Alternatives

What else you could do about this.

The realistic alternatives are a general AI assistant, your document library and its search box, your technical phone line, and doing nothing. Each is a reasonable choice in some circumstances, and this page says where each one holds and where it stops.

Four options, and each of them is the right answer for somebody. This page says where each one holds and where it stops, including the one where you do nothing at all.

Option one

A general AI assistant.

This is not a hypothetical. Specifiers and installers are already asking general assistants about your products, today, and the conversation happens whether you participate in it or not.

Where it holds

It costs nothing, it is already in the hands of the people asking, and it will answer instantly and fluently in the asker's own words. For a general question about a category of product it is genuinely useful.

Where it stops

It has no defined corpus, so it cannot tell you when an answer is outside your documentation. It draws on training data, a competitor's page, a forum post, or a ten-year-old version of your own bulletin, and it presents all of them with the same confidence. The answer is unattributable, so it is worth nothing in a dispute. And the surface is invisible to you: you cannot see what was said about your product, and you cannot correct it.

Option two

Your document library and its search box.

Every manufacturer in this industry resolves technical questions this way, and there is a reason it has lasted.

Where it holds

The documents are authoritative and, if the library is maintained, current. Nothing is invented, because nothing is generated. It already exists, it costs nothing further, and for a person who knows which document they need it is the fastest route there is.

Where it stops

It is organised by document and people search by question. Search returns documents, not answers, so a specifier who needs one number gets ten PDFs and ninety seconds of patience. It also produces no record of what anyone was looking for, which means the questions your literature cannot answer stay invisible. If your library is small and the questions are simple, this is probably enough.

Option three

Your technical line, answered by a person.

The highest-quality answer available, and the one we are least interested in replacing.

Where it holds

A person can interpret, exercise judgement, hear what is actually being asked, and handle the question the documentation does not cover. On anything that requires an engineering view, this is the only correct option, and it is where a refused question should end up.

Where it stops

It is expensive and it does not scale. The same forty routine questions consume somebody paid to do harder things. It stops at five o'clock, and on the Friday of a long weekend nobody answers at all. And what is said on the phone leaves no record, so when a job goes wrong and the dispute turns on what your company told somebody, the file is empty.

Option four

Doing nothing.

The default, and the hardest one to argue against internally, because its costs do not appear in any report.

Where it holds

It costs nothing, it breaks nothing, and it needs no project, no budget line and nobody's time. If your products have no published technical literature, or nobody inside the company is absorbing repeat questions, or you cannot name a consequence of somebody reading a detail wrong, then doing nothing is the correct answer and we will say so on the call.

Where it stops

The specifications you lose are lost silently, and nothing about them reaches any report you receive. The general assistants keep answering on your behalf. The phone keeps ringing. The file stays empty. None of that shows up as a number, which is exactly why it persists.

The difference

Four things none of the above does together.

A defined corpus, so there is a boundary to be outside of. A citation on every answer, so it can be checked against the document rather than believed. A refusal when the answer is not there, so nothing is invented to fill the gap. And a record of every question asked, so the gaps stop being invisible.

Any one of those on its own is available elsewhere. The combination is what a technical department can put its name on.

Questions

Why not just use a general AI assistant?

Because it has no defined corpus, so it cannot tell you when an answer is not in your documentation. It answers from training data, a competitor's page or an old version of your own bulletin, on a surface you cannot see or correct.

We already have a document library. Is that not enough?

If your library is small and the questions are simple, it may well be. A library returns documents and people ask questions, so the gap shows up as library size grows and as the questions get more specific.

Will this replace our technical rep?

No, and it is not meant to. It absorbs the routine questions so that the person who can exercise judgement is free for the question that needs it.

When is doing nothing the right answer?

When nobody internally is absorbing repeat technical questions, or when you cannot name a consequence of somebody reading a detail wrong. We will say so on the call rather than sell past it.