Company Second Brain · Knowledge for AI
Most of what your company knows isn’t written down anywhere.
We put it in one place your team, your new hires and your AI systems can all read. That order matters.
The hard part isn’t the AI. It’s the information.
Why this customer gets those payment terms. Which supplier was trouble
in 2023 and why. What you agreed on the phone in March. Where the spare
key is.
None of that is in a system. It is in three people’s personal WhatsApp, in mail threads, in a spreadsheet only one person opens, on paper, and a good part of it in the manager’s head.
None of that is in a system. It is in three people’s personal WhatsApp, in mail threads, in a spreadsheet only one person opens, on paper, and a good part of it in the manager’s head.
In that state there is not much an AI can do with it, whichever one you
pick. The model is rarely the bottleneck.
It is also not much use to the person you hired last month, which is the version of this problem you were already paying for before anyone mentioned AI.
It is also not much use to the person you hired last month, which is the version of this problem you were already paying for before anyone mentioned AI.
What it actually is
One place in the middle, and everything else pointing at it.
What we build
Four parts, and the fourth is the one that decides whether it survives.
A place for it to go
One store, structured around how your business is actually organised:
customers, suppliers, products, decisions. Plain files and a database,
not a proprietary format you cannot leave.
Getting the existing pile in
Years of mail, documents and chat, sorted and summarised into entries
rather than dumped in a folder. This is the part AI is genuinely good at,
and the part nobody would do by hand.
A way to ask
Questions in plain language, answered with a link to where the answer
came from. An answer you cannot trace is a rumour, and people stop trusting
the system after the second one.
A way to keep it current
Capture built into work that is already happening, so writing it down
is not a separate chore somebody has to remember. This is where every internal
wiki you have ever seen died.
How it works
Three steps. You can stop after any of them.
Find out what you know
One to two weeks. We work out what knowledge your company runs on, where
each piece currently lives, and which pieces would hurt most to lose. You
get that as a map, and a lot of people find the map alone uncomfortable
reading.
One area, done properly
Four to six weeks on the area where not knowing costs you most, usually
customers or suppliers. Store, import, search, capture habit, all of it,
for one slice. A small store people use beats a complete one they do not.
Widen it, then connect it
The remaining areas, and then the part everyone was waiting for: agents
that can answer from it and act on it. That step is genuinely easy once
the store is real, which is the whole reason it comes last.
Who this is for
This is for you if
- There is one person whose holiday is a company risk.
- New people take months to become useful, and nobody can say exactly why.
- The same question gets asked and answered several times a week.
- You tried AI on your own documents and the answers were confidently wrong.
Probably not, if
- You want a document dump with a search box over it.
- Nobody on your side is willing to own it after we hand it over.
Questions we get asked
- Is this just a wiki?
- A wiki is one possible shape for it, and most wikis die within a year because keeping them current is nobody’s job. What matters more than the tool is where the entries come from. If writing something down is a separate chore it will not happen, so we take it out of work that is already being done wherever we can.
- Do we need this before we do anything with AI?
- Not before everything. Plenty of useful automation runs on data that is already structured, in your accounting system or your shop. You need this for the questions where the answer lives in prose: why this customer gets those terms, how we handle that supplier, what we decided last March.
- Where does it live?
- Wherever your rules allow, and we ask that first. It can be a server you own, a data centre you pick, or a hosted service. The store is plain files and a database rather than a proprietary format, so moving it later is an afternoon and not a migration project.
- Who is allowed to see what?
- Access is part of the design, not a setting bolted on afterwards. Salaries, contracts and personnel notes sit behind their own permissions, and an AI answering a question only ever sees what the person asking is allowed to see. Getting this wrong is the fastest way to make the whole thing unusable.
- How long does it take?
- The first useful version is weeks, not months, because it does not start with everything. We pick the area where not knowing costs you most, usually customers or suppliers, and get that one right. The rest grows from a working example instead of from a plan nobody reads.
- What happens when the person who knows everything leaves?
- That is the reason most companies call us about this, usually a little later than they should have. It is much easier to capture what somebody knows while they are still answering questions cheerfully than during a handover month. If that person exists in your company, start there.
Who in your company can’t go on holiday?
Call the office nearest you. That question usually tells us within ten minutes whether this is worth doing. No sales sequence.