Services · Building block

Processes that run on their own. And stop when something is missing.

Workflow automation links ERP, CRM and mailbox into one flow with rules and approval. An AI step is added only where judgement is needed.

Free intro callDiscuss your project
Incoming invoicesWorkflow live
  1. New email to invoices@Accounting mailbox
  2. Read attachmentInvoice RE-26-1187, 14 linesConfidence 0.97
  3. Find purchase order in ERPPO-40211, goods received
  4. Amount within tolerance?Rule: up to 1 % difference
yesno
  1. Create posting proposalin the ERP, not posted
  2. Purchasing approvesS. Albers
  1. Clarification listRE-26-1187, €42.80 difference

Today 214 completed, 6 in clarification, 0 posted silently.

  • No more retyping

    Data goes into the system once and moves on from there. Nobody copies numbers from one window to the next.

  • Nothing slips through

    Anything that moves money or leaves the company waits for approval. If a step fails, the process stops.

  • Rules before AI

    What a rule can decide, a rule decides. That is cheaper and can be checked.

  • Not tied to one person

    The process is written down and keeps running when your colleague is on holiday.

Where it gets stuck today.

  1. Retyping at every system boundary.

    The order is in the CRM, it has to go into the ERP, and the confirmation goes out by email. In between sits a person copying it over, with the odd transposed digit.

  2. The process depends on one person.

    Only one colleague knows in which order the windows have to be worked. When she is on holiday, the cases pile up.

  3. Nobody notices when it stalls.

    An export breaks overnight, an approval has been waiting for a week. It comes out when the customer calls.

Process

From trigger to record in the ERP.

An invoice arrives in the mailbox, rules check it against order and supplier, and the record is ready in the ERP. Nobody retypes anything in between.

invoices@Mailbox
  • Hesse Industriebedarf07:58Invoice RE-26-1187PDF, 2 pagesTrigger
  • Brandt GmbH07:41Invoice 2026-0412PDF, 1 pageTrigger
  • Hesse Industriebedarf07:30October newsletterno trigger
Check 2026-0412Brandt GmbH
  1. Order number presentBE-40198yes
  2. Supplier knownVendor 70388yes
  3. Amount within tolerance, up to 1 %as orderedyes

All rules met

Incoming invoiceERP
Vendor70388 Brandt GmbH
OrderBE-40198
Invoice number2026-0412
Net amount€1,186.40
Line items6 of 6 matched to the order
Send for approval

Rule or AI?

AI only for the step that needs judgement.

Most steps are if-then: same number, amount within tolerance, customer known. An AI step is added only where free text, a PDF or a photo has to be read.

  • A rule, where if-then is enough

    Fields, numbers, amounts: the decision can be written down. That is cheaper, faster and auditable.

  • An AI step, where something must be read

    Free text, PDF, photo or speech. Narrowly scoped, with a confidence score and a threshold below which a person takes over.

  • An agent, where the path differs per case

    Then it is no longer a fixed process. The workflow moves the data, an AI agent takes the one step that needs judgement.

Steps in the processIncoming invoices
  1. Recognise what the email is aboutInvoice, confidence 0.97AI step
  2. Read line items from the PDF14 items, confidence 0.95AI step
  3. Find the order in the ERPBE-40211 foundrule
  4. Amount within tolerancedeviation up to 1 %rule
Clarification listJ. Wendt
  1. Email from Lange & SöhneRequest unclear, confidence 0.61below threshold
  2. RE-26-1187, HesseDeviation €42.80rule applies

Operations

Stop rather than carry on quietly.

If a step fails, the case lands with its current state on a named person's desk. An overview shows every day what ran through and what is stuck.

Overviewtoday
  1. RE-26-1185, Brandt GmbHPurchasing approval, S. Albersapproved
  2. RE-26-1187, Hesse Industriebedarf€42.80 differencein clarification
  3. Stock export, 02:00failed, state savedreported

Clarification list: J. Wendt, accounting

AccountingTeams

  1. Stock export at 02:00 aborted. State saved, nothing written twice.

    Workflow stock export
  2. J. Wendt

    Thanks, I will restart it after the ERP update.

Approaches

n8n, Power Automate or Make: which tool when.

The tool follows your environment, not the other way round. What matters more than the name: licence model, data location and who will still understand the flow in three years.

  • n8n, self-hosted

    Our default

    Self-hosted, Sustainable Use License

    Fits when
    Our default when operation should stay in Germany or on your own servers. Flows can be versioned as files, and custom code fits in at any step.
    Mind
    Source available, but not an open source licence: free for internal flows, offering n8n as a paid service to third parties is excluded. Some enterprise features need a licence.

  • Make

    Managed, cloud

    Fits when
    flows should be up quickly without your own server and the business team wants to help maintain them. Visual editor with many ready-made modules.
    Mind
    Billed in credits per executed action, so volume drives cost. Hosted, according to Make, on AWS in the EU or North America.

  • Microsoft Power Automate

    Managed, Microsoft 365

    Fits when
    your flows live in Microsoft 365, SharePoint and Teams. For legacy applications without an interface there are desktop flows that operate the screen.
    Mind
    Premium connectors and unattended desktop flows need their own licences (Premium or Process). Do the maths first.

  • Zapier

    Managed, cloud, US provider

    Fits when
    simple flows between widely used cloud tools are needed. A very large number of ready connectors, quick to set up.
    Mind
    Billed by tasks. Rarely the first choice for ERP and on-premises servers; clarify data location first.

  • Camunda and other workflow engines

    Process engine, BPMN

    Fits when
    a process runs long, has many parties, deadlines and escalations, and should be documented and audited as a BPMN model.
    Mind
    A development project, not a click tool. Since version 8.6, Camunda 8 self-managed needs an enterprise licence for production.

  • Workflow module of your ERP

    Inside the existing system

    Fits when
    the flow starts and ends in the ERP, such as approving orders or invoices. No extra platform; permissions and data are already there.
    Mind
    Stops at the system boundary. As soon as mailbox, CRM or a specialist application join in, a platform in between is needed.

Where it is used.

Tasks and limits.

What the system takes on

  • Starts on an unambiguous trigger and handles every case the same way
  • Checks rules: order number, amount within tolerance, customer known
  • Uses AI only where free text, a PDF or a photo has to be read, with confidence and a threshold
  • Runs alongside the manual process first; it takes over only once results match
  • Shows what went through and what is stuck, and alerts when a run does not happen
  • Versions every workflow and logs every write

What your team takes on

  • Approves anything that moves money, touches contracts or goes out
  • A named person works the clarification list
  • The business team changes tolerances and rules, not the workflow
  • Decides the cases that differ every time
  • Rare processes, and those just before a system change, stay manual

What it connects to.

Works with the systems you already use

Workflow automation

If the path is always the same and a rule can decide, it is workflow automation: predictable, cheaper, auditable. AI gets only the one step that needs judgement.

Which one when?

  • AI agents

    Rule or AI? If the path differs from case to case, it is an AI agent. Often it is a mix: the workflow moves the data, the agent takes the one step that needs judgement.

  • Data & documents

    The input is a PDF, a scan or a photo? Then Document Intelligence reads the fields the process passes on.

  • AI integration

    A system has no interface, or the question is rights and logging? Then AI integration settles the way in.

The way there.

From the first call to a system in operation. Each phase stands on its own; none ties you to the next.

  1. Intro call

    free

    30 minutes. We work out where AI fits for you and whether an AI strategy is worth it.

  2. AI strategy

    €1,990

    AI strategy for mid-sized companies: which of your processes suit AI, what the build costs and brings back, and which ones are not worth the effort.

  3. Pilot & implementation

    from €5,000

    AI pilot at a fixed price: one use case, fixed scope and acceptance criteria in writing up front. Built in your systems, tested with your data.

  4. Operation & improvement

    from 20% a year

    Running AI systems: hosting in Germany, monitoring, model and prompt maintenance, ongoing measurement, and the next use case on the same foundation.

Works together with.

  • Building block

    AI agents

    AI agents for companies: the agent reads the enquiry, looks things up in ERP and CRM and creates a draft. What leaves the company, your team approves.

  • Building block

    Data & documents

    Data preparation for AI: we bring data from ERP, CRM, spreadsheets, mailboxes and file shares together, clean it, resolve duplicates and keep it current.

  • Building block

    AI integration

    AI integration where the work happens: in Outlook, Teams, the ERP and the CRM, with your permissions and your log. Your systems stay, old ones included.

Frequent questions

Common questions about it.

The questions that almost always come up.

What is workflow automation?

A fixed process between your systems that runs on its own: a trigger starts it, rules set the steps, and where an approval is needed it waits for a person. If something is missing or a step fails, it stops and reports.

Are you an n8n agency?

We set up and run n8n, and it is our default when processes should run in Germany or on your own servers. Every client gets their own self-hosted instance, in a German data centre or in your infrastructure. We build the flows, version them as files and, if you want, take on updates, monitoring and maintenance. For internal processes n8n is free to use; some enterprise features need a licence.

When do you not use n8n?

If your processes live in Microsoft 365, SharePoint and Teams, Power Automate is usually closer. If a process starts and ends in the ERP, its workflow module is enough. Long processes with many people, deadlines and a BPMN model belong in a workflow engine such as Camunda. And anyone who wants to sell flows as a paid service to third parties may not do so with n8n under its licence.

Does workflow automation need AI?

Not always. Many processes can be automated with rules, and then that is the better choice: cheaper, faster and traceable. We add an AI step only where free text, documents or a judgement call are involved.

Which tool do you use for workflow automation?

Whatever fits your environment. Power Automate is the obvious choice if you work in Microsoft 365; n8n if it should run on your own servers or in Germany; Make for quick cloud processes. Where none fits, we build the process as a service of its own.

What about systems with no interface?

Every business has them. We work through the options in order: documented interface, database or file export, and only then automation at the user interface. The last option costs the most to maintain, which is why it comes last.

What happens when a step in the process fails?

The process stops and the case goes to a named person in the state it stopped in. Automations that quietly carry on after a failure create work rather than removing it.

How do we know an automated process is running?

From an overview showing what went through, what is stuck and how long it took, and from an alert when a run does not happen. Without that, a stalled process is noticed when somebody phones.

Do the automated processes run in Germany?

Yes. Processing in German data centres or in your own infrastructure, with a data processing agreement under Article 28 GDPR before any data flows. AI steps use only model access where the provider contractually commits not to train on your data.

The next step

Let us find out where AI pays off for you.

30 minutes, free and without obligation.

Free intro call