Back to Blog & News

AI

From process data to AI agents: how a Knowledge Graph comes about

Getting from “processes documented somehow” to “an AI agent acting on them” normally takes twelve months. With a structured approach it takes four weeks — because processes are built as a machine-readable structure from the start.

Getting from “processes documented somehow” to “an AI agent acting on them” normally takes 12 months.

With a structured approach it takes four weeks. Not as a promise — as the result of a different starting point: processes are built as a machine-readable structure from the start.

Step 1: raw data — what the company already knows

Every company already holds process knowledge — scattered and unstructured across system logs (ERP, CRM, ticketing), existing documentation (manuals, wikis, SOPs), and the key people who carry the implicit knowledge.

In the Knowledge Graph approach it is extracted automatically — through process mining and short structured interviews.

Your own effort: 3–5 hours a week, for 1–2 weeks.

Result: the complete raw material.

Step 2: extraction — raw data becomes entities

Entities are identified: processes, roles, systems, rules, exceptions — each as its own node.

Properties are captured: duration, frequency, ownership.

Conflicts become visible: when system data shows that a step is skipped in 40% of cases although the manual describes it as mandatory, that gets flagged. Not resolved, just visible. That is information.

Result: clean entities with properties.

Step 3: linking — entities become a Knowledge Graph

A Knowledge Graph is a network of relationships:

FromRelationshipTo

“Check invoice”

is performed by

Role “Accounting”

Role “Accounting”

uses system

“SAP”

Exception “>50k€”

escalates to

Role “CFO”

Rule “regular-customer goodwill”

applies to

“Complaint handling”

Thousands of connections. No human reads this graph — machines query it in milliseconds.

Result: a machine-readable, connected model of the company.

Step 4: activation — the graph becomes usable for AI agents

Query layer: Agents query the graph: “Which rules apply to suppliers above 200k?” → a structured answer.

Context injection: Relevant context is injected automatically. The agent doesn't get everything — only what matters for the task at hand.

Continuous learning: Corrections flow back. The graph learns along with the company — with no manual upkeep.

Result: an agent that acts on company knowledge — not on assumptions.

The path end to end

  • Weeks 1–2: capture raw data
  • Weeks 2–3: extraction
  • Weeks 3–4: linking (Knowledge Graph)
  • From week 4: activation (AI agents)

After week 4 the graph isn't finished — it is operational. It keeps improving.

The difference from classic BPM: the result isn't a document that goes out of date. It is a system that learns.

About the Authors

Dr. Christian Graup

Managing Director · Product & Direction

Leads product vision and OIS architecture at aiio. Establishes the scientific and practical foundations for machine-readable organizations.

LinkedIn Profile →

Jobst von Heintze

Managing Director · Market & Communications

Responsible for brand leadership, communications, and market positioning at aiio. Translates complex B2B process technology into clear market narratives.

LinkedIn Profile →

Next step

Move from reading to organizational capability.

Twenty minutes, no pitch.

Get in touch