Anton Yurchenko. Fourteen years in duty-free distribution across the Baltics, my own company since 2015, an INSEAD MBA, and a habit of building a system for whatever costs hours. Today the company runs in under 40 hours a month, my customers still talk to a person, and the rest of my time goes into AI animation for children.
Not a consultancy. One person, a few companies, and the same discipline applied to each: automate the work, never the understanding.
Three years with a duty-free distributor serving diplomats across Estonia, Latvia and Lithuania. Alcohol and tobacco, bonded warehouses, customs paperwork: an industry that runs on documents. I learned it from the inside before building my own.
Digitising a cemetery archive in Riga: every grave photographed, inscriptions read by OCR in four languages, checked by native readers. Field work time fell 40%, and I learned that an accuracy number means nothing until a person who understands the material has looked at the output.
I founded a duty-free and diplomatic distribution business of my own in the Baltics. Five people, the usual paperwork, and from the first year the question that has driven everything since: which of these hours does a person actually need to spend?
For a research centre I designed the data architecture of a database that unifies many cemeteries and hundreds of recorded interviews. The platform was finished by their own team; the architecture I laid down is still in daily use.
I built a service marketplace in Estonia that aggregated the records of about 70% of the country's cemeteries and automated search, ordering and fulfilment. About three times faster than anyone else. The automation worked; the business model did not.
After a partnership ended I rebuilt the distribution company from scratch and never rehired the five roles it used to take. Website, warehouse and purchasing connected; invoicing scripted; partner emails generated; marketing and accountant reporting automated; analytics that never existed before. Every customer conversation is still mine.
Claude-based forecasting for an electricity trading company in Ukraine: the imbalance on solar power station forecasts fell from 40–50% to 8–10%. And a new venture, WOOPSEE: AI-native animated content for children, meant to grow into their AI assistant and a parenting tool.
The clearest example from my own company. Minutes per invoice, by hand in 2019 and by the system today.
| Step | By hand | By the system |
|---|---|---|
| Order created and checked against stock and prices | 20 min | 1 min |
| Documents compliant with warehouse, bookkeeping and customs | 40 min | 1 min |
| A person reads it before it goes out | — | ~2 min |
| Per invoice | ~60 min | ~4 min |
Marketing and accountant reporting each took two working days a month; both now run themselves. About 37 hours of admin a month removed from invoicing, marketing and reporting alone.
Work done before 2022 was automation and IT; the AI came later, on top of the same discipline.
Website, warehouse and purchasing connected into one system. Invoicing scripted and reconciled against stock. Partner emails generated. Marketing and accountant reporting each two working days a month lighter. Financial analytics that did not exist before.
Still human, on purpose: every customer conversation.
An electricity trading company forecast the output of solar power stations by hand, with a 40–50% imbalance between forecast and actual generation. Claude-based forecasting plus automation of the processes around it, inside the tools the team already used, brought the imbalance down to 8–10%.
Reviewed by a trader before anything goes out. That is the design, not a caveat.
A fragmented phone-and-paper market for grave care, headstones and renovation. Records of about 70% of Estonian cemeteries aggregated, then search, ordering and fulfilment automated across the whole pipeline.
Still the fastest way anyone has processed that kind of order in Estonia.
Burial books digitised, every grave photographed, a field workflow system for the photographers, OCR of inscriptions in four languages verified by native readers. Then the data architecture for a research centre's database.
Architecture I laid down in 2017 is still in daily use.
Already in 2014 the OCR was 90% accurate. That number told us nothing. Was it nine stones in ten read perfectly, or one wrong letter in every tenth character across half the records? Until a person who understood the language looked at the output, nobody knew where the 10% was.
I met the same thing again in 2026, building solar forecasts with a language model for an electricity trader. The output was confident every time; only the trader could tell which forecast was plausible. So the system drafts, and the trader signs.
Every system I have built since 2014 follows that rule: automate the work, never the understanding. It is why, in my own company, the customer conversations are still mine.
The trade press talks about AI agents shopping on the traveller's behalf. The AI that paid for itself this year sits behind the counter.
Agents running the back office unsupervised. Not this year. What works is narrow: one process, inside the tools you already use, with a person who signs off.
"Our data ends up in ChatGPT." Only on free personal accounts. Business API access keeps it out of training; the contracts say so.
"It makes things up." It does. So don't ask it to be trusted; ask it to be checked. Every figure my invoicing produces is reconciled before it leaves.
WOOPSEE is my new venture: AI-native animated content for children that is meant to grow into their own AI assistant and a tool for parents. There is no website yet, only the animation. These are working clips: an alpaca, a dog called Monti, a bear called Mischa, Papageno, an excavator, a crab and an octopus, a bear and a fox.
Alongside WOOPSEE I take part in AI film and advertising competitions, including both Runway AI ad contests, mostly to learn what the tools can and cannot do yet.
I take on a few projects a year for companies like mine: operations-heavy, small teams, too much paperwork. No packages, no platform to buy.
We map where the hours go and pick the processes worth automating, with a payback estimate for each.
One process at a time, live in weeks, inside the tools you already use. A person signs off before anything reaches a customer or a partner.
I maintain it, adapt it as the business changes, and move on to the next process, or hand it over to your team.
Founder and operator. Fourteen years in duty-free and diplomatic distribution across the Baltic states, the last eleven running my own company; alongside it a startup and a string of IT and automation projects. When I rebuilt the company in 2025, I automated it until it needed only me.
INSEAD Global Executive MBA, class of 2023. Based in Vienna. I work in English, Russian, Ukrainian, Latvian and Spanish.
Lynstone is the name I work under for projects with other companies. It is not an agency: the engineers on a project are senior and assembled for that project, and you deal with me from the first call to the last handover.
Operations, AI, animation for children, duty free, or something I haven't thought of yet. If I can help, I will say how; if I can't, I will say who might.