Workflows that run themselves

What repetitive manual work costs your organisation

Many businesses run workflows every day or every hour that follow the same steps: pull data from system A, modify it, push it to system B, send a confirmation. Someone does this manually, makes mistakes, forgets once, or simply burns out on it.

The real cost of that kind of work is not just in the hours. It is also in the errors that surface three steps later, in delays during peak load, and in context lost when that one employee is unavailable. I see this often in growing SMBs: at some point manual data management starts to pull the organisation sideways.

Automation solves this by removing the repetitive steps from human hands and placing them in a reliable, logged system. The result is not just time saved but predictability: the process works the same on Monday as it does on Friday afternoon.

n8n, custom script or AI-orchestration: when to use which

The right tool depends on the complexity of the logic, not on what sounds most impressive. I work in three layers:

• ▸n8n: when the workflow runs on integrations between existing tools. Think: new order in WooCommerce triggers a Slack message, a PDF upload sends an email to a client, or a CRM update syncs with your accounting software. n8n has hundreds of ready-made connectors and your team can adjust it after handover without writing code.
• ▸Custom Node.js script: when the logic is too specific for a visual workflow tool, or when performance matters. Think batch processing of large datasets, complex transformations, or integrations with APIs that have no standard connector. A script gives full control over error handling and retry logic.
• ▸AI-orchestration: when the decision in the middle of the workflow is context-dependent and cannot fit in an if-else. An AI model reads the input, decides based on context which action is needed, and drives the rest of the process. This is the heaviest layer and I only use it when the business case justifies it.

I always present the choice with the trade-offs included. No overkill, no underestimation.

My approach to an automation project

I start with reality, not with wishful thinking. That means I first find out which processes are actually repetitive, how much variation exists, and where the pain is greatest. I do this by looking at existing tools, asking who does what, and looking at where things go wrong.

I then propose a priority order based on impact per build unit. I work iteratively: each step delivers something working that you can test immediately. No intermediate deliverables that are only usable when everything is done.

• ▸Workflow audit: inventory of repetitive tasks, data flows and tool integrations
• ▸Prioritisation based on volume, error risk and implementation complexity
• ▸Minimal working automation per use case, verified with real data
• ▸Error handling and notifications built in from the first version
• ▸Handover with documentation so your team can manage or adjust it
• ▸Optional monitoring retainer for critical workflows

Examples of what I build

Not abstract possibilities, but concrete automations I have built or build regularly:

• ▸CSV-to-database pipelines with validation and deduplication, including error reporting
• ▸Email-trigger actions: incoming messages are classified and forwarded to the right system or person
• ▸File processing with context-aware logic: read, modify based on content, export to target format
• ▸Multi-tool synchronisation between CRM, ERP, accounting and communication tools
• ▸Data validation pipelines that check input before processing and flag deviations
• ▸Webhook receivers that translate external events into internal actions in your tools
• ▸Periodic reports assembled from multiple sources, automatically sent

What I do not do

Automation is not a sales argument in itself. I do not build solutions that sound impressive but do not solve a measurable problem.

• ▸No Zapier-only setups without code control: if you have no grip on the logic, you have no grip on what goes wrong
• ▸No marketing automation without a business case: drip campaigns are marketing, not operational automation
• ▸No AI features as a gimmick: I use AI only when the task requires context-aware reasoning that a script cannot handle
• ▸No solutions where you are fully dependent on me afterward: I always deliver source code and documentation as your property

Who this is suitable for

Automation is most effective when there is a clearly repetitive process that costs time and is prone to errors. I see this most often with:

• ▸SMBs that are growing and find that manual data management does not scale with them
• ▸Scale-ups that move data between tools manually because the integrations do not exist
• ▸Teams that spend hours per week on the same steps in the same system
• ▸Companies that want to keep control of their own processes without monthly per-task subscription costs
• ▸Sectors with high volumes of repetitive file processing, order processing or communication handling

If you are unsure whether your situation fits, a short intake is the fastest way to find out. I give an honest assessment there, including if automation is not the right next step for you.

Further reading