Skip to content
All case studies

Case study · Manufacturing

Every request arrived differently. Every quote was built by hand.

Quoting a configurable product is not a price lookup. It is a small construction exercise repeated for every request, and it was being done by hand, out of spreadsheets, for requests that arrived in whatever format the customer chose to write them in.

63%

Faster quote preparation

Customer
Glass manufacturer, Hungary
Workflow
Quote requests by email
Built on
AI Quote Agent
Assessment to live
2.5 months
Status
Running in production

At a glance

Measured against the company's own quote preparation process before the agent was deployed. The figure is the customer's, published with their approval. Their name is confidential.

The company manufactures glass. Its customers do not order a catalog item; they describe what they need, and what they need is assembled from a set of choices. That is the whole problem in one sentence, and it is why quoting consumed as much of the commercial team's week as it did.

Why a quote here is not a price lookup

A distributor quoting from a catalog answers a question with a number that already exists. A manufacturer of a configurable product does not have that number. It has a set of options the customer can combine, and a price that follows from the combination they chose.

So every request has to be read, understood as a specification, and then priced from the company's own data. Two requests for the same product in different configurations are two different pieces of work. There is no line to look up.

The product is a set of choices. The quote is what you get when you resolve the customer's choices against your own data.

And the requests arrived in every format

The second half of the problem was the intake. Requests came in by email, written the way each customer happened to write them. There was no form to fill in, no structure to rely on, and no way to know in advance whether the next message would contain a tidy list or a paragraph of prose.

This is the normal state of B2B quoting and it is precisely what defeats conventional automation. A form-based system needs the customer to fill in the form. A scripted integration needs the input to be identical every time. Neither describes an inbox.

What the work looked like before

The team assembled each quote from spreadsheets. The specifications and the pricing data lived in Excel files, and producing a quote meant reading the customer's email, working out what had actually been asked for, finding the relevant figures across those files, and building the quote from them.

Nothing about that is unusual, and nothing about it is cheap. It is accurate work done by people who know the product, and it is slow for exactly that reason: the knowledge is in their heads and the data is in a file they have to go and open.

What the agent does now

  1. 01Reads the request

    Takes the incoming email and works out what is being asked for, whatever form the customer wrote it in.

  2. 02Resolves it against their database

    Matches what was requested to the company's own product and pricing data, rather than to a copy or an exported list.

  3. 03Assembles the quote

    Builds the draft from the resolved data: what was asked for, in the configuration asked for, priced from the company's own figures.

  4. 04Drafts the reply

    Prepares the response as a draft email, ready for a person to read, adjust if needed, and send.

  5. 05Writes and logs it

    The draft goes into the company's own calculation system, and the transaction is logged in the ERP, so the record exists where the business already keeps its records.

What actually changed

Quote preparation is 63% faster than the process it replaced. That is the company's own measurement against their own previous process, published with their approval.

The mechanism behind the number is worth being precise about, because it is not that the agent is quick at arithmetic. It is that the reading, the resolving and the assembling now happen without a person moving between an email and a set of spreadsheets. What is left for the person is the part that needed them: checking the result and deciding to send it.

BeforeNow
Reading the requestA person interprets the emailThe agent reads it, in whatever form it arrived
Finding the dataOpening spreadsheets and locating the figuresResolved against the company's own database
Building the quoteAssembled by hand from those filesAssembled by the agent from the resolved data
Sending itA person writes and sends the replyA person reviews the draft and sends it
The recordWhatever was saved afterwardsIn the calculation system, logged in the ERP

How long it took

Two and a half months, from the first assessment of the workflow through the strategy, the build and the implementation, to an agent running in production.

That is one deployment and not a promise about yours, which depends on your data and your rules. It is stated here because it is a measured fact about this project rather than an estimate, and because it is a more useful number to a buyer than most of the ones this industry quotes.

Why this one worked

Three things, and they are the same three that predict whether any quoting deployment will repay the effort.

  • The volume was real and constant. Quoting was not an occasional task, it was the commercial team's week.
  • The data existed and was reachable. The configuration and pricing information was theirs, in their systems, and could be resolved against rather than copied.
  • The outcome was defined. A quote is either prepared or it is not, which makes both the automation and the measurement unambiguous.

Choosing your first workflow sets out those criteria in general, and from an RFQ to a quote goes through the workflow step by step.

Questions

Questions we get about this deployment.

Does the agent send quotes to customers?

No. It prepares the quote and drafts the reply; a person reads it and sends it. A quote is a commercial commitment, so releasing it stays with the person who owns the customer relationship.

How does it handle requests that arrive in different formats?

That is the point of using an agent rather than a form or a scripted integration. It reads the request as written, in whatever form the customer sent it, and works out what was asked for, instead of requiring the input to be structured in advance.

Where does the pricing come from?

The company's own database, resolved at the moment the quote is prepared, rather than a copied or exported list. That is what makes the output current rather than approximately right.

What does 63% refer to?

Quote preparation time, measured by the customer against their own process before the agent was deployed, and published with their approval. It describes this deployment and is not a projection for another company.

How long did it take to deliver?

Two and a half months from the first workflow assessment through strategy, build and implementation to production. Another project's timeline depends on its data and rules, so this is a fact about this deployment rather than a standard.

See what this would look like in your operation.

A Workflow Review traces one of your real workflows end to end and names the first step an agent could take over. No platform to evaluate first.

Start with one workflow

Fixed scope for the first workflow, agreed before we start. Expansion is your decision once it proves value.

How it works

Security, stated plainly

Where the agents run, what they can and cannot do, and exactly what we do and do not claim.

Security