What actually happens when you work with us
ACOS has seven phases. Each answers a different question and ends with a decision about what happens next. You do not need to arrive with an AI specification. Bring the business problem.
Before we get into the detail
- You do not have to buy all seven phases. ACOS is a way of working, not software or a subscription.
- An engagement can stop. We do not build first and ask questions later.
- You remain involved in the information, boundary and decision points.
- Every phase ends in a decision. Any of those decisions can be to stop, and stopping early because the evidence says so is a good outcome, not a failed project.
THE SEVEN PHASES
DISCOVER
We start by understanding how your organisation actually works — not how an org chart says it works.
ASSESS
Before anything is built, we establish what your organisation actually knows and how reliable that knowledge is.
ARCHITECT
The design phase covers the knowledge the system will draw on and the boundaries it will work within.
BUILD
The phase people imagine is the whole project. In practice it is the most straightforward, because the difficult decisions were made in the three phases before it.
DEPLOY
Deployment is controlled and staged. Nothing reaches the public, your customers or your clients without a separate, explicit decision.
GOVERN
Governance is a phase of work with its own outputs, not a document produced at the end because someone asked for it.
EVOLVE
Organisations change. Policies get updated, services change, people leave. A system built on last year's information quietly becomes wrong.
DISCOVER
Understand the business before discussing the technology
We start by understanding how your organisation actually works — not how an org chart says it works.
| At this phase | What it means for your business |
|---|---|
| What happens | Conversations with the people who do the work and the people accountable for it. We look at the problem you have described, the pressures behind it, the constraints you operate under, and what has already been tried. |
| What we need from you | Your time, mostly in conversation. Access to the people who understand the problem. Honesty about what has not worked before, which is usually the most useful thing anyone tells us. |
| What you receive | A written statement of the problem, the objectives, the constraints and the intended outcome — plus an explicit view on whether AI is an appropriate answer at all. |
| Decision at the end | Proceed with an agreed problem and objective, investigate further, choose a non-AI response, or stop. |
| Why it matters | Most disappointing AI work is technically fine and aimed at the wrong problem. This phase exists to prevent that, and occasionally to conclude that the problem you have is not one AI should be solving. |
ASSESS
Separate what is known from what is assumed
Before anything is built, we establish what your organisation actually knows and how reliable that knowledge is.
| At this phase | What it means for your business |
|---|---|
| What happens | We review the information relevant to the problem and classify every item as verified, assumed, or requiring your confirmation. Gaps and contradictions are recorded rather than smoothed over. |
| What we need from you | You confirm, correct or reject each item we have flagged. You see exactly what we intend to build on before we build on it. |
| What you receive | An evidence and assumption register — a plain list of what is confirmed, what is uncertain, and what is missing. Also an honest view of whether the organisation's information is in good enough shape to proceed. |
| Decision at the end | Proceed only if the relevant information is sufficiently clear and approved; otherwise resolve gaps or stop. |
| Why it matters | An AI system inherits the quality of what it is given. This phase is where most of the real risk is removed, and it frequently surfaces inconsistencies the organisation did not know it had. |
ARCHITECT
Design what will be built, and what it will refuse to do
The design phase covers the knowledge the system will draw on and the boundaries it will work within.
| At this phase | What it means for your business |
|---|---|
| What happens | We design the knowledge foundation, decide what the system will and will not answer, define when it must hand off to a person, and agree what it must disclose. Boundaries are designed here, not patched on later. |
| What we need from you | You review and approve the design, particularly the boundaries. This is where you tell us what the system must never do in your organisation. |
| What you receive | A design specification: the approved knowledge source, the rules governing behaviour, the refusal and hand-off conditions, and what success would look like. |
| Decision at the end | Approve the design and its human hand-off boundaries, revise it, or do not build. |
| Why it matters | What a system refuses to do is a design decision. Made here, it holds. Made after something has gone wrong, it is a repair. |
BUILD
Construct it
The phase people imagine is the whole project. In practice it is the most straightforward, because the difficult decisions were made in the three phases before it.
| At this phase | What it means for your business |
|---|---|
| What happens | We build and configure the solution against the approved design, working from approved information provided by your organisation. Nothing goes in that was not agreed in Assess and Architect. |
| What we need from you | Relatively little of your time. You may be asked to clarify points as they surface, and anything ambiguous comes back to you rather than being resolved by guesswork. |
| What you receive | A working system built to the approved design, ready to be tested rather than ready to be used. |
| Decision at the end | Move the bounded system into controlled testing, return it for correction, or stop. |
| Why it matters | Building is where an unclear problem or unverified knowledge becomes expensive. Done in sequence, this phase is uneventful — which is the point. |
DEPLOY
Put it in front of people, carefully
Deployment is controlled and staged. Nothing reaches the public, your customers or your clients without a separate, explicit decision.
| At this phase | What it means for your business |
|---|---|
| What happens | We test the system against a written protocol before anyone uses it, recording every result, pass or fail. Then it goes to a limited, nominated group in a controlled environment, and we watch how real people use it. |
| What we need from you | You nominate who gets access. You see the test record — including anything that failed — before your people use it. |
| What you receive | A tested system in the hands of a defined group, plus a written record of what was tested and what was not. |
| Decision at the end | Keep access restricted, expand it under an approved plan, return for correction, or withdraw it. |
| Why it matters | Real users ask questions no test protocol anticipates. Finding that out with a controlled group is very different from finding it out in public. |
GOVERN
Phase six, not an appendix
Governance is a phase of work with its own outputs, not a document produced at the end because someone asked for it.
| At this phase | What it means for your business |
|---|---|
| What happens | We establish who is accountable for the system, what it is permitted to do, how it is monitored, what happens when it gets something wrong, and how changes are approved. We also record plainly what has not been tested. |
| What we need from you | You nominate the accountable owner. You approve the rules and the escalation path. This is a business decision, not a technical one, and it stays with you. |
| What you receive | A governance record: accountability, boundaries, monitoring, review cycle, escalation, and a stated list of known limitations. |
| Decision at the end | Approve the accountability, monitoring and escalation arrangements, revise them, or do not continue use. |
| Why it matters | The decisions that make an AI system safe to rely on are business decisions about accountability and boundaries. Left to the end, they get made by default. Given a phase, they get made deliberately. |
EVOLVE
Because capability that is not maintained decays
Organisations change. Policies get updated, services change, people leave. A system built on last year's information quietly becomes wrong.
| At this phase | What it means for your business |
|---|---|
| What happens | Scheduled review of how the system is performing in real use, what it is being asked that it cannot answer, what information has gone out of date, and what should change. Corrections are made through the same discipline as the original build. |
| What we need from you | A review rhythm you agree to, and someone on your side who owns it. The accountable owner named in Govern is usually the right person. |
| What you receive | A reviewed and updated capability, with a record of what changed and why — and a live view of where it is still weak. |
| Decision at the end | Retain, change, restrict or retire the capability based on current evidence and accountable human review. |
| Why it matters | This is the phase most often skipped, and it is why AI systems that worked at launch stop being trusted a year later. Nobody decided to stop maintaining it. It simply was not anyone's job. |
Which phases each engagement covers
Managed and Integrated cover the same ACOS phases. The difference is not how much of the methodology you receive — it is the complexity of the AI environment being supported.
| Engagement | Phases it covers |
|---|---|
| Free 30-Day AI Assistant Experience | Demonstration only. The Free 30-Day AI Assistant Experience uses appropriate public or supplied business information. Any assumptions are identified, and authoritative knowledge is verified and refined with you. |
| AI Assistant Essentials | Discover, Assess, Architect and Build. |
| ACOS Managed | Discover, Assess, Architect, Build, Deploy, Govern and Evolve. |
| ACOS Integrated | Discover, Assess, Architect, Build, Deploy, Govern and Evolve, for a more technically complex AI environment. |
| Enterprise AI Architecture | The full ACOS lifecycle plus enterprise requirements. |
You do not need to commit to a full AI project
Start with a conversation or the Free 30-Day AI Assistant Experience. Either can lead to continuing, clarifying, changing direction or stopping.
Start your free 30-day experience