Technical Machinery Enquiries: AI Chat With Sources

AI assistant in mechanical engineering: answer technical questions on nameplates, spare parts and documentation with a source, not a guess. How to plan it.

In plain language

For questions about a machine, the exact machine comes first. Without type, serial number and year, any answer about parts is a guess. So the assistant asks for these three things first. Business buyers spend only a short part of their buying time with vendors. If you have no answer then, you drop out. Often a customer finds nothing online that fits the question. The firm has to keep its manuals online for a long time anyway, and the assistant can learn from them. Approvals, safety checks and finding faults belong with a person.

August 31, 2026 13 min read By XICBOT

Key points

Technical enquiries in mechanical engineering hinge on identification. Without type, serial number and year of manufacture, any spare parts answer is guesswork. The assistant settles those three facts before it talks about parts.
77 percent (Gartner) of business buyers describe their last purchase as very complex or difficult, and only 17 percent (Gartner) of the entire buying journey is spent meeting suppliers at all. Anyone who cannot answer within that narrow window drops out of the shortlist.
Self-service fails on data, not on willingness: only 14 percent (Gartner) of service issues are fully resolved without contact, and in 43 percent (Gartner) of cases the customer finds no content that fits the question.
Machinery Regulation (EU) 2023/1230 requires the operating manual to stay online for at least ten years (Machinery Regulation). Manufacturers who have to keep those documents anyway already hold the raw material for the knowledge base.
Identification, documentation lookups and availability belong in the chat. Approvals, safety assessments and fault diagnosis belong with people: 75 percent (Gartner) of business buyers will prefer to speak to a human on complex matters by 2030.

Why technical enquiries in machinery work differently

In retail, an article number identifies the product and the only open question is whether it is in stock. In mechanical engineering, the article number identifies the series at best. What actually stands on the shop floor only emerges from the serial number, the year of manufacture, the variant and the question of which modifications have accumulated over fifteen operating years. An answer that skips this chain is not an answer, it is a guess with a number attached. That is exactly why boilerplate replies from a help page carry so little weight here: they answer the question about the series while the caller is asking about a machine.

Then there is the time pressure. Business buyers spend only 17 percent (Gartner) of the entire buying journey meeting suppliers, and any single supplier representative gets 5 to 6 percent (Gartner) of their time. Service is just as tight: 73 percent (Gartner) of customers try to solve the issue themselves first, but only 14 percent (Gartner) of issues are fully resolved that way. The reason is rarely a lack of effort, it is a missing source: in 43 percent (Gartner) of cases the customer finds no content that matches the question. Where the time actually disappears in your own operation only shows up in a review of the conversations, not in the service manager's gut feeling.

What the assistant answers from the machine documentation

Not every question from the plant belongs to an assistant. The dividing line runs where an answer turns into an assessment. As long as the answer sits in a document the manufacturer maintains anyway, it is a case for the chat. As soon as somebody has to decide whether a machine may keep running with a particular component, it is a case for a person with access to that machine. Inside that boundary, plenty remains to take real load off the phones:

Identification: determine the specific machine from type, serial number and year of manufacture, and name the documents that belong to it
Spare parts lookup: which assembly belongs to this variant, which part number is listed for it and whether a successor part exists
Maintenance intervals: which service scope is due by operating hours or calendar time, and which consumables it requires
Operating fluids: approved oils, greases and coolants including fill quantities, with a reference to the chapter of the operating manual
Tightening torques and settings: numerical values that sit in the documentation and otherwise get looked up in a binder during the call
Document delivery: operating manual, wiring diagram or spare parts catalogue in the right language version and for the right build state
Order status: whether a spare part has been ordered, reserved or shipped, provided the ERP system is connected in the Custom package and the assistant is allowed to read it

The difference from a search function lies in the final step. A search returns the document, the assistant returns the passage. Someone asking about the oil change interval gets the value, the unit and the chapter it comes from, not a three-hundred-page appendix. For that to hold, the knowledge base has to be separated by machine and build state rather than by product category. How such a knowledge base is built decides the hit rate far more than the language model behind it.

Nameplate first

No spare parts answer without type, serial number and year of manufacture. The assistant asks for all three and accepts a manually typed plate.
Answer with a source

Every technical figure arrives with a chapter, section or page number. Anyone who wants to read the source does not have to call.
Build state separates variants

The same series carries several variants over ten years. Documents are selected by build state, not by product name.
Language of the enquiry

The question comes from an export market. The assistant replies in the language of the enquiry and points to the language version the customer is entitled to.
Stock from the system

Availability and lead time are read from the ERP system instead of being recited from a maintained text block.
Handover with a case

If the question turns into a service case, the entire conversation and the machine data go to the responsible colleague.

From the nameplate to the right spare part

The most common call in spare parts sales starts with a photographed nameplate and ends with somebody leafing through a catalogue. That path can be put in order without letting an automated system place the order at the end. The assistant runs the identification, proposes the listed part and opens the case. The order itself remains a step a person confirms. If the enquiry turns into a quotation, the same separation applies as for any other price request in the chat.

Capture the identification: record type, serial number and year of manufacture, and ask for the order or delivery note number when the plate is illegible
Resolve the machine file: determine the build state and load the documents that belong to this exact machine
Narrow down the assembly: search by the function the caller describes instead of a part name they do not know
State the part number: return the listed number with its designation and chapter reference, and the successor part for discontinued items
Read availability: retrieve stock and lead time from the ERP system, with a timestamp rather than a blanket promise
Hand over the case: pass the enquiry with machine data, part number and quantity to spare parts sales, where it is checked and confirmed

Access to stock and case status is the point where a lookup path becomes a tool. Technically this happens through a controlled connection to the existing systems in the Custom package: the assistant may issue defined queries and receives defined fields in return. What it may not query, it does not see. What this permission model looks like in detail, and which mistakes get expensive, is described in the article on function calls and tool permissions.

A part number is information, not an approval

An assistant that gives a part number says: this part is listed for this machine. It does not say: this part solves your problem. Between the two lies a fault diagnosis, and that belongs to a person with access to the plant. This boundary belongs inside the answer itself, not in a footnote below the chat window.

Export means the enquiry arrives in another language

German mechanical engineering exports machinery worth 199 billion euros (VDMA) and accounts for 14.0 percent (VDMA) of world machinery exports. Every one of those machines stands somewhere where maintenance does not speak German. For service that means the enquiry arrives in the evening from another time zone, in a language no shift covers, and it concerns a machine delivered eight years ago.

Machinery Regulation (EU) 2023/1230 shifts the situation in a helpful direction. It requires the operating manual to be drawn up in a language determined by the member state concerned and easily understood by users, to remain accessible online for at least ten years (Machinery Regulation) after the machine is placed on the market, and to be supplied free of charge in paper form within one month (Machinery Regulation) on request. The regulation applies from 14 January 2027 (Machinery Regulation). Manufacturers who have to build up this document base anyway already hold the raw material for a multilingual assistant. How language versions can be kept apart without a translation quietly changing a numerical value is covered in the article on multilingual assistants in international business.

Type of enquiry	Phone and email today	Assistant with the machine file
Identifying the machine	Follow-up question about the nameplate, often two contacts	Asked in the first step of the conversation
Spare parts lookup	Leafing through the catalogue, depends on the agent	Listed part number with a chapter reference
Fluids and settings	Search in the operating manual	Value with unit and source
Language	Tied to the shift on duty	In the language of the enquiry, around the clock
Availability	Callback after a look in the system	Stock and lead time with a timestamp
Fault on a running machine	Service technician, often without prior information	Direct handover including machine data

When the line stops: escalation and handover

A stoppage is not a lookup question. Someone writing that the press has been down for twenty minutes does not need an assistant asking three more questions. They need the shortest route to a person who knows the machine. That is why every configuration needs a rule that gets the assistant out of the way: certain terms, a safety reference or an explicit request for a callback end the lookup and open the route into the service area.

The most expensive service case is the one where the technician arrives with the wrong part because nobody asked for the serial number on the phone. If it is in the case file, two site visits become one.

Project experience from service and spare parts projects

The handover decides the impression the assistant leaves behind. 75 percent (Gartner) of business buyers will prefer human contact for complex matters by 2030, and 62 percent (Bitkom) of customers want to turn to a quickly reachable human contact when something goes wrong. At the same time, 36 percent (Bitkom) explicitly want automated help for simple questions. Together the two describe the job precisely: answer quickly where the answer sits in the documentation, and hand over without detours where it does not. What a clean transition looks like, without the customer repeating their details, is set out in the article on handover to human agents. And what happens when the assistant itself goes down is covered in the article on fallback and failover.

Where the assistant deliberately stays silent

In mechanical engineering, a wrong answer costs more than an unhappy customer. The configuration therefore needs a list of topics where the assistant does not answer but forwards: safety functions and guards, modifications and retrofits, conformity questions about the machine, warranty and goodwill, binding delivery dates, and anything to do with a fault on a running machine. That list is not a sign of weakness, it is the reason the remaining answers can be trusted. It is backed up by a knowledge base that ties the assistant to stored content. The approach and its pitfalls are described in the article on hallucinations and the knowledge base.

A few steps to an assistant on the shop floor

The effort rarely sits in the assistant, it sits in putting the documents in order. Among companies in Germany that use AI or plan to, relief from routine work is the most frequently named goal: 84.5 percent (German Economic Institute) state it, and 37 percent (German Economic Institute) want to use AI to cushion a shortage of skilled workers. Spare parts and documentation service is where both meet, because that is where experience sits with individual people. How to make that knowledge available internally is shown in the article on the internal assistant for employees; the technical side of connecting systems is described under integration, and further fields of use are shown in the overview of industries. The following order has proven itself:

Review the last hundred technical enquiries and note which detail was missing before anyone could answer
Check the document stock per series: operating manual, spare parts catalogue, bill of materials, wiring diagram and maintenance schedule, separated by build state
Define the identification rule: which details the assistant requires before it states a part number
Define the answer format: value, unit and source, in exactly that order
Write down the blocked-topic list where the case is handed to a person without further questions
Set up read access to stock and case status, with a clearly limited set of fields
Map the language versions and verify that numerical values are identical in every version
After four weeks, review the conversations and close the documentation gaps that became visible

Frequent questions

Can an AI assistant give a binding spare parts answer in mechanical engineering?

It can state which part is listed for a machine and name the source in the spare parts catalogue. That is not a binding commitment, because it depends on the build state, on modifications and on the actual condition of the assembly. The clean wording is therefore: listed according to the catalogue, chapter and page named, confirmation by spare parts sales. That helps the caller without turning information into a promise.

How does the assistant know which machine is being discussed?

Through type, serial number and year of manufacture from the nameplate. If one of those is missing, it asks before answering and accepts the order or delivery note number instead. Only once the build state is established does it load the documents belonging to that exact machine. Without that step it answers the question about the series rather than the question about the machine, and in spare parts that difference comes with an invoice.

Which documents does the knowledge base need as a minimum?

Operating manual, spare parts catalogue, bill of materials, wiring diagram and maintenance schedule are enough to begin with, provided they are separated by build state. Machinery Regulation (EU) 2023/1230 requires the operating manual to stay online for at least ten years after the machine is placed on the market anyway. Manufacturers meeting that obligation already hold the raw material and only have to order it by machine and build year.

What happens when a running machine breaks down?

Then the assistant hands over instead of advising. A safety reference, a reported stoppage or an explicit request for a callback ends the lookup and opens the route into service, together with the machine data and the full conversation. The technician does not start from scratch but with the type, the serial number and an assembly that has already been narrowed down.

May the assistant collect contact details when a callback is requested?

Yes, provided the purpose is stated and consent is obtained and logged properly during the conversation. For handling a service request that purpose is usually a sufficient legal basis; for promotional follow-up it explicitly is not. Which wording holds up in the chat and what has to be documented is covered in the article on consent in the chat.

Is this worthwhile for a mid-sized machine builder?

The calculation runs through the calls, not through the technology. Anyone handling daily enquiries in spare parts and documentation service whose answers sit in a document is moving exactly that portion, while keeping the cases that need experience. 41 percent (Bitkom) of companies with 20 or more employees already use AI. A reliable picture comes from one series and four weeks of operation; pricing can then be matched to the measured volume, and an initial conversation about your machine documentation clarifies in advance which documents exist.

Ask it yourself, Assistant of this website

Automated assistant

Settings