Company brain AI for multi-brand equipment service

The manager of such a team deals with scattered knowledge. Manufacturers publish manuals in their portals, technicians keep downloaded copies, and records of equipment modifications end up in service tickets. Searching for a model name can produce several convincing answers. Each still needs to be checked against the machine at that customer's site.

A similar manual may cover a different machine

Consider an illustrative system evaluation. A service company maintains equipment from several manufacturers. A ticket concerns a machine family with different control system variants. The document library contains a current manual for an earlier serial number range and another document that applies to the customer's machine. Both use the same family name and similar headings. A more recent publication date on the first file does not make it the right source for this ticket.

The service history also records a controller replacement. The machine's serial number alone may therefore be insufficient to select its documentation. In the proposed test, the assistant should identify which details establish the current configuration and request confirmation of missing information. If the equipment register conflicts with the replacement report, the person responsible for the documentation needs to resolve the discrepancy.

The expected result is the right document, an explanation of why it applies, a link to the relevant passage, and a visible version. The evaluation should also retain the reason for rejecting the similar material. That lets the manager check whether the system recognized the difference between variants or happened to choose the correct file.

This scenario concerns finding the evidence needed for the work. Technical assessment and repair decisions remain the responsibility of authorized personnel working under the service organization's procedures.

Answers should come from an available source

A language model can use a manufacturer's terminology even when it has not received the relevant manual in the conversation. A fluent answer does not establish that its information is current. A knowledge assistant should retrieve material that applies to the case and base its answer on that content. This approach is called retrieval-augmented generation, or RAG: generating answers using retrieved sources.

A reference must lead to a passage that supports the answer. A citation from the wrong machine variant is still an error. Acceptance testing should assess both whether the answer accurately reflects the document and whether that document applies to the specified machine.

Facts in this type of system can be updated by changing the sources available to the assistant. This does not require additional model training, known as fine-tuning, every time a fact changes. You still need to check when revised material starts affecting answers and what happens to the previous version. A custom or privately deployed model may address other project requirements, but selecting one does not resolve outdated documentation by itself.

Who maintains the catalog and access rules

Before requesting a quote, identify a documentation owner for each brand or equipment group. That person should be able to confirm where material applies, resolve conflicts, and approve its withdrawal. The system provider needs these decisions to reflect the service organization's rules.

Microsoft's SharePoint content guidance for the Employee Self-Service agent describes the importance of current, structured, authoritative material and managed permissions for answers with references. This guidance covers a specific product. When buying a custom assistant, establish how those responsibilities will be handled in the repositories your service team uses.

Withdrawing a manual from current use does not necessarily mean deleting it from the archive. An older document may be needed to explain a past service ticket. The system should clearly distinguish that historical lookup from an answer about the machine's current state. Also agree on what happens when a manufacturer removes a file or publishes a correction without a clear version label.

Customer history needs separate access rules. A manufacturer's general manual may be available to the whole team, while reports and contractual arrangements belong to a particular customer's records. A technician serving one company should not receive excerpts from another company's tickets simply because both involve similar machines. If you plan to provide a customer portal, treat its knowledge scope and permissions as a separate part of acceptance testing.

When a library is enough and when an assistant helps

An organized document library or a manufacturer's search tool may be sufficient when a technician knows the equipment designation and needs a single document. First assess how much a clear catalog of variants and access to applicable material would resolve.

An assistant has a stronger case when an answer requires documentation from different brands to be combined with individual equipment records and service history. The purchase then also covers relationships between sources, handling uncertainty, and access controls. Our article on choosing a ready-made platform or a custom company brain discusses the broader differences in scope.

An agent that performs work adds another scope. Creating a ticket, updating an equipment record, or ordering parts changes data or initiates a process. Each action requires separate agreement on permissions, approval conditions, and how its result will be checked. The first project can end with an answer supported by documentation if that meets the team's needs.

Test cases that are easy to confuse

Ask the provider to evaluate cases selected by the service team, with expected results approved by the knowledge owner. The set should include:

• Current documents for the same machine family that cover different variants and serial number ranges.
• A ticket with insufficient information to identify the individual machine.
• A conflict between the equipment register and modification history.
• A withdrawn manual and a question about a past service visit.
• Similar tickets from two customers to whose records the user has different access rights.

For each case, record which source may be used, what the answer must exclude, and when the case should be referred to an expert. Also agree on how errors will be reported after launch and who decides when corrected knowledge can be used again. Useful measures include the time needed to reach a verified document and the accuracy of document selection.

What to discuss with a provider

Syntalith builds a company brain tailored to a team's sources and work, as well as applications, automations, and agents. For a service organization, this allows knowledge retrieval and any subsequent work in the ticketing system to be considered separately. The scope of development, integrations, and maintenance needs to be agreed. See Syntalith's pricing page for the current offer.

Include the brands you service, a description of your equipment register, and the locations of your documents in your inquiry. Prepare an anonymized example of a mix-up between variants and identify who can resolve such a case. This will help establish whether the initial engagement should cover organizing the catalog, building a knowledge assistant, or connecting it to service history.