HOW WE WORK
A staged engagement, with a decision at every gate.
Every engagement moves through four gates. Each one ends with a decision that is yours: go, no-go or later. A successful proof earns the right to production. A failed one saves you from scaling the wrong thing.
Start a DiagnosticThe path
Four gates, one accountable team from the first call to the final measurement.
- 01
Diagnostic
2–4 weeks · fixed fee. We map your priorities, processes, economics, data and risk to the handful of workflows with a credible case. Output: ranked portfolio, sized levers with visible assumptions, a frozen measurement ruler. Gate: go, no-go or later.
- 02
Prove
15–30 days · fixed fee. One workflow on live port calls, voyages, crew changes or dockings, your people approving every consequential step. Output: a measured result on the ruler frozen before day one. Gate: did it move the line we named?
- 03
Deploy
6–12 weeks; 10–20 for multi-system or regulated · milestones. Production in your estate: integrations, controls, runbooks, training, acceptance tests. Output: software your team runs. Gate: acceptance passed, users trained.
- 04
Run
annual licence. Thirty-, sixty- and ninety-day measurement; attribution split between what we did, what the market did and what remains unattributed; you sign it. Output: the next workflow, or not. Gate: repeatable?
Engagement terms
What production includes
What production includes.
We deliberately test the conditions a polished prototype hides: missing data, conflicting documents, an agent's portal that is down, a tariff page that no longer exists, a user who was not in the demo. The aim is not to make AI look impressive. It is to make the workflow dependable on an ordinary day.
| Workstream | Detail |
|---|---|
| Workflow | The process that changes. |
| Data | What is observed, and where it comes from. |
| Integration | APIs, events, governed views, files, approved email. |
| Intelligence | Connect, contextualise, retrieve, calculate, reason, act. |
| Evaluation | Acceptance tests on every production workflow. |
| Security | Identity first. Least privilege. Engines hold no credentials. |
| Governance | Staged, not committed. Named person for every commit. |
| Experience | The people who run the work. |
| Change | What has to change around the software. |
| Measurement | The same ruler used to approve the work. |
What we need from you
Three things, before the first week.
An owner.
One person who can change the workflow and say yes or no.
Access.
Read access to the systems and documents the workflow uses, through your IT, on your terms.
A baseline.
Whatever you have: the last quarter's disbursement accounts, the SOFs, the crew matrix. We build the ruler from it.
How we price
Fixed fee to diagnose. Fixed fee to prove. Milestones to deploy. A licence to run it yourself.
We agree scope and fee before work starts. Licence pricing is by vessel for fleets and by site for ports and yards, so a five-vessel manager and a national port operator both pay for what they run. No open-ended discovery. No licence before we know what it is supposed to achieve.
| Stage | Model |
|---|---|
| Diagnostic | Fixed fee. |
| Prove | Fixed fee. |
| Deploy | Milestones. |
| Run | Annual licence. Vessel for fleets, site for ports and yards. |
How we measure
We measure business outcomes, not activity.
Capacity released is reported as capacity, not as a payroll saving. Avoided cost is reported on an expected-loss basis. Revenue is incremental contribution, never headline revenue. The ruler is the one we agreed at the Diagnostic; success is not redefined at the end.
| Category | How we report it |
|---|---|
| Banked cash | Banked cash |
| Avoided cost | Reported on an expected-loss basis. |
| Released capacity | Reported as capacity, not as a payroll saving. |
| Working capital | Working capital |
| Revenue | Incremental contribution, never headline revenue. |
| Risk | Risk |
| Compliance | Compliance |
FAQ
Questions
How long does this take?
Diagnostic two to four weeks. Proof of Value fifteen to thirty days. Production six to twelve weeks, or ten to twenty for multi-system or regulated work.
What if the best answer is not AI?
A failed proof saves you from scaling the wrong thing. The Diagnostic may recommend not using AI.
Can we enter at proof of value?
A successful proof earns the right to production. The measurement ruler is frozen before day one of the proof.