Data platform

Data & management

One source of truth for your business — every system in one clean store, with an AI layer you can ask in plain words.

What changes for the person who owns the company: you stop asking people for numbers. The question goes to the system in plain words — this month against last, which warehouse is short, why the margin moved — and the answer comes back with the figure traceable to the row it came from.

Problem

A real number today means going through analysts, and you still get four different "sales" figures — nobody can just ask the business a question.

What it does

• Every system feeds one clean store — raw kept, facts matched, ready numbers on top.
• One true record per product ties the same item across every system.
• An AI layer sits on top: ask in plain words, or let your agents read it directly.

Inside CLAD

Built from open tools, one isolated copy per client — your data never mixes with anyone else's. CLAD is the foundation the other projects plug into.

Why the numbers can be trusted

This is the part that decides whether any of it survives contact with a real company, and it is the part almost nobody builds. Ask two people for last month's revenue and you get two numbers. Not because either is careless — because the meaning of "revenue" lives inside the query each of them wrote, and the two queries disagree.

So the definitions live in the store instead of in people's heads. Every figure has one written contract: what it means, which system owns it, what shape the data has to arrive in, and what happens when it does not.

• Checked on the way in, by machine, against that contract. Data that does not match is refused at the door and reported, instead of being quietly averaged into a report three weeks later.
• Stored under one regime, with the raw record kept. Nothing is overwritten, so a number can be re-derived rather than argued about.
• Any figure traceable to its source: a total on a dashboard follows back through the calculation to the rows it came from, in a few clicks, by whoever is asking.
• Which is also what makes AI agents safe on top of it. An agent reading a store with written contracts has far less room to invent, because it is not guessing what a word means — it is reading a definition.

The module on top: your own people build the agents

The platform is the foundation. The module that sits on it is where a company stops depending on whoever happens to know how to write a query: somebody who needs a process automated builds it themselves, that week, instead of joining a queue behind a developer. How far that goes depends entirely on how much of your work has a right answer, and that is a thing to measure in your company rather than a number to promise in advance.

• An interface an ordinary employee can use to assemble an agent for their own process or their own task. No code, no ticket to IT, no waiting a week for a developer.
• An AI assistant that helps them build it: it asks what the process actually is, writes it down, helps set a hypothesis about what should improve, and then keeps watching whether it did.
• An auditor that keeps the agents under control — what each one may touch, what it did, and where it was wrong — in one place a person can actually review.
• And underneath all of it, the contracts above. That is the whole difference between your people building agents and your people building a mess.

The AI answers from your own numbers — it doesn't invent them, and when it isn't sure it says so.

What decides the size of it

Three things, and none of them is how big your company is on paper: how many systems have to feed the store, how badly they disagree today, and how much history has to be brought in with them. Those are also the three things worth establishing before anybody quotes anything, which is what the free hour and the free diagnostic after it are for.

• A small company, a handful of systems, one warehouse: the smallest version of this, with something working in weeks rather than months.
• A mid-sized operation with several systems that disagree, real transaction volume and more than one warehouse: bigger, and staged — the store and the one true record land first, and you can ask questions of those while the rest is still being connected.
• A group with many systems, history worth keeping and a team who will go on to build their own agents: months rather than weeks, and staged for the same reason — nobody should have to wait for the whole thing to see the first honest number.
• The first conversation is an hour and costs nothing. So is the diagnostic that follows it — what you actually have, where the numbers disagree, and what a first version would need to cover. You keep whatever comes out of it whether or not anything gets built.

The people who use it every day

Said plainly, because it matters: these are employees of the company that runs the system, and that company has the same owner as this agency. It is a top-50 seller of women's clothing on the two largest marketplaces in its region, a family business. Nobody below is an arm's-length customer, and none of this is a quote written for a website — it is what they reported, in summary. The percentages are their own estimates of their own weeks, not a measurement anybody took.

• First, the routine that already existed. Igor, who runs the commercial side, uses every tool in the set and puts the saving at about a fifth of the week he used to spend on routine work.
• Sofia leans hardest on the content production and puts her saving at about a third of the week. She rates the reject rate at no worse than one in four, and the thing she named first was not a number: the feeling of up to a thousand product photos coming out overnight from one button.
• Polina mostly uses the analytical agents, and puts her saving at up to a third of the week.
• Second, and it is a different kind of thing entirely, the work that never happened at all. All three said the same about the funnel: it could not have been assembled by hand, so nobody ever had, and none of them had seen the whole picture before. They see it now every day with the analysis already on it. That is not a saving and it should not be added to the numbers above — nothing was being spent on it, because it was not being done.
• And the most honest part of their feedback is that not one of them can put a number on the platform itself. It went in long ago as version one — internally it is still called Mission Control — and the analytics, the stock ordering and the distribution across warehouses are all either done in it or done with it. They cannot compare it against not having it, because none of them has ever worked that way.

Is this for you?

• You pull numbers from several systems and they never agree.
• You have lots of small transactions that need to be analysed, not just stored.
• Two people in your company would answer "what was last month's revenue" differently, and both would be able to defend their answer.
• You want your team — and your AI assistants — to ask the business a question and trust the answer.