How to automate spare parts quote requests without quoting the wrong part

Where quote time actually goes

Open the parts@ mailbox at a machinery maker on a Monday and the requests only look alike from a distance. One customer sends a serial number and a part number from a 2014 parts book. Another writes "the blue seal on the gearbox, machine is down" and attaches a phone photo with the nameplate half painted over. Each one needs a quote, and the customer with the stopped line needs it before lunch.

Producing the quote document is the quick part. Most ERPs, whether SAP S/4HANA, Dynamics 365, Infor, Epicor or IFS, produce a priced quote from a part number and a customer in a few clicks. The minutes go into everything before that: working out which part is meant for this serial, following the supersession chain to the current number, finding the customer's contract price, checking stock and lead time, screening the order and writing the reply.

That split matters for automation. Form filling in the ERP is already fast. Identification and screening are where an agent saves the most time, and they are also where a careless setup quotes the wrong part to the wrong customer.

It matters for revenue too. Bain's 2012 study of industrial goods makers found that spare parts still made up more than half of service revenue, and that companies typically captured only 10% to 25% of the service potential of their installed base. McKinsey's 2017 analysis across 30 industries put the average EBIT margin of aftermarket services at 25%, against 10% for new equipment. The same analysis found that OEMs lose aftermarket business once warranties and initial contracts expire, often to third-party parts makers and independent service providers.

Speed is part of that contest. In a case study published by InstaLILY, a software vendor, the chief commercial officer of a US industrial automation distributor said a reply within an hour gives the best chance of winning the order, and that after four or five hours it is hardly worth replying. Treat it as one executive's view at a distributor rather than a measured result. When McKinsey surveyed 250 senior buyers of aftermarket services in North America and Europe in May 2024, poor spare parts availability ranked among their top three concerns, and fewer than 30% said they used their suppliers' digital offerings. If your requests arrive by email, start with the shared mailbox.

What to automate and what to keep with the parts specialist

A workable rule: let the agent do what a careful new hire could do in their first month from your parts catalog and the ERP, and keep anything that needs judgment with a person. In practice the line falls roughly here.

The hard cases are the ones where the data runs out. Think of a serial whose as-built BOM exists in two variants, or a casting number that matches several sales part numbers. The agent's job there is to show the candidate matches with the evidence for each. Unresolved requests go to a specialist. It should never pick the most likely part and quote it with a confidence it does not have.

Fix supersession and serial data first

The biggest accuracy gain usually comes from cleaning the data the agent will read, before any AI touches a quote. Most equipment makers have years of supersession records typed by different people, with a one-way replacement recorded as two-way here and a kit that replaced several loose parts but was never linked to them there. An agent reading messy tables will produce wrong quotes, faster.

The second link is serial to BOM. A parts desk can only identify the right part for a 2016 machine if someone can see what that serial shipped with. Some makers hold the as-built BOM by serial in the ERP. Others have it in scanned build records or only in the memory of a senior specialist. Without it, the agent can still narrow the candidates from the model and the photo, but the share of requests it can draft with confidence drops.

Once both are in order, most repeat requests become a lookup plus a stock check. Give one named person ownership of the supersession data. When the agent meets a number it cannot resolve, it proposes the link and that person approves it. That small gate stops one wrong link from turning up on every future quote for that part.

Photos help, within limits. DB Cargo described how even experienced staff in its rail maintenance plants sometimes needed 10 to 20 minutes or more per part to find the right material number, because parts come back worn and soiled. Visual search can rank candidates from a photo, but a worn part seldom gives a single answer. Treat a photo match as one piece of evidence next to the serial, and never as the identification on its own.

The approval queue and the controls that protect you

For the first months, every quote the agent prepares should land in an approval queue rather than the customer's inbox. The specialist sees the original request, the part the agent identified and why, the supersession steps, the price source and the screening result, then releases it or sends it back. Machine-down requests go to the top of the queue with a flag, so the draft that matters most is read first.

Export control belongs inside the workflow. Under the US Export Administration Regulations, a part for use in or with an item on the Commerce Control List can be specially designed, subject to the releases in 15 CFR 772.1, such as the release for fasteners and washers. Trade compliance has to classify each part and each service document under the applicable EAR provisions. The agent reads the approved ECCN (export control classification number) or EAR99 classification from your item master and holds the request when it is missing. It also screens every party on each draft, including the end user, against the US Consolidated Screening List, which brings together the restricted party lists of several federal agencies.

One US date matters for the workflow design. The BIS rule that extends Entity List restrictions to companies owned 50 percent or more by listed parties is suspended until 9 November 2026 and returns on 10 November unless BIS publishes an extension. From then on, screening by name alone may not be enough, so the workflow needs ownership data or a documented hold for unknown owners.

European exporters have their own rule. Article 12g of Council Regulation 833/2014 requires a contractual ban on re-export to Russia when certain goods, including bearings, are sold to a country outside the regulation's partner list. The agent inserts your approved clause from a template and records the check, so the clause text always comes from your legal team.

This is general information, not legal advice. Have your legal and trade compliance teams confirm the requirements that apply to your products and markets.

These are the controls worth insisting on before go-live:

• Each draft records how it identified the part and where the price came from.
• Prices and lead times come only from the ERP, and anything below the price floor goes to a manager.
• Part numbers come only from approved records, and an unresolved number goes to a specialist with the candidates listed.
• Screening runs on every quote, and a hit or an unknown owner holds the quote for trade compliance.
• Export classification is read from the item master, and a missing field blocks the quote.
• A log shows who released each quote and when, with retention periods set by your counsel.

Worked example: what the hours look like

This is an illustrative example with assumed numbers, not a client result. Take a 900-person machinery maker with a parts desk of six people handling 350 inquiries a week. Each takes 18 minutes from opening the email to sending the quote: about 8 minutes identifying the part and 10 on price, stock, screening and the reply. That is 6,300 minutes, or 105 staff hours a week.

Now assume 60% of inquiries, 210 a week, are routine: a serial or a current part number that matches clean BOM and supersession data. With an agent preparing the draft, the specialist spends about 4 minutes reviewing and releasing each one, so 840 minutes. The other 140 still take the full 18 minutes, another 2,520 minutes. Total time falls to 3,360 minutes, or 56 hours, which gives back 49 staff hours a week.

The routine share depends almost entirely on the state of your supersession and serial data, so the first week goes into counting requests by type. Time to first quote is the second number to watch. In this example we assume it moves from a same-day reply to under an hour on routine requests, and that is the number your customers notice.

Starting small with parts quotes

Pick one product line and one ordinary week. Have the parts desk tag every inquiry as routine or not and note the minutes it took. As an illustrative starting sample, pull the supersession records for the 200 parts you quote most and check them with a senior specialist. Then run an agent against copies of last month's inquiries with nothing going to customers, and compare its drafts with the quotes your team actually sent.

That comparison becomes the acceptance test. With us it goes into the agreement before we start, with criteria such as the share of routine drafts identified correctly and the share released without edits. We build the workflow in your environment in two weeks, and you pay $10,000 for it only after every criterion passes. Running it afterwards starts at $1,200 a month. Whoever you work with, insist on a test against your own inquiries before a single quote reaches a customer.

Common questions