The problem
An AI assistant for service manuals addresses a problem every equipment maker with a dealer network knows. A technician is at a customer's site with a machine that will not start. The fix is documented somewhere: in the service manual for that model, in a bulletin issued last spring, or in the parts list for that serial range.
Finding it is the hard part. Manuals run to hundreds of pages, bulletins pile up in a separate folder, and parts lists change between production runs. So the technician calls your help desk, and your engineers answer the same questions they answered last week.
Wrong parts are the other cost. Order a part from the parts list for the wrong serial range and the machine stays down until the right one arrives.
What we would build: an AI assistant for service manuals in your dealer portal
We would add an assistant to your dealer portal that answers only from your technical documents. It relies on retrieval-augmented generation, or RAG:
Split. Manuals, bulletins and parts lists are broken into passages. Each keeps its document, page, figure, model and serial range.
Index. The passages go into a search index that matches by meaning, so "engine cranks but won't fire" finds the no-start troubleshooting table.
Search, then answer. For each question, the system retrieves passages for that machine only. A language model then writes the steps from those passages and cites where each one came from.
If the documents do not cover the question, the assistant says so plainly and offers to open a support case.
How it works
The technician signs in to the dealer portal on a phone or tablet and enters the model and serial number, or scans the plate. Then they describe the fault in their own words.
The assistant replies with the likely checks and the repair steps, each citing its manual page or bulletin number.
If a bulletin has changed a step, the bulletin comes first and the old text is marked as replaced. Every part named in the answer shows its part number for that serial range, linked to the parts portal with stock and dealer price.
When the fault is not covered, the technician raises a case from the same screen. Factory support receives the machine details, the question and the passages already tried. Their answer can then feed a new bulletin, and the next technician finds it.
What the AI does, and what your team decides
The AI matches a described fault to the right passages, puts bulletins and manual text in the right order, writes cited steps and pulls the correct part numbers. It also groups unanswered questions by model.
Your people stay in charge. Technical publications decide which documents go in and when. Support engineers own every answer the documents cannot give. Dealers decide whether a repair goes ahead and what the end customer is charged. Warranty claims follow your existing approval process.
What the research says
Finding answers costs real time. A survey of 12,000 knowledge workers and 200 executives by software maker Atlassian found that leaders and teams lose 25% of their time just searching for answers (Atlassian, State of Teams 2025). A technician hunting through manuals and bulletins on site pays the same tax, so the assistant brings the right passage to them.
Retrieval makes answers more specific. The 2020 paper that introduced retrieval-augmented generation found RAG models produced more specific, diverse and factual language than a model answering from its training alone. It also named giving provenance for answers as an open problem (Lewis et al., NeurIPS 2020).
Grounding still needs checking. A preregistered 2024 evaluation of commercial RAG-based legal research tools found each still hallucinated between 17% and 33% of the time (Magesh et al.). In repair work, that means every step should show its source page.
Guardrails
Data access. Dealer tier, region and role decide which documents a person can search. Search results respect the same rules as the document library.
Human approval. Safety warnings are quoted exactly from the manual. Warranty work, field fixes outside the documents and part substitutions need approval from your support team.
Logging. Each question, the machine, the passages used and the answer are logged, which helps with warranty reviews and shows where documents need work.
Versioning. When a manual or bulletin is reissued, the old passages leave the index the same day.
Is this right for you?
Are your manuals and bulletins digital, and is each tied to models and serial ranges?
How many help desk calls are questions the documents already answer?
Do dealers already order parts online, and could answers link to it?
Which procedures should only authorized service dealers see?
How does the assistant stay inside your manuals? Our explainer on retrieval-augmented generation walks through it. For the build, see our AI agent development and custom software development pages, or explore our ecommerce and B2B commerce work.
