AI Assistant for Small Business

Questions people ask

The cost of asking a colleague

Every company accumulates knowledge that lives in one person's head, one old thread, or a document nobody can find. The visible cost is the interruption: someone stops what they are doing to answer a question they have answered before. The larger cost is the work that simply does not happen, because finding out how is harder than the task itself.

An internal assistant is worth building when that pattern is recognisable to you. Not for the interesting questions, which need a person, but for the ones with a definite answer that is already written down somewhere nobody thinks to look.

Where it helps most

• Onboarding. New staff ask the highest volume of answerable questions and are the most reluctant to keep asking. An assistant that answers without costing them social capital gets used heavily in the first months.
• Multi-location operations. When the person who knows is in another building or another timezone, the question waits. The assistant does not.
• Procedure-heavy work. Anywhere the correct answer depends on a policy that has been revised several times, and the risk is somebody confidently following the old one.
• Pre-task gathering. The information chasing that has to happen before a job can start, which is rarely anyone's actual role and therefore always late.

Access, and what it may change

It runs against your systems with scoped access, agreed explicitly during discovery. Two questions get answered per system: what may it read, and what may it change. Those are separate decisions, and for most organisations the read list is long while the write list is short.

Anything that changes a record runs through deterministic code with an approval step wherever you want one, so a mistake is caught before it commits rather than discovered in a report next month. Scoping is not a formality here. An assistant that can read everything anyone might need is also an assistant that can surface a salary or a disciplinary note to the wrong person, so the boundaries are set deliberately and they usually follow the permissions your systems already enforce.

It works where your team already works

Nobody has to learn a new tool. It is connected to the systems in use, and most interaction happens in the places people are already in during the day. An internal tool that requires a deliberate visit gets used for two weeks and then quietly abandoned, which is the most common way these projects fail.

What it will not do

It will not tell you something that is not recorded anywhere. If the answer only exists in a conversation from two years ago, the assistant does not have it either, and it says so rather than constructing something that sounds right. That is the behaviour you want, even though it is the one that makes it look less impressive in a demo.

It also will not fix an organisation whose documentation is badly out of date. It will make the gaps very visible, which some teams find is the more valuable outcome, but the writing is still work someone has to do.

Whether it is worth it for a small team

Sometimes it is not, and that is worth saying plainly on a page trying to sell you one. The value tracks the volume of repeated internal questions and repeated admin. A small team carrying a lot of both gets more from this than a large team that has already written everything down properly.

If the repeated work is not there, we will tell you during discovery. Where the need is a specific process run end to end rather than questions answered, that is the custom workflow agent instead.

Related reading

• How agent memory architecture actually works
• AI agents in facilities management operations
• Custom workflow agent: automating a defined process