// Agentic applications and systems for industrial, supply chain and project-based operations
Your operations at full potential.
With the people and the data you already have.
We build AI agents and agentic applications: software that connects your systems, shows your people the whole picture, works out what should be happening, and does it.
One hour
a meeting with us.
One euro
for the AI prototype.
One day
you see it running.
- Your peopleThey stop chasing data and building reports. They see what nobody could see before, and what should happen next. They build their own AI agents too — enterprise-grade ones that act in your systems, not chatbots. The person who knows the process is the one who can build it too — on the same data, so nobody ends up with their own version of the numbers. That is potential you already have.
- Your dataMost of it sits in systems nobody looks at — and the numbers that would make it useful, like real cost parameters, are in no system at all. We connect the systems and add the missing numbers, so your data can finally answer what something actually costs. You own the data, it never leaves your ecosystem, and you decide the data security level.
- Process firstSometimes the answer isn't AI. Tape on the warehouse floor and a two-bin kanban cost almost nothing and beat anything we could build. Lean principles and process first, then AI if it earns its place — and you hear that in the first hour, not after the invoice.
- Operations knowledgeOur foundation is operations and Lean Six Sigma. Any vendor can run a model over your data. The hard part is knowing whether the answer is any good. Lot sizes, safety stocks, lead times, reorder points — we know what those should look like, and what the gap is costing you every month. That knowledge goes into the agents — not into PowerPoints.
- Reliable AI, built for operationsThe black box is what raises most concern in operations, and rightly so: a wrong number here is stock that never turns, or a production line stopped. The AI is not what produces the numbers — it reads numbers that already exist, and explains them. That is how it is built, not a promise we make.
- Value, or you don't payMost business cases are written to clear the approval gate and never checked again. Ours is one sentence: if it wasn't worth it, you don't pay. And you are the one who decides that. No conditions. That has always been the deal.
Seven levels. Most companies can see two.
You know the first one, probably the second. Past that it gets vague fast — which is why AI decisions get made with no picture of what is actually possible. This is the picture: what each level does for you, and which ones your own people can build.
Level
What it does for you
Productivity
Who builds
01
AI chat / Copilot
Answers questions. You still do all the work.Most companies are here
Personal
You — we teach you how
02
Light agent
Performs one task, when asked.Most companies are here
03
Knowledge agent
Knows your data. Answers from your own numbers and definitions.
Team / process
Either — we build or teach
04
Action agent
Does the work, in your systems. You decide how far it may go.
05
Multi-agent system
Runs a whole process end to end, the way a team would.
Business
Us — you keep building on it
06
Agentic application
An application your people work in every day — not a tool they visit.
07
Agentic system
Several applications across the operation, largely running themselves. You set direction.
We eliminate the two wastes nobody measures: unused talent and unused data. The potential was always there.
// Problems
What we usually find
None of this is a system failing. Each one does its job on what it can see — and nobody can see across all of them, which is why these survive for years in companies that are otherwise well run. The first thing we do is check whether it's true of you.
01
No system holds the whole picture
The ERP has the transactions, quality has the claims, the project system has the schedule, finance has the cost. Each one is right about its own part. Then finance asks whether one forwarder invoice matches what was agreed — and the contract is a PDF in a folder, the rates are in an annex to it, and the shipments are in the transport system. Checking one invoice means opening all of them, so mostly nobody does. When somebody does, the answer is usually no.
So the version everyone argues about lives in a spreadsheet on somebody's laptop.
02
Your ERP isn't wrong. It's doing exactly what it was told.
The purchasing parameters were set once. Reviewing them properly is a real job — tens of thousands of items, and skilled work rather than something to hand an intern — so it gets started and then buried. The ERP goes on doing exactly what it was told, thousands of times a year, and the inventory ends up wrong in both directions: too much of what nobody needs, too little of what everyone does.
Excess stock and poor availability at once — one cause, paid for twice.
03
The master data is a mess
Every item needs fields somebody was supposed to fill in — a classification, dimensions, a supplier, a status. Nobody has the time, so they sit blank, the same part appears under two numbers, and stock that should have been written off years ago is still in a rack. Bills of material and routings drift the same way: the standard times were set once, and the line has worked around them ever since. It isn't neglect — at this scale, by hand, it isn't a job anyone can finish.
And every number calculated downstream inherits it.
04
Every KPI has a victim
Purchasing gets a better unit price by ordering more, from further away. The target is hit. Then inventory rises, the lead time doubles, and the quality problems turn up six weeks later in somebody else's number. Nobody did anything wrong — the targets were set function by function, so the trade-off belongs to no one.
Every number on the report is green, and the bottom line isn't.
05
Nobody knows what good looks like
Your inventory turns 2.4 times a year. Is that good? Nobody in the building can say — so next year's target becomes 2.6, an improvement on a number nobody has judged. Comparing one plant against another doesn't settle it either: they may all be poor. So the effort goes where a number is easy to move, rather than where the money is.
Without an outside reference, a metric is just a number that moved.
06
Nobody has priced it
What does it cost to store a pallet for a month? To place one purchase order? To stop the production line for an hour? Those numbers decide lot sizes, batch sizes, safety stocks and what you keep on site — and they are not in the ERP, or in any other system. Somebody estimated them once, or nobody ever did.
So the decisions that depend on them get made on a feeling.
07
Every gap got its own tool
Quality needed to log deviations, so quality got a system. Maintenance needed work orders, so it got one too. Then someone built a tool for OEE on the line. Every one of those decisions was right on the day it was made, and none of them was made with the others in view. Repeat that for fifteen years.
Now nobody can say where a number comes from.
08
Delivery ends at the door
The moment the pallet leaves the outbound door, the system books the order as delivered — and that is the last thing the ERP records. On-time delivery reads 98%. Nobody can see when the customer actually received it, or whether your own installation crew turned up the same week as the materials.
You measure at your door. Your customer measures at theirs.
// Behind this
Janne Kilpua
Over twenty years running operations across manufacturing, distribution, spare parts, retail, e-commerce and 3PL. S&OP, production planning, purchasing, transport, warehouses and distribution centres, supply chain development, construction projects, then as CLO and COO. I have been the person these problems land on. I also ran the operations side of two ERP implementations, so when I say a system is doing exactly what it was told, I know who told it. Doctoral researcher at Aalto University. Lean Six Sigma Black Belt.
See it running before you decide anything.
One hour on one process. No integrations, no commitment.
Then a pilot on your real data, typically within two weeks.