollars a month. The real number is measured in the pilot as cost per request. **In custom quoting software you mostly pay for the rules and the connections; the model costs little.**

Then there is hosting and, if you want it, maintenance, which is a separate contract paid on the work done, as the guide on [maintenance](/guides/custom-software-maintenance) explains. There is no licence fee on the software we develop for you.

> **The figures in this guide**
>
> Prices and durations are those of Volcanic Minds, current at the date at the top of the page. The products' price lists are those published by the suppliers and read on 30 October 2026. The numbers in the request for quotation example and the model usage cost are estimates with declared assumptions, to be redone with yours.

## How much time does it really save?

On requests that resemble jobs already done, most of it; on the others, little. Take a shop that receives about thirty requests for quotation a week by email, with PDF drawings, material and quantity, and where an estimator spends about an hour on each. The AI assistants we gave the case to estimate, with quoting software, 5 to 20 minutes of checking per request and 20-25 hours freed a week. They are estimates with no measurement behind them.

An example with hypothetical numbers. Of the thirty requests, twenty resemble parts already made, and the estimator checks the draft in a quarter of an hour; the other ten, new parts or ambiguous drawings, take an hour as they do today.

| Item | Today | With quoting software, under the assumptions |
| --- | --- | --- |
| requests a week | 30 | 30 |
| work per request | one hour | 15 minutes for 20 similar requests, one hour for 10 new ones |
| time a week | 30 hours | 15 hours |

These are assumptions to show the calculation. The cases published by suppliers speak of days becoming hours: in [AMFG's Solmeto case](https://amfg.ai/case-studies/from-3-days-to-2-hours-solmeto-transforms-quoting-with-amfg) three days became two hours on 60-70% of requests. They are cases the supplier chose, and only the test on closed requests gives you your own numbers. The time freed, then, is worth something only if you use it: to answer the same day instead of after three days, or to look at the actual costs nobody looks at today.

## Where do you start?

With the pricing rules, before the software. Among Italian companies with at least 10 employees that considered AI without adopting it, 45.2% name the unavailability or poor quality of data as an obstacle ([ISTAT, Enterprises and ICT 2025](https://www.istat.it/comunicato-stampa/imprese-e-ict-anno-2025/), in Italian). For quoting software the data is the rates, the price lists and the history, and if they are in disorder no product puts them back in order.

1. Measure the starting point: how many requests arrive, how long each takes, how many you win, how many jobs close above the quoted cost.
2. Write the pricing rules in one place: rates per machine, setups, material price lists, outsourced operations, markups, minimum order quantities. It is needed whatever the route, ready-made or custom.
3. Test on closed requests: 30 to 50 quotes with their actual cost, comparing the draft with the estimator's price and with the real cost.
4. Run in parallel for a few weeks, with no automatic sending, and measure the share of drafts signed without changes and the time spent checking.

If the route is custom, the phases and prices are those of the guide [AI in business: where to start](/guides/start-with-ai-in-business). The feasibility study, 1-2 weeks and up to 2,500 €, says whether a ready-made product is enough or custom software is needed, how to connect to the management system, what model usage costs and the criterion for stopping. **The study stays yours even if you do not sign, and if you sign its cost is deducted from the pilot.** The pilot, 4-6 weeks from 10,000 €, puts the quoting software to work on one family of parts or jobs and measures the result against the starting point.

We applied the principle of rules in one place, without AI, in the [quoting for bicycle workshops](/case-studies/mechanical-workshop-management): the operator builds the offer from hours and parts in the bill of materials, the system lays it out as a PDF and the work order is followed in the same place. On a whole order, from quote to workshop job, time went down by 40%, according to internal estimates and measurements made together with the client.

**Would you rather talk it through?**

Book a call with people who work on these topics every day: we will look at your case together.

- [Book a call](https://vms.srl/booking)

## When is AI quoting software not needed?

When requests already arrive as data, when they are few, or when the problem is the rules. In the first case, catalogue codes and quantities, a price list in the management system or a configurator is enough, with no model reading anything. In the second, do the sums first: calibrating rates and times only pays off if there are enough requests to free real hours.

In the third, AI would only make the same mistakes faster. If two estimators give two different prices for the same part, the software does not know which to follow, and an agreed rule comes first: it is the [clean data](/insights/data-quality-why-ai-fails-on-dirty-data) every AI project needs. If the rules live in three Excel sheets nobody dares to touch, the guide [Excel or business software](/guides/excel-or-business-software) says when it is worth leaving them.

**Many processes need a well built form and a clean database, not a new system, and in that case we tell you so.**

## Frequently asked questions

### Does AI quoting software replace the estimator?

No. The software reads the request, finds similar jobs and prepares the draft with the price worked out by your rules; the estimator checks it, corrects it and signs it. Their work changes, from copying data to checking drafts and deciding on new cases, margins and customers.

### Can AI set the price on its own?

It is not advisable. A language model gives plausible prices, but cannot say where they come from and may give different ones for the same request. It is better for the model to read the documents and for your rates and price lists to work out the price, in rules you can read, with a person approving before sending.

### How much does AI quoting software cost?

Ready-made products with a price list in euros cost from 59 to 799 € a month, excluding VAT, according to the price lists read on 30 October 2026, and are paid by users, by quotes or by archived parts. Custom quoting software from Volcanic Minds starts from 10,000 € for the first process, plus model usage, after a feasibility study that costs up to 2,500 €.

### Does it need the 3D model or is the PDF drawing enough?

It depends on the product, because some start from the 3D model and others also read the PDF drawing. The 3D model gives the exact geometry; the drawing carries tolerances, finishes and notes the model often lacks. When both exist both are needed, and a damaged scan has to be flagged as uncertain data.

### Do my customers' drawings stay confidential?

It depends on where they are read. With a ready-made product, ask the supplier where the files end up and whether they are used to train models. In the systems we build we choose Enterprise-grade cloud services that contractually exclude the use of your data for training, and Zero Data Retention services when more is needed; if the data is particularly sensitive we consider a private AI infrastructure.

### Does the quoting software connect to my management system?

A ready-made product usually exports the quote, and a direct connection, where there is one, is built as a project. Custom quoting software connects to the management system you have, through APIs, periodic imports or the other routes the system allows, so the accepted quote becomes a job without being retyped.

### Does the quoting software learn from past quotes?

Yes, but it also learns their mistakes. If the history has quoted times that are always lower than the real ones, the draft repeats them with confidence. That is why the software is tested and corrected on actual costs, the hours and costs recorded once the job is closed, not only on quotes.

### Who is liable for a wrong quote, and what does the AI Act ask?

The company that sends the quote is liable, as it is today, which is why a person signs the draft. Under the AI Act, whoever uses quoting software is a deployer; a system that prepares quotes for other businesses is not among the high-risk uses of Annex III, and what remains is the obligation of article 4, in force since 2 February 2025 and rewritten by Regulation (EU) 2026/1744, which asks for training measures for those who use it without imposing a level. The details are in the guide [on the AI Act obligations](/guides/ai-act-deployer-obligations).

## Quoting software that prices with your rules?

Tell us what you sell, how requests arrive and how you set prices today. The feasibility study tells you whether a ready-made product is enough or custom software is needed, what it costs and the criterion for stopping.

- [Let's talk about your project](/en/contacts)

## Articles on the same topics.

*Keep reading*

- [Web Bot Auth: visitors with no eyes](/en/insights/web-bot-auth-agentic-traffic.md): AI agents read your site and convert better than organic, yet they never show up in analytics. How to measure them and verify them with Web Bot Auth.
- [Generative UI: for humans and machines](/en/insights/generative-ui-agents-dual-mode.md): Explore how Generative UI and Dual-Interaction UX transform enterprise SaaS by creating adaptive interfaces for human operators and AI agents.
- [Data quality: data decides, not the model](/en/insights/data-quality-why-ai-fails-on-dirty-data.md): When AI gets it wrong, blame the data, not the model. How to identify, sanitize and refresh your sources so AI actually works in your next business project.