02 · Services
Automation and AI
The hours a team spends retyping documents, copying data between systems or reporting by hand can be reclaimed. We build automation that runs reliably for years and wire in AI so it helps rather than decides.
We start by describing the process as it runs today: who does what, where it waits and where errors appear. Then we propose what to turn into code and what to leave to people. We build automation as queues with a visible state for every item: to process, processing, manual, failed, done. Nothing may silently disappear. We use language models where they read, sort and propose, for example turning a document into a booking entry or classifying an enquiry and drafting a reply. A human confirms. That is how we built a bookkeeping robot, the AI account audit in ClientDeck and our own cookie banner, which cut deployment on a new site to minutes.
- Describe the process as it runs today and measure the time
- Design: what goes to code, what to AI, what stays with people
- Prototype on real data
- Queue with visible states and a log of every step
- Deployment, monitoring and evaluation of time saved
- Automation in a repository that is yours
- An admin with the state of every item
- Integration with your systems via API
- A log that lets you trace every decision
- Time saved quantified after the first month
What people ask most
What is worth automating first?
A process that repeats daily, has clear rules and is done by a person in a spreadsheet today. Typically documents, enquiries, reports, feeds, post-purchase e-mails. Processes where every case is handled differently we leave to people and automate only their preparation.
When does AI belong in a process and when not?
AI belongs where the input is unstructured: a photo of a document, an e-mail, the text of an enquiry. There it reads, sorts and proposes. It does not belong where money or a customer is decided on without a check. So with us AI only proposes and a human approves, except for cases you explicitly configure as automatic.
What if the automation goes down?
Every item has a state and every step a log, so after an outage it continues from where it stopped and nothing is processed twice. Idempotence is a rule for us, not an option. The worst automation failure is a double send or a double booking.
Do you work with Make or Zapier?
We do, and for a simple link between two services we will recommend them. Custom code makes sense when the process is more complex, the volume larger, or when you need to keep data and logic in-house and not pay per operation.
Related projects
Products we run every day
Účtobot: bookkeeping that books itself
A layer on top of the accounting software: I forward or photograph a document, AI reads it and pre-books it from templates, I only approve. It watches the correctness of the books, VAT, cash flow and prepares filings.
ClientDeck: a campaign control system
Our own SaaS that brings Google Ads, Meta, Sklik and comparison engines into one control system with alerts and an approval queue.
Our own cookie banner with Consent Mode v2
A banner that sets Consent Mode v2 before any other tag, proves consent and deploys on any site with no licences.
Other services
Next.js websites that are fast, readable for search engines and AI, and ready for campaigns from day one.
More about this service →03Applications and internal toolsCustom systems where off-the-shelf software falls short: control panels, admin interfaces, tablet apps, SaaS.
More about this service →04Integrations and trackingE-shop, CRM, ad accounts and analytics connected so the numbers match orders. Consent Mode v2, server-side tracking, feeds.
More about this service →05Hosting and operationsGit deploys with health checks and automatic rollback. Updates, backups, monitoring. On our panel or on your infrastructure.
More about this service →Contact
Have a project? Write to us.
A few sentences about what you are dealing with is enough. We reply within one working day with questions and a proposed approach.