AI automation
AI automation takes a repetitive task that needs some judgement, like reading an invoice or sorting an email, and runs it as a workflow. A language model handles the reading and classifying, ordinary code handles the rules, and a person approves the cases the system is unsure about. We start with one process, measure the time it saves, then add the next.
An example
A PDF invoice arrives by email, the lines are read, the totals are checked and a draft entry appears in the accounting system for someone to approve.
What we build
- Document and invoice extraction
- Pulls supplier, amounts, dates and line items from PDFs and scans, checks them and posts them to your accounting system.
- Email triage
- Reads incoming mail, labels it, drafts a reply or creates a ticket, and sends urgent cases to the right person.
- Lead routing and enrichment
- Cleans up form submissions, adds company data and assigns each lead to the right salesperson in your CRM.
- Reports that write themselves
- Collects numbers from your tools every week and sends a short written summary with the figures that changed.
- Back-office glue
- Connects tools that do not talk to each other, so an order, a contract or a new hire updates every system once.
How a project runs
- Process audit. We sit with the people who do the work today and write down every step, exception and system involved.
- Pick the first workflow. We choose the task with the best mix of volume, clear rules and low risk, and agree how we will measure success.
- Build with a review step. The first version asks a person to approve uncertain cases. Approval data shows us where the automation can be trusted.
- Hand over and monitor. You get alerts when something fails, a log of every run and documentation your team can follow.
What to have ready
- A rough description of the task, how often it happens and who does it
- 10 to 20 real examples (emails, documents, forms) with sensitive data masked if needed
- Access to the systems involved, or a test account
- The person who knows the exceptions
Tools we use: n8n, Make, Python, Node.js, Claude, OpenAI, Google Workspace, Microsoft 365, REST and webhooks
Questions we get asked
Should we use n8n, Make or custom code?
For simple connections between popular tools, n8n or Make is faster and cheaper to maintain. Custom code makes sense for high volume, complex rules or systems without ready connectors. We often combine them.
What if the AI makes a mistake?
We design each workflow with confidence checks and a human review step for uncertain cases, and every run is logged so errors can be traced and corrected.
Can you work with our ERP?
If the ERP has an API, a database we can read, or even a reliable export, we can usually connect to it. We check this in the first conversation.
Where does our data go?
We choose model providers and hosting regions with you. For sensitive data we can use providers with no training on your data, or open models on your own servers.
Tell us what you need
A few sentences are enough. We reply by email with questions or a time for a call.
Other services: AI voice agents, AI integration, Web and SaaS development, Mobile app development, Blockchain development