Process automation
I turn manual, repetitive tasks — moving data between spreadsheets, preparing reports, processing documents, syncing systems — into scheduled, traceable, maintainable workflows.
Updated:
The same data is typed into the ERP, a spreadsheet and an email. Every manual step is a chance for error.
Someone spends hours gathering data from several sources, formatting it and sending the same summary.
Invoices, delivery notes or forms arrive by email or as PDFs and someone keys in the data by hand.
If that person is away, the process stops. Nobody knows exactly how it is done.
A useful automation is usually small and concrete. For example: every Monday at 7:00 a process downloads the exports from the online shop and the ERP, matches them by reference, calculates stock differences and sends a summary to purchasing. It used to be two hours with three spreadsheets open; now it is an email. The value is not the technology, but that the process no longer depends on someone remembering and not making a mistake.
I mainly use Python for its ecosystem for data, spreadsheets, PDFs and APIs. Each automation includes activity logging, validation of input data and a dry-run mode that changes nothing, so you can check the result before going live. As a process grows, I structure it with the same discipline as a data pipeline: separate steps, tests and orchestrated execution.
A practical rule: multiply the time each run takes by the number of runs per year and by the hourly cost; compare it with the effort of building and maintaining the automation and with the cost of the errors it would avoid. If the result is doubtful, the answer is usually not to automate yet, or to automate only part of it. I would rather tell you before we start.
Python scripts or services with scheduled execution (daily, weekly, event-based), activity logs and error handling.
Connections between the tools you already use (online shop, ERP, CRM, email, spreadsheets, storage) where they offer an API or export.
Checks that the data is consistent and clear notifications (email or message) when something fails or does not add up.
What each automation does, where it runs, how to restart it and how to change it, written for non-programmers.
We write down the real steps, exceptions and who is involved. Exceptions are what shapes the design most.
I weigh frequency, time and error cost against the effort of building and maintaining. Some parts stay manual on purpose.
I automate the main path first with real data and run it alongside the manual process to compare results.
Scheduled execution, a log of every run, failure notifications and a follow-up period.
Documentation and a hands-on session so you know what to do if something changes or breaks.
Consolidate several Excel files or exports into one source, apply rules and produce the final sheet ready for review.
Pull data from several sources, calculate indicators and send a PDF or sheet to whoever needs it, always at the same time.
Read PDFs and emails, extract amounts, dates and references, compare them with orders and leave a list of discrepancies.
Keep stock, prices or customers aligned between the online shop, the ERP and working sheets.
Match movements against internal records and flag what does not add up for a person to review.
Reminders, backups of business data or file clean-up following clear rules.
Those that repeat often, follow reasonably stable rules, take noticeable time and cause errors or delays when done by hand. A simple way to estimate it is to multiply time per run by runs per year by hourly cost and compare that with the cost of building and maintaining the automation.
Usually not. I start from what you use (Excel, email, your ERP or online shop) and add the automation around it. Changing tools only makes sense when they are the real bottleneck.
It depends. They can run on a small server or cloud service, on a computer in your office or as scheduled tasks on your hosting. I recommend the simplest and cheapest option that meets the requirements.
Each automation logs its activity and alerts on failures. Format changes at the source are the most common cause, and they are documented so they can be fixed quickly — by me or by whoever you choose.
You can see my published projects on this site. Client work is always described at a high level, without identifying data, for confidentiality.
Data engineeringPersonal project
A personal lab that recreates an analytics engineering environment on synthetic financial-crime data, with base, core and marts layers, quality tests and orchestration.
Machine learningPersonal project
A system combining computer vision and data processing to identify and classify items and support inventory analysis.
Find where AI adds value in your operations and take it to production with human oversight.
View serviceAn assistant that answers from your documents, cites its sources and respects permissions.
View serviceReliable pipelines, clear models and quality tests so you can decide with data you trust.
View serviceForecasts and classifications honestly evaluated against simple alternatives.
View serviceA first conversation is enough to know whether it makes sense and where to start.