Questions operators ask first.
Straight answers on how we build, run and secure what we ship.
The engagement
Who will actually do the work?
A named engineer, from discovery to production. You meet them before you sign.
Where does the engineer actually work?
Discovery is at your operation, for a day or two, sitting with the people who handle the documents rather than only with the owner. That is where the real exceptions show up. After that the engineer works remotely inside your systems, through a shared channel, and comes back each quarter with the numbers. Travel is never billed separately.
What counts as a first system?
One piece of manual work taken over end to end, running in your environment, with a before-and-after measurement. Assembling a demurrage claim pack before the time bar, turning proofs of delivery into invoices, preparing a customs entry for your broker to sign. Not a chatbot and not a slide deck.
How long does a first system take?
It depends on the workflow and your data. We set the date in writing in the Decide phase, once the work is measured. If scope or system access changes the build, we say so then.
What happens if the system misses the target?
If the first system misses the target we agreed in writing, measured in your own systems, we refund the deployment fee. The target is set before we build: the metric, today's baseline, the level we will reach and the window we measure over. The terms fit on one page: what you provide, what is not covered, and how to claim.
What if we want to stop?
You can. The engagement is one workflow at a time, and each one has an exit: after the look back, if nothing is worth building, and after the parallel run, if the numbers do not hold. What was built stays in your accounts either way.
The work
What does the parallel run look like?
The system works on the same files as your team, for a few weeks, without writing anything your team has not approved. We compare its output with theirs, field by field, and fix what differs. It goes live only when the numbers hold, and the person who did the work before becomes the person who approves.
What about AI making things up?
Every field is matched to your records and shown next to its source. Anything uncertain goes to a person. Nothing enters your system on the model's word alone.
Is this another ERP project?
No. Nothing in your ERP changes. The engineer reads from it through the interface it already has and writes to a review queue or a sheet your team controls. No migration, no vendor rollout, no retraining of the whole company.
What if our data is messy or our system is old?
That is the normal case. Discovery measures what your data actually looks like on real documents, and we connect through whatever interface your system already has.
Does this work with our legacy software?
That is the point. We connect to trading systems, ERPs, customs portals, shared inboxes and file-based archives through their existing interfaces, and build the missing pipeline where none exists.
What if AI is not the answer for a workflow?
Then we say so in the Decide phase. Sometimes the answer is a step to delete or a connection to build, not a model.
Why not buy an AI product?
Products stop at the interface. Someone still has to connect them to your ERP, your inbox and your documents, measure them against how the work is done today, and keep them running. That last mile is where most AI projects die, and it is the whole job of the engineer we put on your team.
Your data
Is our data safe from public AI models?
Yes. Nothing we build sends your data to a public model for training. The system runs against private model endpoints or models deployed in your environment, with retention limits written into the contract and your own audit log.
Can it run on-premise or air-gapped?
Yes. Open-weight models can be deployed entirely inside your network. Where a hosted model is acceptable, we use private endpoints with contractual retention limits.
Who owns what we build?
You do. Code, prompts, evaluations and fine-tuned weights are delivered into your repositories and accounts at handover.
Put an engineer inside your operation.
Tell us about the work your team still does by hand.
Book a call