The essentials
Six terms, six words to remember
Ontology
= Meaning
The rulebook for your data. It says what a customer, an order, a product or a supplier is, and how they’re linked.
A customer places orders. An order contains products. Every product has at least one supplier.
What we put in place
- A shared glossary of your terms
- A map of what links to what
- Rules written down in plain words
Knowledge graph
= Connections
It links your customers, orders, products and suppliers, instead of leaving them in separate tables.
Which customers are hit if one supplier ships late? The graph follows supplier → product → order → customer.
What we put in place
- A structure that follows your rulebook
- Data that keeps it up to date
- Questions ready to ask, for your teams and your AI
Semantic layer
= Consistency
One definition for every number that matters. Revenue, churn or active customers mean the same thing everywhere.
Without it, three teams report three revenue numbers. With it, everyone gets the same answer.
What we put in place
- Definitions agreed together
- One controlled way to get the numbers
- A link to your dashboards and your AI
Entity resolution
= Identity
The same customer often shows up under different names in your CRM, accounts and billing. We find the duplicates and tie them to one identity.
“ACME Corp.” in the CRM, “Acme Corporation” in accounting, “acme corp GmbH” in billing: one customer, one ID.
What we put in place
- Rules for spotting duplicates
- A person checks the doubtful cases
- The right ID sent back to every system
Vector embeddings
= Similarity
A way of turning text into numbers that keep its meaning. AI can then find related content even when the words differ.
A customer asks “Can I get my money back?” Search finds your refund policy, even though the word “refund” never appears.
What we put in place
- Your documents split up and indexed
- Search that understands meaning
- Tests on real questions
GraphRAG
= Context
Search that combines the meaning of words with the links in your data. It finds the right documents, follows the useful links and answers with more context.
“Which suppliers put our biggest customer’s orders at risk?” Ordinary search finds documents about suppliers. This one also follows the link from customer to orders to suppliers.
What we put in place
- Working out who and what the question is about
- Searching the text and following the links
- Checks, and answers that cite their sources
Method
The order it all gets built in
Each step sets up the next. Skip one and the AI ends up reading your data the wrong way.
- 01OntologyAgree on meaning
- 02Entity resolutionClean up duplicates
- 03Knowledge graphLink the data
- 04Semantic layerSettle the numbers
- 05Vector embeddingsIndex the documents
- 06GraphRAGAnswer with context
Warning signs
Signs your data is holding your AI back
- Your AI gets your own data wrong.
- Three teams quote three different revenue figures.
- The same customer exists several times across your systems.
- Your assistant gets stuck as soon as a question needs more than one step.
- Nobody agrees on what an “active customer” is.
- The documents exist, but search can’t find them.
Working together
How we go about it
First an honest look at your data. Then the foundations. Then a first real use.
Taking stock
We measure quality, duplicates and definitions that contradict each other. You know where to start.
The foundations
Data rulebook, duplicates, shared numbers, then the links. In that order.
A first real use
An assistant or agent that builds on these foundations and cites its sources.
Tell us what’s getting in the way
A few lines are enough. Within two working days you’ll get our honest first take, and a time to talk if it makes sense.
- 15
- years in sales and marketing
- 25+
- countries where we’ve launched products and campaigns
- €10M+
- in revenue from the marketing we’ve run