Skip to content
All insights

Industry operations

Manufacturers sit between two streams of paperwork.

The work on the shop floor gets the investment. The work that surrounds it, which is reading documents and keying the result into a system, is usually done by people who were hired to do something else.

Published June 2, 20265 min read

What this covers

  • Quoting from a specification is harder than quoting from a catalog, and the split between what an agent prepares and what an engineer decides has to be explicit.
  • Sourcing to support a live quote is the dependency that makes purchasing delays a commercial problem.
  • Three-way matching is where the two document streams finally meet, and it is the most rule-governed step in the business.
  • The same customer part number chaos as distribution applies, with the addition of drawings and revisions.

A manufacturer has two document streams flowing in opposite directions. Customer RFQs, purchase orders, change requests and schedule questions arrive from one side. Supplier quotations, order confirmations, advance shipping notices and invoices arrive from the other. Production happens in the middle, and the administration that surrounds it consumes sales, purchasing and finance in roughly equal measure.

What makes it harder than distribution is that the two streams are coupled. You often cannot answer the customer until the supplier has answered you.

Quoting from a specification

A distributor's RFQ names items that exist in a catalog. A manufacturer's RFQ frequently names a thing that does not exist yet, described by a drawing, a specification sheet, a material callout and a tolerance.

That does not put the workflow out of reach, but it changes the split. An agent extracts the line items, quantities, materials and requested dates from the specification, identifies which elements are standard and which are not, resolves the standard content against your catalog and existing routings, and identifies what has to be bought in. Where a price depends on engineering judgment, it prepares everything around that judgment rather than attempting it.

The engineer should receive a prepared quote with one open question, not a raw request with an attachment.

This is the honest version of AI in engineered products, and it is more useful than the version that claims to price anything. Most of the elapsed time in an engineered quote is not the engineering. It is the waiting before an engineer looks at it, and the assembly afterwards.

The coupling problem

Here is the sequence that costs manufacturers deals, and it is entirely administrative.

  1. 01The RFQ arrives

    With a specification and a date the customer needs an answer by.

  2. 02Something has to be bought in

    A material, a component, a finish. The price and lead time are not known internally.

  3. 03Purchasing is asked

    Usually by email, often without the full specification attached.

  4. 04Suppliers are contacted

    And then chased, because two of the four do not reply.

  5. 05The quote waits

    For days, on the slowest supplier, while the customer's deadline approaches.

  6. 06The answer arrives late

    Or goes out with a padded assumption, which is a margin decision made by a deadline rather than by a person.

Steps three, four and five are the automatable ones, and automating step four alone (the chasing) changes the elapsed time more than anything else in the sequence. Sourcing and purchase order automation covers that workflow directly.

Where the streams meet

Three-way matching is the point at which the customer side and the supplier side finally reconcile, and it is the most rule-governed process in the business: the invoice against the purchase order on price, the goods receipt against both on quantity, within tolerances somebody set.

In manufacturing it carries an extra wrinkle. Partial deliveries against blanket orders, over-deliveries within an agreed percentage, tooling amortized across a first production run, and freight lines that never appeared on the purchase order are all normal rather than exceptional. Each is a rule that exists informally in somebody's head today. Invoice processing automation goes through how those get written down.

Where agents fit, by function

FunctionWorkflowThe agent's part
SalesCustomer RFQs with specificationsExtracts requirements, resolves standard content, identifies what must be sourced, prepares the quote
SalesCustomer purchase ordersValidates against the quote and contract, creates the order, acknowledges
PurchasingSourcing materials and componentsIssues requests, chases, normalizes offers, raises the order once approved
FinanceSupplier invoicesThree-way match, coding, approval routing
Customer serviceOrder status and delivery questionsAnswers from live production and shipment data

Where to start

Manufacturers usually have a clearer answer than distributors, and it is not always quoting. If engineered quotes dominate and each one genuinely needs an engineer, the elapsed-time problem is in purchasing, and sourcing is the better first workflow. If most quotes are repeat or configured work, quoting goes first.

The diagnostic is simple: take ten quotes that went out late and ask what they were waiting on. AI agents for manufacturing sets out the workflows, and choosing your first workflow covers the selection criteria in general.

Questions

Questions we get about this.

Can it quote engineered products?

It prepares them rather than pricing the engineering. It reads the specification, drawings and bills of materials, extracts line items, quantities, materials and dates, resolves the standard content against your catalog and routings, and identifies what has to be bought in. Where a price depends on engineering judgment, that judgment stays with your team.

Does it read drawings and bills of materials?

Yes, including PDF specifications, scanned documents and spreadsheet bills of materials attached to an RFQ. It reads the whole thread rather than a single attachment, which is what catches a revision sent after the original request.

How does it handle partial deliveries and blanket orders?

As rules rather than exceptions. Partial receipts against an open order, over-deliveries within an agreed percentage and tooling amortized across a first run are all normal cases that need a stated tolerance, which a deployment establishes with finance rather than inferring.

Should a manufacturer start with quoting or purchasing?

It depends on what your late quotes were waiting for. Where engineered quotes dominate and the delay is a supplier who has not replied, sourcing is the better first workflow. Where most quotes are repeat or configured work, quoting goes first.

Does it work with our production system as well as the ERP?

It works with the systems that hold the answers it needs, which for status questions usually means production and shipment data as well as the order record. Which systems those are, and how they are reached, is scoped during the review.

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