Approach

What rollout looks like, day by day

No leap of faith and no multi-month projects without results. A 14-day pilot shows value on your real documents — and everything after it builds only what was proved.

Workshop desk with a thin blue line leading scattered sheets into one neat stack

Inside the pilot

14 days, three steps, one team

What actually happens from day one to the report — and how much of your time it really takes.

  1. Knowledge and access map

    days 1–3

    We capture where knowledge lives today, where it leaks and where hours are lost — before we build anything.

    • Inventory of tools and places where knowledge lives
    • In black and white where hours are lost
    • Pick 1–2 automations with the biggest gain
  2. Automations on real documents

    days 4–10

    We build and run chosen automations on your real documents — locally, with approval for every write.

    • Working automations, not a demo on someone else’s data
    • Every write goes to quarantine, for human approval
    • Your team does their job — we work around you
  3. Measurement and plan

    days 11–14

    We measure what changed and hand over a rollout plan, step by step.

    • Before/after comparison on agreed metrics
    • Rollout plan with deadlines
    • Everything stays yours even if we stop

The pilot engages one team, and only in a limited way. If it does not show measurable value on your documents — we do not charge for it.

Row of tiles arranging from left to right out of scattered grey clutter into a connected layout linked by blue lines
  • Day 1
  • Day 14

Under the hood

From scattered document to answer

Scattered information does not become useful on its own. The system guides it through five steps — from collection to an answer with a source — with human approval before every write.

01 — Collection

Collection

Conversations, documents and decisions flow into a connected knowledge base — on your machines, not in someone else’s cloud.

02 — Cleaning

Cleaning

Duplicates are resolved, personal data sidelined, and everything gets consistent labels.

03 — Linking

Linking

Related notes, projects and people are connected, so knowledge becomes searchable as a whole — not folder by folder.

04 — Answer

Answer

A question gets an answer with a source, in seconds. Drafts and reports appear overnight — nothing is written without review.

05 — Maintenance

Maintenance

Quarterly cleanup removes what is stale, so the graph gets denser, not messier.

Where data live

Two layers: private stays private

Company knowledge is split into two layers — private, which never leaves your machines, and shared, which the team and system draw from. AI comes to the data; data do not travel to AI.

Shared layerDocumentsResearchDecisionsProceduresApproval gatePrivate layerContractsFinanceClientsPersonal data
The biggest gain is where contracts, finances and client data live — that is why the system is built to be allowed to work there.

The system writes only to quarantine. No automatic writes. No source — no note.

Agenti predlažu, ljudi odlučuju.

Independence

We are tied to no vendor

AI models change month to month. The system is built so that costs you nothing.

  • When a better or cheaper model appears, we swap it — and your system does not notice.
  • For the strictest cases the model runs on your hardware, so cost per task falls too.
  • Everything the system can do we write and verify ourselves — nothing installs unseen.
  • Exit is built in: everything is ordinary files in an open format, readable without us.

Process questions

What it looks like in practice

How do you measure whether the pilot succeeded?

Before we start we agree what is measured — usually time to answer and hours of manual work on chosen tasks. At the end of the pilot you get a before/after comparison on those metrics, on your documents, not someone else’s examples.

What if our procedures change?

The system is maintained from real practice: when a procedure changes, the record updates, and quarterly cleanup removes what is stale. That is why maintenance exists as a phase — a system without a gardener decays.

Who on our side must be involved?

For the pilot: one person who knows where knowledge lives and the team whose documents we use. For long-term work we train 1–2 internal champions who keep the system alive — hours per week, not days.

Does rollout disrupt day-to-day work?

No. Setup fits around your schedule and the team keeps doing their job. The first change people feel is a finished draft waiting in the morning — not a new programme they must learn.

Ready to see the method on your example?

30 minutes of conversation, no sales pressure. We walk through your case: where knowledge leaks, what the pilot would fix first and what it would look like for you.

Book 30 minutes

We reply within 24 hours