Skip to main content
FromNine
Menu

How we work

Understand. Design. Build. Run.

Every engagement passes through four stages, and you come away from each with something you can check and hold on to: a clear statement of the problem, an architecture with measures of success, a tested build, and a handover backed by an operating plan.

Government departments, State agencies and Irish companies all get the same four stages, delivered by teams one hour ahead of Irish time.

Talk to us about your projectHow we test AI

Four stages

From the first conversation to a system in daily use

Every stage has a named result and a defined part for your own people.

  1. Understand

    We get to grips with the business problem, who is involved and which systems already exist, and we decide with you whether AI is the right tool at all.

    What we do

    • Interviews with the staff who do the work now
    • A look at systems, data access and constraints
    • An early view of risk, data protection and AI Act category
    • A shortlist of opportunities and how each could be measured

    Result

    • A clear problem statement

    Your part

    Give us the background, introduce us to the right people and tell us what matters most.

  2. Design

    We settle on an architecture, define what success means and plan a first release, including how we will demonstrate quality, security and human oversight.

    What we do

    • Design of the architecture and the integrations
    • A test plan for AI: test sets and quality criteria
    • Design for security, privacy and human oversight
    • Release planning and team set-up

    Result

    • Architecture and measures of success
    • Plan for the first release

    Your part

    Agree the measures and priorities, and confirm the choices on data, hosting and oversight.

  3. Build

    We deliver in short steps that you can see and use, and each change is reviewed by a colleague, tested and documented as it goes in.

    What we do

    • Demonstrations on software that works
    • Automated testing, code review and security scans
    • AI output measured against the criteria we agreed
    • Documentation written next to the code

    Result

    • A tested build
    • Results of the AI tests

    Your part

    Look at each increment, try it with real users and accept what is ready.

  4. Run and improve

    We hand the system over or operate it ourselves, watch it in production and improve it with you, guided by how people actually use it.

    What we do

    • Tracking of quality, cost and response time
    • Review of cases that were unclear or escalated
    • Runbooks and knowledge transfer to your staff
    • Improvement releases at a regular pace

    Result

    • Handover and a plan for operations

    Your part

    Take ownership of the system or agree who runs it, and choose the next improvements.

After that the cycle begins again: what we see in production decides what comes next.

Collaboration

Working alongside your people

A single team with common goals, progress you can see and decisions on paper.

  • One lead who answers to you

    A single delivery lead is responsible to you for scope, quality and progress, with architects and specialists brought in as the work needs them.

  • One team, one backlog

    Your specialists and ours share a single backlog. We set priorities together, and everyone can see what has changed.

  • A regular rhythm

    Short planning cycles, demonstrations of software that works and frequent reviews, so problems show up early instead of at the end.

  • Decisions written down

    We record architecture and design decisions with the reasons behind them, so they can be reopened, explained to others and handed over.

Governance

Clear roles, risks in view, change under control

  1. A steering group that decides

    A small steering group meets on fixed dates and takes the calls on scope, priorities and risk. We agree how to escalate at the outset, before anyone needs to.

  2. Risks out in the open

    Risks, dependencies and assumptions sit in a single log with an owner each, and the steering group goes through it at every meeting.

  3. Governance for AI

    For each AI system we record its intended purpose, risk category, data sources and the points where a person decides, following what the EU AI Act and Ireland’s Regulation of Artificial Intelligence Act 2026 expect of providers and deployers.

  4. Change under control

    Any change to scope, data or how a model behaves goes through an agreed change procedure that spells out the effect on cost, quality and compliance.

Quality

Tested as a production system, and that includes the AI

Software quality and AI quality are built in together, starting with the first increment.

  • Automated tests

    Unit, integration and end-to-end tests run on their own in the release pipeline. If one fails, the release stops.

  • Code review and security scans

    A colleague reviews every change before it goes out. Scans of dependencies, code and infrastructure are automatic, and we plan security testing for each release.

  • Accessibility tests

    We test interfaces against WCAG 2.2 level AA, with automated checks plus manual testing using assistive technology. That covers the WCAG 2.1 AA level that public bodies meet through EN 301 549 under S.I. No. 358 of 2020.

  • Testing the AI output

    Together we agree quality criteria and test sets, measure the AI against them before release and continue to measure once it is live. When the system is not sure, a person makes the call.

Testing AI before release and after it

The test cycle we agree with you for each AI feature. Criteria, thresholds and review steps are set for each use case.

Read the diagram as text

The diagram is a cycle of five steps.

Step 1, criteria and test set: your experts agree quality criteria with us, along with a test set built from real examples. Step 2, test before release: the AI output is scored against those criteria and that test set. Step 3, release decision by a person: we go through the results with you, and you decide if the feature goes live. Step 4, watch in production: quality, cost and response time are followed on real use. Step 5, review unclear cases: staff deal with the cases the AI cannot handle reliably.

What staff decide in step 5 is added to the test set in step 1, so each new round of testing builds on real use.

Documentation and handover

You keep the know-how as well as the code

We plan the handover from week one, so your own staff can run and change what we deliver.

  • Architecture documents

    Up-to-date diagrams and descriptions of components, integrations and data flows, revised whenever the system changes.

  • Decision records

    The reason for each important choice, the alternatives we looked at and what would make us look again.

  • Runbooks and an operating plan

    How to release, monitor, restore and support the system, and who is responsible for what once it is live.

  • Passing on the knowledge

    Pair work, walkthroughs and handover sessions with the team that will own the system, spread over the whole engagement instead of crammed into the last weeks.

Delivery teams

Teams chosen from 2,400+ colleagues

Our teams combine AI engineering, software, cloud, data and integration skills with our SAP, Salesforce and Odoo practices, drawn from 2,400+ colleagues. A team starts small and grows with the work, from a handful of people in discovery to a full programme.

We try to keep the same people with you from design to handover, because continuity is what makes a project last as a system. The make-up of the team, and how it works with your own staff, is agreed for each engagement. Our teams work from Antwerp, Ghent and Amsterdam, one hour ahead of Irish time.

Public service

Working with the public service

Our work on 600+ government projects across the EU shapes how we approach Government departments, State agencies and local authorities. Tender competitions, the procurement rules and the paperwork they call for are part of the plan from the outset.

Accessibility, security, data protection and services in Irish and English are requirements in discovery and design, not checks for the last test cycle. We write decisions down so they can be explained to auditors, regulators such as the Data Protection Commission, and the public.

  • Public serviceWhat we build for public bodies in Ireland and across Europe.

Ireland

Planning for Irish and EU rules

Projects in Ireland face three rule sets that are moving at once. The AI Act now has an Irish implementing law, the Regulation of Artificial Intelligence Act 2026, with the AI Office of Ireland coordinating supervision. NIS2 has not yet been transposed: as of autumn 2026 the National Cyber Security Bill is still being drafted, and S.I. No. 360 of 2018 continues to apply. And GDPR, supervised here by the Data Protection Commission, governs every system that handles personal data.

We work through these questions in the Understand and Design stages, so they shape the architecture rather than hold up a release. Where Irish rules differ from those in other Member States, the plan names the difference. Your own legal advisers make the legal call.

Common questions about our approach

How long does each stage last?

That depends on the scope, the systems involved and how fast decisions get made. We agree a plan for each stage at its start and report progress all the way through. Until we understand the problem, we won’t commit to a duration.

Can we begin with something small?

Yes. Plenty of engagements begin with the Understand stage alone, or with one use case taken through all four stages before anything is scaled. Each stage leaves you with a usable result, even if you then decide to stop working with us.

What is ours when the work ends?

You get the documentation, decision records and runbooks you need to operate and change the system, and we plan the handover from day one. Who owns the code and the intellectual property is set out in the contract for each engagement.

How do you know the AI output is good enough?

Your experts agree quality criteria and test sets with us; we test before release and keep measuring in production. Unclear cases go to a person. We share the test results, and we don’t promise accuracy percentages.

Can you work with our own staff and with other suppliers?

Yes, and we often do, in joint teams and next to other suppliers. Clear interfaces, one shared backlog and written decisions make it obvious who is responsible for what.

Can you work within Irish public procurement rules?

Yes. Our 600+ government projects across the EU include tender competitions and the documents they require, and in Ireland we answer each requirement of the request for tender published on eTenders. Get in touch with the tender reference and we will pass it to the right team.

Start from the problem

Tell us what you are working on and which systems you have. We will suggest the stage to begin with, and what you would hold at the end of it.

Talk to us about your projectSee what we do