Most companies no longer need to be convinced that AI will affect the way they work.
The more useful question is:
Where does it actually make a difference in day-to-day operations?
A powerful AI model on its own does not automate a business process.
It may answer a question, but it does not create a lead in your CRM, assign a task, check a calendar, route an invoice or make sure the right person receives the information.
That is where we use n8n at Bergstone.
For us, n8n is the layer that connects AI, data and the software a company already uses. We use it to build chatbots, internal assistants, business process automations and content workflows.
What is n8n?
n8n is a workflow automation platform.
In simple terms, it allows different systems to work together.
For example:
A potential customer submits a form on a website. n8n receives the information, identifies the service they are interested in, creates a contact in the CRM, assigns the enquiry to the right person and sends an appropriate response.
Or:
An invoice arrives by email. A workflow detects the attachment, extracts the relevant information, associates the invoice with the correct project and sends it to someone for approval.
None of this sounds particularly dramatic.
That is precisely the point.
Automation creates the most value when it handles the small tasks that happen every day and consume a few minutes every single time.
Why we use n8n
There are plenty of workflow automation platforms available.
We tend to use n8n for three main reasons.
We can control the infrastructure
n8n can be run on your own infrastructure, for example on servers located in Germany or elsewhere in the EU.
That matters when workflows handle customer information, invoices, internal documents or conversation histories.
For sensitive projects, we can also combine n8n with locally hosted AI models. In that setup, document analysis and language-model processing can remain inside the same infrastructure.
Not every workflow needs this level of control.
But when privacy matters, we want the architecture to make it possible.
Costs do not increase with every tiny action
Many automation platforms charge according to the number of tasks or operations they execute.
That is barely noticeable during a small test.
It becomes much more relevant once workflows are running thousands of operations every day.
With a self-hosted n8n installation, running costs are mainly determined by the underlying infrastructure rather than by every individual step in the workflow.
For larger automation projects, this can make costs considerably more predictable.
We are not limited to standard integrations
n8n already includes integrations with many popular tools.
But the interesting cases usually involve systems that do not have a ready-made connector.
If a system exposes an API, we can normally connect it anyway.
That matters particularly for established businesses, where the software environment is rarely made up entirely of modern cloud applications.
In reality, Microsoft 365 may need to communicate with specialist software installed ten years ago, an accounting platform, a telephone system and three Excel sheets that everyone is terrified to touch.
Automation has to survive that reality.
Chatbots that keep working after the conversation ends
Many chatbots have one fundamental weakness:
They talk a lot, but they do very little.
For us, a chatbot becomes useful when it becomes part of an actual business process.
Imagine a potential customer asking through the website, WhatsApp or another messenger:
“How much would it cost to survey our building?”
The assistant can ask the obvious follow-up questions: Where is the building? How large is it? What kind of data is required? When does the survey need to happen?
But the process should not end with a polite chat response.
The workflow can structure the enquiry, create the lead in the CRM, assign it to the appropriate person, check availability and prepare the information required for a quotation.
If someone sends an enquiry at 10 p.m., nobody has to respond immediately.
But by the next morning, the company can already have a structured lead with the conversation history and the most important project details.
That is where we see the real value of an AI assistant.
Not in how convincingly human it sounds.
But in what happens after the conversation.
The best automations are often the boring ones
When we analyse processes inside a company, we do not start by searching for the most impressive AI use case.
We look for repetitive work.
Where are people copying data between systems? Where is information being manually extracted from emails? Where does someone remember every Friday that a spreadsheet needs to be updated? Which five steps does somebody repeat for every new customer or project?
Those are usually the areas with the greatest potential.
Some typical examples include:
Lead processing
Enquiries from websites, advertising campaigns and messaging platforms can be collected, structured and automatically assigned to the right employee or project.
Invoice preparation
Incoming invoices can be detected, analysed, associated with the appropriate project and then presented to a person for approval.
Project status updates
When the status of a project changes, a workflow can automatically create tasks, notify stakeholders and move documents to the appropriate location.
Follow-up processes
If a quotation remains unanswered for a defined period, the system can create a reminder or follow-up task.
AI is not always the most important part of these processes.
Sometimes a well-designed workflow is all that is required.
That is worth saying because the current fashion seems to be adding a language model to absolutely everything, including problems that were already solved perfectly well by an if statement twenty years ago.
Content as a workflow rather than a permanent unfinished task
Many companies have another recurring problem.
They want to publish regularly.
There are enough projects, experiences and ideas to talk about. Yet weeks go by without anything being published because nobody has the time to turn those raw materials into a blog article, LinkedIn post or newsletter.
This is where a content workflow can help.
The process starts with real material from the business: customer questions, completed projects, recurring problems, new technologies or developments in the industry.
A workflow can collect this material and prepare a first draft.
That core piece of content can then be adapted into different formats:
- a detailed website article,
- a shorter LinkedIn post,
- several social posts,
- a newsletter section, or
- a script for a short video.
For us, one part of this process remains deliberately manual: publication approval.
We automate production, not responsibility.
Before anything is published, a person reviews the facts, wording and context.
That also prevents a common failure mode of fully automated content systems: producing enormous quantities of perfectly acceptable text that absolutely nobody wants to read.
One idea, multiple channels
The real advantage of a content workflow is not simply that AI can write faster.
It is reuse.
One strong project report can become:
- a technical article for the website,
- a LinkedIn post aimed at decision-makers,
- a short visual post for Instagram, and
- a section in the next company newsletter.
Each channel gets an appropriate version, while the underlying information remains the same.
That can reduce editorial effort considerably because the team does not need to start from zero every time something needs to be published.
We use these systems ourselves first
One principle at Bergstone is that many of the workflows we recommend to clients are first used inside our own ventures and projects.
Our automations process our own leads, projects, documents and content.
That creates a useful reality check.
We quickly discover which ideas look impressive in a presentation and which ones survive actual day-to-day use.
A workflow that constantly needs to be manually corrected after two weeks is not really automation.
It is just a new category of work.
How an automation project starts
We rarely begin with a large platform.
We begin with one process.
First, we look at how that process works today.
Then we identify where time is being lost, mistakes happen or information gets stuck between systems.
We build a small first workflow and put it into real use.
If it works and creates measurable value, we expand it.
If it does not, we change it or remove it.
This incremental approach is usually faster, cheaper and less risky than attempting to automate an entire organisation in one go.
Which processes are worth automating?
A good automation candidate usually has at least one of these characteristics:
- It happens frequently.
- It follows reasonably clear rules.
- People transfer information between multiple systems.
- Copy-and-paste causes mistakes.
- Or someone regularly says: “Surely this could happen automatically.”
That is normally a good place to start looking.
Frequently Asked Questions
What is n8n used for?
n8n is used to connect software systems and automate workflows between them.
A workflow can react to an event such as a new enquiry, an incoming email, a changed project status or a new document and then perform a series of predefined actions automatically.
Typical examples include CRM automation, invoice processing, lead qualification, reporting, document workflows and AI-assisted customer communication.
Is n8n an AI platform?
Not primarily.
n8n is a workflow automation platform, but it can connect to AI models and make them part of a larger business process.
That distinction is important.
The AI model may analyse a document or understand a customer request, while n8n handles everything around it: retrieving data, creating records, triggering approvals and sending the result to the right system.
Do we need AI for automation?
No.
Many valuable automations do not require AI at all.
If a process follows clear rules, conventional workflow logic is often more reliable, cheaper and easier to maintain.
We add AI when the workflow needs to understand unstructured information such as emails, documents, natural-language requests, images or voice recordings.
Can n8n be self-hosted?
Yes.
n8n can be deployed on your own infrastructure or on infrastructure managed specifically for your organisation.
This can be useful when a company wants greater control over data location, access, backups and integrations.
Can n8n workflows be GDPR-compliant?
Yes, but GDPR compliance does not come automatically from using a particular tool.
It depends on how the entire system is designed.
This includes where n8n is hosted, which external services receive data, what information is stored, how long it is retained, who has access and whether appropriate data-processing agreements are in place.
For sensitive use cases, we can design workflows so that relevant processing remains on infrastructure in Germany or the EU and, where appropriate, use locally hosted AI models.
Can you connect n8n to our existing software?
In many cases, yes.
If the software provides an API, webhook, database interface or another structured way to exchange data, it can usually be integrated.
For systems without a standard n8n connector, we can often build the connection directly through their API.
Older or closed systems need to be assessed individually.
Can n8n work with Microsoft 365, CRM and accounting software?
Yes.
Typical projects involve systems such as Microsoft 365, Gmail, CRM platforms, databases, accounting tools, project-management software, messaging services and internal applications.
The exact integration depends on the APIs and permissions available in the customer’s existing environment.
Can a chatbot create leads and appointments automatically?
Yes.
A chatbot can collect information from a potential customer, structure the enquiry and pass the data into a workflow.
The workflow can then create or update a CRM contact, notify the responsible employee, check available appointment slots or prepare the information required for a quotation.
For important actions, we normally include explicit human approval rather than giving the chatbot unrestricted autonomy.
Can AI process invoices and documents automatically?
Yes, within clearly defined limits.
AI can help classify documents, extract information and understand unstructured text.
But financial workflows should not rely on a language model alone.
For example, AI might identify the supplier, invoice number and project reference, while deterministic software handles calculations, validation and approval rules.
Can content publishing be fully automated?
Technically, a large part of it can be.
That does not mean it should be.
We usually automate research assistance, first drafts, formatting, adaptation for different channels and distribution preparation.
Final approval stays with a person.
The goal is to reduce repetitive editorial work, not to build a machine that publishes unchecked content at industrial scale.
How long does it take to build an automation?
That depends heavily on the process and the systems involved.
A simple workflow connecting a few existing tools can often be implemented relatively quickly.
A more complex process involving legacy software, AI models, approvals, custom APIs and sensitive data requires more architecture, testing and monitoring.
For that reason, we normally start with one clearly defined process rather than attempting a company-wide automation project immediately.
How much does an n8n automation cost?
There is no meaningful flat price because two workflows that look similar from the outside can require very different amounts of integration work.
The main cost drivers are the number of systems involved, API availability, process complexity, security requirements, AI usage, testing and the amount of custom logic required.
We therefore scope the process first and then provide a project-specific estimate.
Will we depend on Bergstone after implementation?
That should not be the objective.
We prefer architectures that the customer’s own team can understand and operate.
Depending on the project, this can include documentation, access to the workflows, technical handover and training.
More complex systems may still require ongoing maintenance, but the customer should not be locked into us simply because nobody else can understand what was built.
The right first step
Automation does not start with n8n.
And it does not start with AI either.
It starts with one very simple question:
Which repetitive task is currently costing your company unnecessary time?
Once that is clear, you can decide whether a workflow makes sense, which systems need to be connected and whether AI is even required.
Sometimes the right solution is a sophisticated AI workflow.
Sometimes it is ten carefully configured automation steps without a language model anywhere in sight.
Both are perfectly valid.
The important thing is not how impressive the architecture looks.
The important thing is whether there is less manual work left at the end.