Build
Process automation
If your team re-types the same data into three systems, that is a fixable problem. We connect the tools you already run and automate the work in between.
Build
AI integration
We connect AI to the systems you already run, so invoices, orders and emails are read and entered without anyone retyping them, and your team gets answers from your own documents. Every AI integration starts with one process and a written estimate of whether it pays back.
What this looks like in practice
- Supplier invoices read from the inbox and entered into accounting, checked before anything is posted
- Orders that arrive as PDFs or free-text emails turned into orders in your system
- Inquiries sorted, summarized and passed to the right person with a draft reply
- An assistant that answers your team from price lists, procedures and product specifications, and shows the source of every answer
- An assistant on your website that answers from your own content and hands over to a person when it should
- Store search that understands questions, not just keywords
How we approach it
- Pick one process where people spend hours on documents, emails or repeated questions
- Test models on a sample of your real data and measure how often they’re right
- Estimate the payback: hours saved against build and running costs
- Run a pilot with a person approving every result
- Hand over only the steps the pilot showed it can handle
The first three steps are an assessment at a fixed 800 EUR, credited in full against the build if you go ahead. You get the result in writing, whether or not you go ahead with us: how often the model got your own documents right, how many hours it would save, what building and running it would cost and what we recommend. If the estimate doesn’t pay the project back within a year, we’ll say so then, before you spend anything more.
What’s built in
- A person reviews anything the checks catch
- Every step is logged: what came in, what the model returned and what was done
- Assistants say they’re an AI from the first message
- Answers show where they came from, so a wrong one is quick to spot
- You know which provider and region your data goes to before anything is sent
- The model can be swapped without rebuilding the rest
- Each integration uses a fixed model version, so it changes only when we switch it on purpose
What you own
The code, prompts and configuration sit in your repository from the first commit, and the model account is in your name. If you move to another supplier, everything moves with you, the same as with custom software.
What we connect it to
Accounting and ERP: Horizon, Visma, Jumis, Microsoft Dynamics. Email: Gmail and Microsoft 365. Documents: Google Drive, SharePoint and your own file storage. E-commerce and CRM: Shopify, WooCommerce, HubSpot, Pipedrive. Plus in-house systems reached over their own APIs or databases.
Frequently asked questions
Do we need AI for this at all?
Often not. If a rule or a direct connection between two systems does the job, it’s cheaper to run and easier to trust, and we’ll build that instead. AI earns its place where the input is messy: scanned PDFs, free-text emails and documents that never look the same twice.
What happens when the AI gets it wrong?
It will sometimes, so we design for it. Extracted data is checked against rules before anything is saved: totals add up, the supplier exists, the VAT number is valid. Whatever fails goes to a person, and every step is logged. During the pilot we measure how often it’s right on your own documents, and it only runs unattended where that rate allows.
Where does our data go?
Before anything is sent, we tell you which provider it goes to, in which region and whether they keep it. We use business AI services that don’t train on what you send and come with a data processing agreement, and where a provider processes data outside the EU, the transfer is covered the way GDPR requires. If your data has to stay in the EU, mention it at the start and we’ll tell you whether your task can be done that way. Anything that doesn’t need to leave your systems stays out of the request.
Which AI model do you use?
Whichever does your task well at a sensible cost, tested on a sample of your own documents before we commit. We build so the model can be swapped, because quality and prices change every few months and you shouldn’t be tied to one provider.
What about the EU AI Act?
Most business uses, such as reading invoices or answering from a knowledge base, aren’t high-risk under the AI Act. What does apply, since August 2026, is that people must be told they’re talking to an AI, and we build that into every assistant. If what you describe could be high-risk, such as screening job applicants or deciding on credit, we’ll tell you before we scope it.
What does it cost?
The assessment is a fixed 800 EUR, credited in full against the build if you go ahead. Building one process, from pilot to production, starts at 4000 EUR and typically costs 5000-12000 EUR over 4-8 weeks. Model usage is billed by volume, on an account in your name at the provider’s own prices, the same way hosting is. We measure it during the pilot and give you a monthly estimate before you commit. After launch it sits on a maintenance plan like any other system, from 100 EUR a month, including re-testing when a provider retires a model. Prices exclude VAT.
Related services
Build
Custom software
When off-the-shelf software forces you to work in ways that do not fit, we build the thing that fits: internal tools, web applications, client portals and integrations between systems that were never designed to exchange data.
Build
E-commerce
We build online stores on the platform that suits your catalog, margins and team, and integrate them with the payment, delivery and accounting systems you already use.
Start a conversation
Tell us what you’re looking to build or improve. From the first conversation, you’ll be speaking with someone who understands the technical side of your challenge.
- Phone
- +371 29384947
- Working hours
- Monday-Friday, 9:00-18:00 Latvian time