Skip to content
NUL-G

AI in your process.

Work the system does for you. Reading, checking and proposing. On its own, on top of what you already have, such as a mailbox, folder or ERP, or inside a custom system. The system does the work; a person decides.

Get a working prototypeGet in touch

No obligation · six questions · reply within one to two working days

Our position.

What we believe in, and how you will notice it.

  1. 01

    No AI where a rule will do.

    If a fixed rule can do it, we build a rule. Cheaper and predictable.

  2. 02

    Measured on your work, before you sign.

    The prototype runs on your own documents. You see how often it gets it right and where it is unsure.

  3. 03

    Every result with its source.

    The system shows where it read something. If it is unsure, the case goes to a person.

  4. 04

    It stays yours.

    Instructions, test set and log belong to you. You can switch to another model.

What it is.

One type of work, with a person who checks.

What we build most often

  • Reading: delivery notes, invoices, SOFs, certificates, requests
  • Checking: invoice against order, certificate against requirement, permit against other permits
  • Proposing: a shift swap, task risk assessment, cost estimate or HS code
  • Searching and summarising: contracts, specifications, fault history
  • Reporting: a draft that a person finishes

When AI fits

  • The input differs every time: PDF, email, photo, free text
  • There is a lot to read or check
  • Someone checks it now and can keep doing so

When it does not

  • A fixed rule is enough
  • A mistake costs more than having a person do it
  • There are no examples to measure against

What you get

  • An application that does one type of work
  • A review screen with the source of every result
  • A log of every proposal and every decision
  • The measurement: how often it is right, per document type

How it works.

First we measure on your own examples, then we build.

How it works

  • Choose: one type of work, the one with the most manual effort
  • Test set: a stack of real examples, including the difficult ones, with the right answer from your people
  • Prototype: measured on that test set, field by field
  • Build: with a screen where a person checks and approves
  • Live: what is certain goes through; what is uncertain goes to a person
  • Keep measuring: in the quarterly report, with the test set

Data and rules

  • Only models covered by a data processing agreement; your data does not train models
  • Data stays in the EU
  • Access by role, as in the rest of the system
  • For each application, we record which AI Act risk class it falls under, and how human oversight works

What we do not do

  • Systems that decide on their own, without a person
  • A general-purpose chatbot as a product
  • Promising percentages before anything is measured
  • A model that quietly learns from your data. Improvements go through the test set and instructions, visibly and with your approval

What it costs.

Fixed price per phase, as with custom software. The model’s usage costs per month, shown in the quarterly report.

See the price indication

Questions.

Questions that come up often, with a short answer.

Will AI take over our work?

The system proposes, you approve. It takes over the retyping and checking, not the decision.

What happens to our data with AI?

We only use models covered by a data processing agreement. Your data does not train models, and it stays in the EU.

What if a better AI model comes along?

Then we test it on your test set. If it is better, or cheaper for the same result, we propose the switch. As part of support.

What will it deliver?

We tell you what we expect, and after go-live we measure what it actually does. No percentages up front.

Other services.

Often together, sometimes on their own.

Show us one process that could be lighter.

Get a working prototype

No obligation · six questions · reply within one to two working days. Or get in touch