A brand, importer or manufacturer of technical products answers the same questions every week: what fault code E-xx means on this model, when the filter is due, how a parameter is set, which part belongs to which serial number. AI customer service for after-sales answers those questions from your own manuals, service documentation and parts catalogues, on the channels owners and installers already use, and passes everything else to the right person with the conversation attached.
This post is for the IT manager, head of customer service or informatiemanager at a 250 to 5,000 FTE organisation in the Netherlands or Flanders that sells through dealers or installers and already runs a ticketing system with an ICT partner. Crux Digits designs and builds AI systems alongside that partner; we do not run helpdesks, hosting or IT management.
What does AI customer service do in after-sales for technical products?
It answers four groups from your own service documentation: owners, installers on a job, the parts counter and new service staff.
After-sales differs from webshop support in one way that matters: the right answer depends on the exact model, and often on the production year or serial number. The owner manual, the installation manual, the service manual, technical bulletins, wiring diagrams and exploded views are written per model and revised over time. An AI assistant is built to work like an experienced colleague: ask for the model first, then read only the documents that belong to it.
What a team gains is concrete. First answers arrive in seconds instead of after a callback. Evenings and weekends, when owners notice a fault and installers finish their jobs, are covered. Answers are consistent, because every channel reads the same documents. And the service desk spends its hours on the cases that need a technician's judgement, a warranty decision or a site visit.
How does it answer owners who ask about fault codes and maintenance?
It identifies the model, looks up the code in that model's owner manual, and gives only the steps the manual assigns to the user.
The first question back is usually the model or the serial number from the type plate, because fault code E-xx on one generation can mean something else on the next. Once the model is known, the assistant answers from the owner documentation: what the code means, which checks an owner may carry out, and when the dealer or installer takes over.
The line between user tasks and technician tasks is not something the assistant has to invent. For machinery, the Machinery Regulation (EU) 2023/1230 requires the instructions for use to describe the adjustment and maintenance operations to be carried out by the user, in Annex III, section 1.7.4.2, point (r). A well-written manual draws that line for every product. The assistant follows it, and everything past the line becomes a handover: a service request carrying the model, serial number, code and the steps already tried.
Maintenance questions work the same way: filter intervals, descaling, winter storage, battery care.
How does it help installers and technicians during a job?
A technician asks via WhatsApp or the dealer portal and gets the setting or procedure from the service manual, with its source named.
An installer next to a unit wants one answer and its origin. So every technical answer cites its source: document title, revision and section. When a technical bulletin has replaced a value in the manual, the bulletin wins, and the answer says so.
Installer content stays with installers. Service manuals, parameter lists and wiring diagrams are often meant for trained professionals only. That is a design choice in the knowledge base: owner documents are public, installer documents sit behind a login, and the assistant on the dealer or installer portal recognises the user by a verified email address before it answers from them.
The calls that still reach your technical helpdesk are the ones worth a senior's time: a fault the documentation does not cover, a pattern across several sites, a product question for engineering.
How does the parts counter match a serial number to the right part?
The assistant reads the serial number or production year, picks the matching exploded view, and returns position number, part number and drawing reference.
Parts catalogues change between model years: a part is redesigned, a number superseded, a kit replaces loose parts. The parts list normally states which serial range or production year each drawing covers. The assistant uses that to pick the right drawing before it names a part, and it states the range it matched, so the person at the counter can check it at a glance.
Ordering stays with people and your systems. The assistant prepares the request and hands it to the parts team with the conversation attached. A useful handover carries eight fields: channel, role (owner, installer or dealer), model, serial number or production year, fault code or position number, part number, steps already taken, and the documents cited. Live stock, dealer prices and delivery dates sit in your ERP or dealer management system, which is where a custom integration starts.
How do new service staff learn the product range faster?
New staff ask the knowledge base the customers use, and learn the range from the source documents instead of interrupting a senior colleague.
A new colleague at the service desk or the parts counter asks the questions they would otherwise take to the person at the next desk: which models share this pump, where the code table for the previous generation is, how a warranty case is passed on. Each answer names the document it came from, so the new colleague learns where things live, not only the answer.
There is a second gain. The questions the assistant cannot answer show exactly where documentation is missing or out of date. That list goes to the technical documentation team.
What do example exchanges look like in four different trades?
The four exchanges below are illustrative, written for this post: no real brand, codes shown as E-xx placeholders, each answer naming its source document.
Heat pump, installer during commissioning. Question: "Outdoor unit shows E-xx shortly after start-up. What do I check first?" Answer: "For this model, E-xx indicates insufficient flow in the water circuit. Check the strainer, the isolation valves towards the indoor unit and the minimum flow rate in the parameter list. Work on the refrigerant circuit is outside this procedure." Source: installation and service manual for the outdoor unit, section on fault codes, and the commissioning checklist.
Forklift, fleet manager at a distribution centre. Question: "A truck shows E-xx after a battery change. Can we keep using it?" Answer: "The operator manual lists E-xx as a battery connection or battery type mismatch. Take the truck out of use and carry out the battery check described under daily checks. If the code remains, I will pass this to your dealer's service team with the truck's serial number." Source: instructions for use, sections on daily checks and display messages.
E-bike, dealer workshop at the parts counter. Question: "Frame number starts with XX. The customer needs a new derailleur hanger." Answer: "That frame number range belongs to the second model year of this frame. The hanger is position 14 on the rear frame exploded view, part number XX-XXXX, which supersedes XX-XXXX. Shall I send an order request to the parts team?" Source: parts catalogue, model year table and rear frame exploded view.
Hot tub, owner on a Saturday evening. Question: "The display shows E-xx and the water is not heating." Answer: "On this model E-xx is a low-flow warning, and the heater pauses until flow returns. Check that the water level is above the skimmer opening, then remove and rinse the filter cartridge as shown in your owner manual. If the code returns after a restart, I will send your model, serial number and these steps to your dealer." Source: owner manual, sections on display codes and filter care.
All four follow one pattern: identify the model, answer from the document that applies, name it, and hand over with full context when the next step belongs to a person.
Which design choices keep after-sales answers correct?
Five choices do most of the work: model before answer, documents only, citations, separate owner and installer libraries, and written handover rules.
Model before answer: the assistant asks for the model or serial number whenever the documents differ per version. Documents only: when the documents do not contain the answer, it says so and hands over instead of filling the gap. Citation: document, revision and section on every technical answer. Two libraries: public owner documentation and login-only installer documentation. Handover rules: which questions go to the service desk, the parts team or a dealer, and which eight fields travel with them.
One rule comes from the AI Act. Article 50(1) requires that AI systems interacting with people are designed so those people know they are dealing with an AI system, unless that is obvious, and under Article 113 it has applied since 2 August 2026. A clear line at the start of every conversation covers it.
The documentation library is getting a legal anchor as well. The Machinery Regulation applies from 20 January 2027 and allows instructions for use in digital format. Under Article 10(7), the manufacturer must then mark on the product how to reach them, make them printable and downloadable, and keep them online for the expected lifetime of the product and at least 10 years after it is placed on the market. A user can still ask for paper at the time of purchase, free of charge within one month, and products for non-professional users keep the essential safety information on paper. For a machinery manufacturer, the versioned per-model library the regulation asks for is the library the assistant answers from. Build it once, use it twice.
Should you buy an off-the-shelf assistant or have one built?
Buy when the answers live in documents; build when they live in your systems, such as ERP stock, warranty claims or a serial-number registry.
SimplyBoost is our own product: an AI support agent that answers from your website, PDFs, FAQ and catalogue on website chat, WhatsApp, Instagram and Facebook Messenger. An account includes three AI agents, which fits after-sales well: one for owners, one for installers and dealers, one for your own team. It hands a conversation to a person, or to Zendesk, Freshdesk, Gorgias or Slack, with the conversation attached. An agent can run on logged-in portal pages and recognise users by verified email, which keeps installer content with installers, and it replies in Dutch, French, German, Spanish and English. See simplyboost.io.
A custom build makes sense when the answer has to come from live data: is the part in stock, is this serial number still under warranty, which service visit is planned, what did the last inspection record. Crux Digits builds that part at fixed prices per step: an AI audit at €2,500, a Production-ready MVP at €20,000 and production from €50,000, all excluding VAT, with the code handed over to you. Many after-sales teams start with documents and add one system integration once the question volume shows where it pays.
Where does Crux Digits fit next to your ICT partner and service team?
Crux Digits builds the AI layer, your ICT partner keeps running infrastructure and ticketing, and your service team owns the answers.
In practice we help you structure the documentation library per model and revision, define the owner and installer audiences, write the handover rules into your existing ticketing system, and build integrations with ERP, dealer management or warranty systems where documents are not enough. We do not take over helpdesk, hosting or IT management. Your service team owns the content, your ICT partner owns the platform, and you own the code.
For the choice between buying and building AI customer service, see AI customer support: buy or build.
Last updated: 28 October 2026.
SimplyBoost, our own product, answers from your documents on website chat and WhatsApp. When the answer has to come from your own systems, we build that part.
AI customer service agent →
