AI in the process model

AI in the process model

Where is AI already
at work in your organisation?

The question is no longer whether AI is used. It is: do you know where, with what and who is accountable for the result – and does your process model show it?

Starting point

Accept it or steer it?

AI has long been part of everyday work – often in a private chat, with company data, without approval and without a trace in the process. The greater danger is not that employees use AI. It is that nobody knows where.

Not like this

Here and there, privately

AI?AI?
Who knows about it?
The person themselves – perhaps the team.
Which data?
Whatever is being copied into the chat.
Who checks the result?
Nobody explicitly.
If something goes wrong
Nobody knows where it came from.
Evidence
None.
Like this

Introduced – at the process

AIAI
Who knows about it?
Everyone who knows the process.
Which data?
What the classification allows.
Who checks the result?
A defined role approves.
If something goes wrong
Origin and approval are recorded on the result.
Evidence
In the model, in reporting, in the Viewer.
In the model

AI is recorded at the step – like any other participant.

A step that an AI carries out looks different in the model from one where it only helps. In both cases the model says who the AI is – and which person is accountable for the result.

HumanRecord enquiry
RCase handler
AI-assistedDraft reply
RCase handlerKAI assistant
AIReconcile data
KAI assistantAHead of customer service
AI at the step “Reconcile data”Example
AI employee (K)
AI assistant
Use of AI
AI carries it out on its own
Operation of the AI
external (provider service)
Data passed to the AI
personal data, pseudonymised
Human oversight
A person checks samples
AI risk level
limited risk (transparency obligation)
AI potential
high
Accountable person (A)
Head of customer service
HumanApprove reply
RCase handlerAHead of customer service
Example: answering a customer enquiry.R carries out · A decides · K: AI participates.
Shape “AI”

The AI carries out the step.

Who the AI is, is recorded at the step as an AI employee with key K. A person stands next to it as A – they are accountable for what the AI delivers.

AIKA
K AI employeeA accountable person
Shape “AI-assisted”

The person works with AI.

The step states whether the person can, should or must use AI – and which AI it is.

AIRK
R Person carries outK AI participates
RACI becomes RACI-KR carries out, A decides, C is consulted, I is informed – K: AI participates. R, A, C and I remain people.
Origin stays visibleWhat an agent has created or changed is marked on the element – with the person and time of approval.
Protection applies to AI tooWhat an AI agent may see is set by the administrator per classification.
Ask across all modelsWhere does this system occur? What is AI-generated but not approved? The attribute directory answers.
See for yourself

In the video, in the model, in the training.

Learning video · 4 minutes

Making AI visible in the process

Where AI works in the business process, who stays accountable and what it delivers – at the step, in the Viewer and in reporting.

In the Viewer

Open a model with AI

Demo model “KI-Agent einführen und freigeben” (introducing and approving an AI agent): a process in which AI takes part – with AI employee, oversight and risk level at the step.

Another model: “Videoschnitt: vom Rohschnitt zur Freigabe” (video editing) →
In TAOM Training

The course for the video

The video as a course in TAOM Training: e-learning, at any time – open to all registered companies.

Risk reporting

See where AI participates – and where something is missing.

What is entered at the step is evaluated by reporting across all models: use, risk, gaps and effect.

42Steps with AI
3AI without an accountable person
3High AI risk
7Personal data passed to the AI
12AI operated externally

Example figures – this is how the “AI & RACI-K” evaluation reads.

Potential × risk

Where much benefit meets little risk – and where it does not.

lowmediumhighprohibited practice001high risk101limited risk465minimal risk675

Rows: AI risk level · Columns: AI potential. Click a cell to see the steps.

Gaps in the use of AI

What is missing is shown in a list.

  • AI without an accountable person (A)3
  • AI without an AI employee (K)2
  • Risk level not classified4
  • Without human oversight1
  • Potential not rated2

Only what is entered at the step is evaluated.

Risk level at the stepFour levels following the EU AI Act, from “minimal risk” to “prohibited practice” – as your own assessment.
Record per AI employeeOne AI employee per row: where it participates, how, and with which risk.
Measure the effectMeasurement series without and with AI: reporting calculates the processing time saved per month.

Reporting provides calculated indications. Assessment and decision rest with the persons responsible.

See it on your own workflows.We go through with you where AI is already at work today – and record it in the model.
Five test questions

Can your process model show this today?

A model is a picture – and what is not in the picture does not exist for the organisation.

  • Can you tell at the step whether an AI participates?
  • Which AI is it – and which data does it work with?
  • Who approves the result?
  • Has the risk been assessed – and is there a control?
  • Can you ask across all processes where AI is used?
Who it is for

Size changes the tool – not the question.

Small businessOne person answers customer enquiries with AI – including names, addresses and prices.
CorporationA hundred teams work with a hundred different assistants.
Public administrationOfficial secrecy and data protection come on top: much must not go into public AI applications at all.
Not sure exactly where AI is at work in your organisation?That is the normal case. Start with one workflow.

More on this: AI Workforce → · Information protection and classification → · TAOM for management →