Creating quotes automatically from price requests: how it works with AI

In brief — A price request arrives as a PDF, an email or a bill of quantities. An AI reads it, recognises items, quantities and specifications, assigns each line the price from the price list, framework agreement or quote history, and produces a quote in the company’s own layout — which a person then approves. Line items without a verifiable price are flagged, not guessed. The finished quote is typically ready for approval within minutes. The bottleneck is not the technology but the question of where in the company the pricing logic actually lives.

Which software creates quotes automatically from price requests?

This post explains the method. The product — automated quote generation with process, integrations and FAQ — is on the solution page.

The short answer: a solution that can do three things in one pass. It reads the request, whether it arrives as a PDF, as email text, as an Excel bill of quantities or as a Word document. It breaks the content down into line items — item, quantity, specification, special requirement. And for each line it fetches the price from the sources the company already maintains: the price list, the framework agreement with this customer, earlier quotes sent to them. The result is a quote document in the familiar layout, presented for approval.

The longer answer is that the word “automatically” has to be read honestly in two places. Reading and assembling are automatic. Sending is not, and that is deliberate. A quote is a legally binding document. The final decision is made by a person, at a point where they only check what the machine has flagged as unclear.

Why has quote creation stayed manual for so long?

Because the input is messy. A purchase order has a format, an invoice has mandatory fields, a price request has neither. One customer sends a clean bill of quantities with article numbers. The next writes three sentences in an email: “Same as last time, but 40 units this time and in the 6 mm version.” A third attaches a scanned form on which someone has changed the quantities by hand.

Inside sales solves this today with experience. They know what “same as last time” means, they know the framework agreement, they know that this customer always expects delivery free of charge. Exactly this knowledge is why the task could never be handed to a form-based tool. And it is why it works now: a language model can read the three sentences in the email, and the framework-agreement logic sits in the ERP.

How does automated quote generation work?

Four steps that run in minutes in practice.

1. Read the request. The request arrives through the mailroom or directly via an interface. PDF, email, Excel, Word — the format is secondary. Free text is understood as well. If an AI-powered mailroom sits in front, it recognises the request as such and hands it over directly, without anyone having to route it.

2. Extract line items. The AI recognises requested items, quantities, specifications and special requirements and stores them as a structured list. Every value points to the position in the original document it came from. That is not a side issue; it is the foundation for step four.

3. Assign prices. Each line item is matched against the price list, the framework agreement and historical quotes. If the customer has a special condition, it applies. If there is no verifiable price for an item, it is flagged. The AI does not guess.

4. Generate and present the quote. A quote document in the layout of your own template is created and presented for approval — as PDF, as Excel or as JSON via the API into the CRM or ERP. Sales sees the flagged items first, completes or corrects them, and approves.

Where do the prices come from, and what happens if one is missing?

The most common worry in sales is not that the AI misreads the request. It is that the AI names a price that does not exist. So the rule is simple: prices come exclusively from sources the company maintains itself. The price list. The framework agreement with exactly this customer. What they were offered last time.

If a price is missing for a line — because the item is new or the specification deviates from the price list — the field stays empty and gets a flag. These items form the work list for the human. On a request with forty line items, three of them flagged, inside sales checks three items, not forty.

This is also why quote automation almost always begins with a question about master data: where does the pricing logic really live? If it is in the ERP, connecting it is an integration task. If it is in the heads of two employees and in an Excel file on the network drive, the first step of the project is to write it down. That is not a weakness of the technology. It is the moment a company makes its own knowledge explicit for the first time, and it benefits from that even without AI.

An example: specialist wholesale with item and price data

Take a specialist wholesaler with a few thousand items, volume-tiered prices and framework agreements with the larger customers. Requests arrive daily, a good share of them as emails with the customer’s requirement list attached — sometimes as Excel, sometimes as a PDF from the customer’s own system, sometimes photographed.

Today inside sales reads every list, looks up their own article numbers for the customer’s descriptions, checks the volume tier, checks the framework agreement and types the quote into the template. A request with thirty line items takes a good half hour. At twenty requests a day, that is one full-time equivalent doing nothing but retyping and looking up.

With automated quote generation, the AI reads the requirement list, maps the customer’s descriptions to the company’s own items, pulls tier and contract prices and produces the quote. The lines where the mapping is uncertain — because the customer uses a description that could fit two items, for instance — are flagged. Inside sales checks those, approves, and the half hour becomes a few minutes. Digitally taking over item and price data from the request is not a separate project here; it is steps two and three of the same process.

What matters about this example is what does not happen: no quote is sent without review, no price is invented, and the framework agreement is not overridden because the price list shows a different value.

How is this different from a configurator or a chatbot?

A configurator assumes the customer sticks to the supplier’s structure. They fill in fields, choose from lists. That works when the customer plays along. Price requests, however, arrive in the customer’s structure, not the supplier’s. Automated quote generation turns this around: it takes the request as it comes.

A chatbot answers questions. It can explain what an item costs. But it does not produce a document in the company’s layout that has been checked against the framework agreement and is stored in the CRM as a quote with a number and a validity period. The difference is in the outcome: a quote is a transaction with accountability, not an answer.

Where are the limits?

Three, stated openly.

First, the pricing logic. If it is not documented, it has to be documented first. That is work that comes before the project, and no better model shortens it.

Second, line items that need a professional judgement. When the customer writes “or equivalent”, a person decides what counts as equivalent. The AI can propose; it should not decide.

Third, recognition itself. In production environments, line-item recognition accuracy is above 90 percent. The remaining cases are flagged and land on the work list. Anyone who budgets zero minutes of review for the automated items should be able to justify it — we have written elsewhere about why the automation rate alone says nothing about the savings.

Frequently asked questions

Which software creates quotes automatically from price requests?

An automated quote generation solution reads the request as PDF, email, Excel bill of quantities or Word document, extracts items, quantities and specifications, assigns prices from the price list, framework agreement and historical quotes, and produces a structured quote that is presented for approval. feld.ai offers this as a standalone solution and as a building block of the AI-powered mailroom.

Which request formats can be processed automatically?

PDF requests, email text, Excel bills of quantities and Word documents. Unstructured free-text requests are understood as well and broken down into individual line items.

Where do the prices in an automatically generated quote come from?

From the systems that already exist: price lists, framework agreements with the specific customer and historical quotes. The AI does not invent prices. If it finds no price for a line item, it flags the item instead of guessing a value.

Is the quote sent without approval?

No. The quote is formatted and presented for approval. Before it is sent, sales can adjust every line item. Unclear items are flagged so that the review concentrates on them.

How accurate is the recognition of requested line items?

In production environments, line-item recognition accuracy is above 90 percent. Unclear items are flagged for manual review, and every correction improves recognition for the next request. The quickest way to see how this looks with your own requests is real examples: see automated quote generation or book an intro call directly.