Our approach
Understand. Design. Build. Operate.
Our engagements move through four stages, each ending in something you can review and keep: a problem definition, an architecture with agreed success measures, a tested implementation and a handover with an operating plan.
The same four stages for public institutions and companies, in every European market we work in.
Four stages
From first conversation to running system
Each stage has a named output and a clear role for your team.
Understand
We map the business problem, the people involved and the systems already in place, and decide together whether AI is the right answer.
What happens
- Conversations with the people who do the work today
- Review of systems, data access and constraints
- First pass on risk, privacy and AI Act classification
- A shortlist of opportunities and what we would measure
Output
- Problem definition
Your role
Share context, open doors to the right people and decide what matters most.
Design
We choose an architecture, agree what success looks like and plan the first release, including how quality, security and oversight will be shown.
What happens
- Architecture and integration design
- Evaluation plan: test sets and quality criteria for AI
- Security, privacy and human-oversight design
- Release plan and team set-up
Output
- Architecture and success measures
- Release plan
Your role
Agree measures and priorities, and confirm decisions on data, hosting and oversight.
Build
We deliver in short, tested increments that you can see and use, with changes peer-reviewed, tested and documented as they land.
What happens
- Increments demonstrated on working software
- Automated tests, code review and security checks
- AI output evaluated against the agreed criteria
- Documentation written alongside the code
Output
- Tested implementation
- Evaluation results
Your role
Review increments, test with real users and accept what is ready.
Operate and improve
We hand over or run the system, monitor it in production and improve it with you, based on what real use shows.
What happens
- Monitoring of quality, cost and latency
- Review of uncertain and escalated cases
- Runbooks and knowledge transfer to your teams
- Regular improvement releases
Output
- Handover and operating plan
Your role
Own the system or agree who runs it, and decide which improvements come next.
Then we start again from what we learned: what happens in operation sets the agenda for the next round.
Collaboration model
How we work with your teams
One team with shared goals, visible progress and decisions on record.
One accountable lead
A single delivery lead answers to you for scope, quality and progress, backed by architects and specialists as the work requires.
Joint team, one backlog
Your experts and ours work from one backlog. Priorities are set together and changes are visible to everyone.
A steady rhythm
Short planning cycles, demonstrations on working software and regular reviews, so surprises surface early rather than late.
Decisions on record
Architecture and design decisions are written down with their reasons, so they can be revisited, explained and handed over.
Governance
Clear roles, visible risks, controlled change
Steering that decides
A small steering group meets on a fixed cadence to decide on scope, priorities and risks. Escalation routes are agreed at the start, not when they are needed.
Risks in the open
Risks, dependencies and assumptions are tracked in one log and reviewed at every steering meeting, with an owner for each.
AI-specific governance
For AI systems we document the intended purpose, risk classification, data sources and where people stay in control, in line with what the EU AI Act asks of providers and deployers.
Controlled change
Changes to scope, data or model behaviour go through an agreed change process, with the impact on cost, quality and compliance made explicit.
Quality assurance
Tested like any production system, including the AI
Software quality and AI quality are engineered together, from the first increment.
Automated testing
Unit, integration and end-to-end tests run automatically in the delivery pipeline, and a failing test stops the release.
Review and security checks
Changes are peer-reviewed before they ship. Dependency, code and infrastructure scans run automatically, and security testing is planned per release.
Accessibility testing
Interfaces are tested against WCAG 2.2 level AA with automated checks and manual review using assistive technology.
Evaluation of AI output
We agree quality criteria and test sets with you, measure AI output against them before release and keep measuring after it. When the system is unsure, a person decides.
The evaluation loop we agree with you for every AI feature. Criteria, thresholds and review steps are set per use case.
Read the diagram as text
The diagram shows a loop of five steps.
Step 1, criteria and test set: quality criteria and a test set of real examples are agreed with your experts. Step 2, evaluate before release: the AI output is measured against that test set and those criteria. Step 3, human release decision: the results are reviewed with you, and you decide whether the feature goes live. Step 4, monitor in production: quality, cost and latency are tracked on real use. Step 5, review uncertain cases: people handle the cases the AI cannot handle reliably.
The decisions made in step 5 feed back into the test set of step 1, so each round of evaluation starts from what real use has shown.
Documentation and handover
You keep the knowledge, not just the code
Handover is planned from the first week, so your teams can run and change what we build.
Architecture documentation
Current diagrams and descriptions of components, integrations and data flows, kept up to date as the system changes.
Decision records
Why each significant choice was made, what else was considered and what would make us revisit it.
Runbooks and operating plan
How to deploy, monitor, recover and support the system, and who does what after go-live.
Knowledge transfer
Pairing, walkthroughs and handover sessions with the team that will own the system, spread across the engagement rather than squeezed in at the end.
Delivery teams
Teams drawn from 2,400+ people
We put together multidisciplinary teams from 2,400+ colleagues across AI engineering, software, cloud, data, integrations and our Salesforce, SAP and Odoo practices. A team starts focused and grows with the work, from a small group in discovery to a full programme.
We aim to keep the same people on your programme from design through handover, because continuity is what turns a project into a system that lasts. How the team is composed and how it works alongside your own people is agreed per engagement.
- Our servicesThe services our teams are drawn from.
- Enterprise platformsSalesforce, SAP and Odoo practices.
Public sector
Working with public bodies
Experience from 600+ European government projects shapes how we work with public institutions. Tender phases, procurement rules and the documentation they require are built into the plan.
Accessibility, security, data protection and multilingual delivery are requirements in discovery and design, not items for the final test round. We document decisions so they can be explained to auditors, supervisory bodies and citizens.
- Public sectorWhat we build for public institutions across Europe.
International
Delivery shaped by European rules
Cross-border programmes meet the same questions in every Member State: which obligations of the EU AI Act apply, how personal data is protected under GDPR, and what the Cyber Resilience Act will ask of software with digital elements. We plan for them in the Understand and Design stages, so they shape the architecture instead of delaying the release. Where national rules differ, the plan says so explicitly; the legal reading remains with your counsel.
Questions about how we work
How long does each stage take?
It depends on the scope, the systems involved and how quickly decisions can be made. We agree a plan for each stage at the start and show progress at every step. We do not promise durations before we understand the problem.
Can we start small?
Yes. Many engagements start with the Understand stage, or with a single use case taken through all four stages, before scaling up. Every stage ends in an output you can use, even if you decide not to continue with us.
What do we keep at the end?
You receive the documentation, decision records and runbooks needed to run and change the system, and handover is planned from the start. Ownership of code and intellectual property is set out in the agreement for each engagement.
How do you make sure AI output is good enough?
We agree quality criteria and test sets with your experts, evaluate before release and keep measuring in production. Uncertain cases go to a person. We report evaluation results; we do not promise accuracy figures.
Can you work alongside our own teams and other suppliers?
Yes. We often work in joint teams and alongside other suppliers. Clear interfaces, a shared backlog and written decisions keep responsibilities visible.
Do you work within public procurement rules?
Yes. Our experience of 600+ European government projects includes tender procedures and the documentation they require. Contact us with the tender reference and we will route it to the right team.
Start with the problem
Tell us what you are working on and the systems you already run. We will suggest which stage to start with, and what you would have at the end of it.