Ask the AI your team uses why you stopped buying from a supplier last year, and you'll get a polite guess. The real answer exists. It's in an email thread, a meeting note and someone's memory. A business knowledge layer is where that kind of answer lives. Below, we explain what goes in one and when it's the wrong thing to buy.
The short answer
A business knowledge layer is one place that holds what your business knows and has decided, in plain files that people and AI can both read. It doesn't copy your numbers. When an AI needs a figure, it asks the system that owns it, such as your accounts or your CRM.
Two different kinds of question
Some questions are about numbers. What did we invoice last month? How many jobs are open? Your accounts, CRM and job system already answer those, and they're the authority.
Other questions are about meaning. Why did we price phase 2 that way? Which supplier did we drop, and why? What's our policy on software spend? Those answers are spread across email, documents and people's heads. A knowledge layer is built for the second kind of question. A data warehouse is built for the first.
What goes in, and what stays out
In the knowledge layer:
- Decisions, and the reasons behind them
- How the business does things: process, pricing rules and templates
- Who owns what
- Policies and the exceptions to them
In your systems, where it already lives: figures, transactions, customer records and anything else a system of record holds. Copying those into a second place only creates a copy that goes stale.
Why it goes out of date
Knowledge drifts. People change roles, prices change and old notes contradict new ones. A knowledge layer needs upkeep: broken links found, contradictions listed, loose notes merged into proper pages, and a way to flag a wrong answer so the page behind it gets fixed.
If nobody owns that job, the layer gets a little less trustworthy every month. Ask who looks after it before you ask what it costs.
When it's the wrong thing to buy
If your systems disagree about the numbers, start there. A knowledge layer can't fix a CRM with three versions of every customer. We'll tell you that on a call, and a data clean-up should come first.
A knowledge layer holds meaning. Your systems hold the numbers.
Five questions to ask any vendor
- What does it hold, and what does it leave to your systems?
- Whose is it, and can you open the files without their software?
- Who can see what?
- Who can change it, and what waits for approval?
- What happens after go-live, and when it doesn't know the answer?
We go through each one, with a test to ask for, in five questions to ask before you buy a business knowledge layer.
A quick example
A 30-person engineering firm. A project manager asks the AI, "Why did we stop using our old steel supplier?" Without a knowledge layer, the AI has nothing to go on. With one, it finds the decision note from the March operations meeting and the two late-delivery emails behind it, and shows both as its sources.
Where to start
Write down five questions your team asked the AI last month that it couldn't answer. If most of them are about meaning rather than numbers, a knowledge layer is worth a look.
Common questions
What is a business knowledge layer?
It's one place where a business keeps what it knows and has decided, readable by people and by AI. It sits above your systems and doesn't replace them.
Is a knowledge layer the same as a data warehouse?
No. A data warehouse copies your figures so they can be analysed. A knowledge layer holds decisions, criteria, owners and rules, and leaves the figures in the systems that own them.
Do we have to move our files?
No. A knowledge layer connects to your sources read-only, so your files stay where they are.
What happens when it doesn't know the answer?
It should say so plainly rather than guess. That gap also tells you what to write down next.