Six stages, each producing something you can inspect and ending with a decision you sign off. Nothing moves forward on an assumption.
Six stages, six gates
We study the problem with the people who live it: workflow, customers, constraints. Output: Problem brief.
Problem brief agreedWe agree what success means, what is in scope and the hypothesis we are testing. Output: Scope and hypothesis.
Scope signed offFlows, interfaces and architecture, tested as a prototype before engineering starts. Output: Validated prototype.
Prototype approvedWorking software and artefacts, demonstrated every week. Output: Working software.
Features acceptedRelease to real users with documentation, monitoring and a runbook. Output: Release and documentation.
Release acceptedWe measure what happened and turn it into the next roadmap. Output: Measurement and roadmap.
Roadmap reviewedWhen the scope needs to change
Good projects learn things. When they do, the change is written down before it is built.
- Raise. Either side can propose a change at any time
- Assess. We write down the effect on scope, timeline and cost
- Decide. You approve, defer or decline it in writing
- Record. Approved changes update the plan and the acceptance criteria
Rule. Nothing outside the agreed scope is built without a written change approval.
One integrated team
Forward deployed engineers sit between the business problem and the technical system, so the build stays tied to how the work really happens.
- Product managersOwn the outcome, the scope and the acceptance criteria.
- Forward deployed engineersSit between the business problem and the technical system, discovering the real workflow and keeping the build tied to it.
- AI specialistsDesign retrieval, tools, evaluations and guardrails.
- Software engineersBuild the applications, integrations and data layers.
- DesignersMake the system understandable and usable.
- QATest function, accessibility, performance, AI behaviour and security from the first component.
What we don't trade away
- Human judgment stays in the loopWhere context, ethics, exceptions or accountability matter.
- Permissioned and loggedEvery AI action that can affect a workflow is controlled in code.
- Evidence, not adjectivesWe measure, and we publish only what we can verify.
- Handover by designCode, documentation and credentials are yours at the end.
Start with discovery
Bring the problem as it is today. Discovery tells us both what to build, and what not to.