The quoting process in manufacturing: steps, KPIs and AI agents

The quoting process in manufacturing runs through eight steps from request to order: intake, technical clarification, configuration and bill of materials, cost estimation, pricing, approval, sending and follow-up. Every hand-off between inside sales, applications engineering and cost estimating adds a queue, so a quote often spends more time waiting in inboxes than being worked on.

Before anyone prices a line, it has to be read and matched. A 40-line RFQ means 40 lookups against the item master; configured machines add a check against the configuration rules, and special designs add a round with engineering. For machinery and component manufacturers, the quote desk is therefore a good first workflow, because volume and ownership are clear and turnaround time can be measured from the ERP and the inbox.

An RFQ (request for quotation) asks for price and lead time on a defined scope, an RFP (request for proposal) asks for a proposed solution with a price, and an RFI (request for information) gathers information about possible suppliers. The type of request therefore decides how much engineering and costing the response needs.

In manufacturing, RFQs are the usual form for spare parts, standard components and repeat orders, often as a spreadsheet of lines and quantities. An RFP usually comes with a requirements specification, and the answer is a technical concept with a price that engineering and sales build together. An AI agent classifies the request type on intake, because that decides whether it prepares a draft quote or a requirements checklist for engineering.

Quotes for spare parts and standard products can be prepared largely automatically, because every line has an item in the ERP, a list price and a discount rule. For engineer-to-order (ETO) equipment, where each order involves its own design work, the agent prepares the case and engineering decides.

A spare parts RFQ often lists part numbers from the spare parts list of an installed machine. The agent maps them via the serial number to the right bill of materials, detects superseded parts, checks stock and lead time and drafts the quote, leaving open only the lines it cannot match unambiguously.

Configure-to-order products sit in between, with the agent translating the customer's requirements into characteristic values and the configurator in the ERP or the CPQ system checking whether the combination is valid. For ETO, the agent finds similar past orders, turns the specification into a checklist and flags every deviation from standard, while engineering decides whether the machine can be built that way and what new assemblies will cost.

RFQ automation with an AI agent covers the work between the incoming request and the draft quote. The agent reads, matches, finds comparable cases, checks price rules and creates the draft in the ERP, following the same mechanics as every workflow: intake, context, proposal, release and write-back.

On costing, the agent does not improvise, so lines with a list price get the valid price, and for lines without one it shows the estimated and actual costs of similar past orders with their range, and the cost estimator decides which figure to use. In the quote preparation workflow, price rules and discount floors are enforced inside the workflow rather than merely suggested.

Technical feasibility, pricing strategy and the release of every quote stay with people, and the agent never sends. There is a legal reason as well as a commercial one, because a quote that goes out can be accepted, and under German law Section 145 of the Civil Code binds the sender to an offer unless binding effect was ruled out. Whether a given quote binds you, in Germany or elsewhere, is for your legal team to assess.

In the control set, the agent reads the ERP, CRM, PLM and price lists but writes only draft quotes. In SAP S/4HANA this runs through the standard Sales Quotation (A2X) API, technical name API_SALES_QUOTATION_SRV, and SAP supports an approval step for quotations that cannot be processed without an approver's consent. Discounts beyond the rule, special terms and strategic accounts are flagged to the account owner, and every draft logs the price list, rule set and comparable cases it used.

In shadow mode the agent drafts quotes for live requests while inside sales works as before, and the comparison per request type decides where the agent may prepare more.

CPQ (configure, price, quote) software holds product rules and pricing logic and generates quotes from a configuration that a salesperson selects in the system. An AI agent works one step earlier and turns an unstructured customer request into the structured input that CPQ or an ERP configurator needs.

The two combine well, because with a maintained CPQ system the agent proposes characteristic values from the specification and the email and keeps CPQ as the rule set and price source; without CPQ, it checks against the ERP's configuration rules and price lists. Either way, the rule set decides whether a configuration is valid, not the language model.

CPQ pays off where product rules are stable and sales configures in the system. An agent pays off where requests arrive in inconsistent formats and reading, matching and chasing missing details make up most of the work.

Four KPIs show whether a quoting process gets faster and better: quote turnaround time, hit rate, quotes per person and the variance between estimated and actual cost. Measure them as a baseline before any change and split them by request type, because a spare parts quote and a custom line have different normal values.

Turnaround time on its own misleads, because quoting faster with a wrong estimate wins orders at too thin a margin, so cost variance belongs in the same report. In a Workflow Assessment we take these four figures from the ERP and CRM before any build decision, and the AI ROI calculator gives you a first order of magnitude from requests per month and minutes per request.

Start with the request type that occurs most often and needs the least engineering, which in machinery is often spare parts or repeat orders. There, matching and price checks can be measured in shadow mode without any quote leaving unreleased.

Configured products come next, and only once matching and price rules run reliably does the agent support engineer-to-order with comparable cases and checklists. When the order arrives, the same mechanics continue in order entry automation, because the customer's purchase order is checked against the released quote before the sales order is created in the ERP.