Norbert Sowinski Freelance web developer

AI automation

Automate repetitive work without losing control of the process.

Automation is most useful when it removes routine hand-offs and data movement while keeping the important decisions visible. I build workflows that connect the tools you already use and add AI only where it contributes something concrete.

The process is mapped first: what starts it, which rules are deterministic, where human review is needed and what should happen when a step fails.

Practical automation

Reduce manual steps without automating blindly

A reliable workflow makes routine work disappear while preserving the checks that protect quality and accountability.

Move information between tools

Data can flow from forms, email, CRM or other systems without repeated copying.

Structure incoming data

Automation can clean, classify and route information before someone needs to work with it.

Keep important decisions visible

Human approval can stay in the loop where judgement or accountability matters.

Typical workflows

Automate the repetitive parts of everyday processes

The strongest candidates are tasks that happen often, follow recognisable rules and consume time without requiring much judgement.

The workflow can then process the information, call another service, use AI where it is genuinely useful, save the result and notify the right people.

Lead qualification

Enrich, classify and route incoming enquiries before the right person follows up.

Document processing

Extract useful fields, rename files, create records or prepare structured summaries.

Support requests

Categorise incoming messages and prepare the next action or draft response.

Tool synchronisation

Keep CRM, spreadsheets, databases and SaaS platforms aligned automatically.

What the project covers

Use AI only where fixed rules are not enough

Many automation steps are better handled by normal logic. AI is added for tasks that genuinely involve language or fuzzy interpretation.

Classification

Route messages, documents or requests based on their meaning rather than fixed keywords.

Extraction

Pull structured fields from natural-language text or semi-structured documents.

Summaries

Create concise working summaries for people who need the important points quickly.

Draft preparation

Prepare responses or internal notes for review instead of sending them automatically.

Analysis

Compare or interpret text according to a defined business question.

Assistance

Use an AI step inside a larger workflow without giving it unrestricted control.

Tools & integrations

Orchestrate your tools in one workflow

Depending on the project, I use n8n to orchestrate the different steps of a workflow and connect applications that expose an API or a suitable connector.

When a standard connector is not enough, custom code or a dedicated API integration can fill the gap.

n8n

Workflow orchestration, triggers, conditions and connections between services.

APIs

Direct connections to tools when their data or actions are available through an API.

Webhooks

Trigger the workflow as soon as an event occurs in another system.

Custom code

Additional logic when standard connectors do not cover the requirement.

Process

Understand the process before automating it

The quality of the workflow depends more on the process design than on the number of nodes in the automation tool.

  1. 01

    Current process

    We map the trigger, manual steps, data sources, exceptions and desired result.

  2. 02

    Workflow & rules

    Deterministic logic, AI steps and human approvals are separated clearly.

  3. 03

    Integration & tests

    The workflow is connected to the required tools and tested with realistic edge cases.

  4. 04

    Deployment & follow-up

    Logging, credentials and operational ownership are set up so the automation remains understandable.

Control & reliability

Important automations should remain observable

A workflow can depend on external services that become temporarily unavailable, receive unexpected data or run into a case that was not anticipated.

Important processes therefore need error handling, logging and a clear way to identify operations that require attention.

Depending on the workflow, that can include:

  • • data validation;
  • • operation logs;
  • • controlled retries;
  • • notifications when a step fails;
  • • limits on AI actions;
  • • human approval;
  • • the ability to resume the process manually.

Automation & chatbot

A workflow can also start with a conversation

An AI chatbot can collect information from a user before passing it into an automated workflow. The assistant handles the interaction while the automation performs the steps required behind the scenes.

This combination can work well for support, sales qualification or processes that begin with a request expressed in natural language.

Example flow:

  1. 1. the user explains the request;
  2. 2. the chatbot collects the useful information;
  3. 3. the data is structured;
  4. 4. the workflow is triggered;
  5. 5. the CRM or business tool is updated;
  6. 6. the relevant team receives the request.
Explore AI chatbots

Existing process

Already have a workflow but some steps are still manual?

You do not need to automate an entire process in one go. Individual steps can be improved first when they already account for a meaningful amount of repetitive work.

A review of the current process helps identify where data is re-entered, which actions are repeated and which decisions could reasonably be assisted by AI.

The work can focus on:

  • • a recurring manual task;
  • • synchronisation between two tools;
  • • an existing n8n workflow;
  • • a classification step;
  • • document processing;
  • • a notification or approval step;
  • • adding an AI step to an existing process.

Frequently asked questions

Questions about AI automation

Which tasks can be automated with AI?

AI can be used in workflows to classify information, extract data, summarise content, prepare responses, analyse requests or route a task to the right step. It is most useful where the work involves interpretation that fixed rules handle poorly.

What is the difference between traditional automation and AI automation?

Traditional automation mainly follows deterministic rules: when a condition is met, a defined action runs. AI can be added when a step needs to understand text, classify a request, extract information or generate content. Both approaches can work together in the same workflow.

Can you connect the automation to my existing tools?

Yes. Workflows can connect to CRM systems, web applications, databases, email, forms and SaaS platforms through APIs, webhooks or available connectors.

Do you use n8n for automation?

Yes, when n8n fits the project. It is useful for orchestrating workflows, connecting services and adding custom steps. Some projects still need additional development or a different architecture.

How do you prevent an AI automation from taking the wrong action?

Sensitive steps can be protected with validation, deterministic rules, action limits or human confirmation. AI does not have to make every decision on its own; it can prepare or classify information before another rule or a person approves the next step.

How does an AI automation project get started?

We start by mapping the current process: trigger, manual steps, tools, available data, exceptions and expected outcome. Then we identify which parts can be automated and which should remain under human control.

Your project

Have a repetitive process you want to simplify?

Describe what starts the process, which tools are involved and where people currently spend time copying, checking or routing information. We can identify the parts worth automating first.