Four kinds of work, usually in this order.
Digitise the process, build the data under it, add AI where it is safe, and connect it all to the systems you already run.
- 01
Process digitisation
Map the operation, remove re-keying, and move hand-offs out of email and spreadsheets into tracked workflow.
You get- A mapped current-state process
- Intake, conditions and routing as workflow
- A short list of what to fix first
- 02
Data foundations
One governed layer of loan, policy, member and external data, with lineage and access control.
You get- A data model that matches how you operate
- Pipelines with lineage
- Permissions you can audit
- 03
AI enablement
Narrow models on clean data, designed so a person reviews the output.
You get- Extraction, triage or decision support for one process
- Evaluation against your own files
- A review step and an audit trail
- 04
Integration and modernisation
Connect the core, the loan or policy system and the vendors around them, and plan careful cut-overs.
You get- APIs and event streams
- Data migration support
- A cut-over plan
Specifics by industry: Financial services · Geospatial and weather
Start with one process. Leave with a plan.
- 01
Pick one process
Choose the process that hurts most. One is enough, and it keeps the first step small.
- 02
Map it with your team
We walk the work end to end and mark where files wait, get re-keyed or get lost.
- 03
Leave with a written plan
What to fix, in what order, what each step needs, and what we would leave alone.
-
Discovery
A fixed-scope first step: map one process and write the plan. You can take the plan to anyone.
-
Build
We design and build the first piece with your team, then hand it over with the documentation to run it.
-
Ongoing support
A small senior team on call as the work grows, for changes, new processes and new data.
-
Senior engineers only
The people you meet are the people who do the work.
-
Your operation first
We start from how work moves, not from a platform we want to sell.
-
Open standards
Our data engine is open source, built on Zarr, Arrow and DataFusion, so you are never locked in.
-
A person reviews the AI
Narrow, auditable models. Decisions stay with people.
Do we have to replace our core system?
No. The system of record is usually fine. We work around and alongside it: intake, workflow, data and integration.
What does the AI decide?
It reads, classifies, scores and flags. People make the credit, underwriting and claims decisions, and every override records the reason.
Is our data clean enough to start?
Usually not, and that is the point. We start with the process and make the data usable as we go, so the first AI step has something to stand on.
Who actually does the work?
A small team of senior engineers, with no pyramid of juniors behind them.
Do you work with Earth and weather data too?
Yes. It is the other half of what we do, and it matters for insurers. See the Geospatial page.
