AI & Data
The technology domain this engagement operates — capabilities, architectures and the full picture.
AI Consulting
From 'we should do something with AI' to a decision you can defend.
The operating model
AI consulting is a structured engagement for organisations that suspect AI should matter to them and want that suspicion tested honestly: which processes are actually candidates, what the data can support today, what a pilot would prove, and what it costs to run the result — not just to build it. The AI capability pages describe the technologies; this page describes the engagement that decides whether and where they are worth your money.
The model is deliberately gated. Each phase ends in a decision you make with evidence in hand, and stopping at a gate is a success of the engagement, not a failure of it — the most expensive AI outcome is a pilot that should never have started, kept alive because nobody was given a clean place to say no. We are a systems integrator: our recommendation is not a sales channel for a model vendor, and build-nothing is always on the table.
The rhythm of the engagement
What actually happens — by day, by week, by month, by quarter. If a provider cannot describe this, they are selling a tool, not a service.
Daily
During active phases: working sessions with your process owners and data people — the engagement runs on your ground truth, not on our slideware.
Weekly
A written status you can forward: what was examined, what was found, what changed in the assessment, and the open questions with owners.
Monthly
A steering session at each phase boundary: the evidence, the recommendation with reasoning shown, and the decision recorded — proceed, adjust, or stop.
Quarterly
For engagements that continue past a pilot: a review of what the pilot's numbers did after handover, because a result that only held during the pilot is a result that needs explaining.
How escalation works
Severity definitions and response objectives are agreed at onboarding and written into the runbook — they are commitments made to you, not marketing figures published here.
Gate 1 — after discovery
The candidate list, ranked by value and feasibility, with the data reality stated bluntly. Decision: which one or two candidates justify a feasibility phase — or none, which is a legitimate and cheap answer.
Gate 2 — after feasibility
For the chosen candidates: what the data supports today, the measurable target a pilot must beat, the baseline it is compared against, and the run-cost estimate. Decision: pilot, fix the data first, or stop.
Gate 3 — after the pilot
The pilot measured against the target agreed at gate 2 — on your data, in your environment, by the metric written down beforehand. Decision: production path with owners and costs, iterate once with a stated change, or stop with the findings documented.
Onboarding, phase by phase
Sequenced so that value starts before the last phase finishes. Dates go into the plan we agree together; phases are what the plan is made of.
Framing
The business problems named in business terms, the stakeholders identified, and the success criteria drafted before anyone mentions a model.
Discovery
Process walkthroughs and data reality checks — where it lives, its quality, who owns it, and what governance permits — producing the ranked candidate list for gate 1.
Feasibility
The short technical proof behind the chosen candidates: retrieval quality on your documents, baseline comparisons, the boring-but-decisive questions answered before a pilot spends real money.
Pilot and handover
A scoped pilot with the metric agreed in advance, then handover of everything — code, evaluation harness, findings — whether the decision is proceed or stop.
What we run, and what stays yours
A managed engagement fails quietly when this table was never written. Ours is agreed before the service starts.
| Area | NexMena | You |
|---|---|---|
| The engagement and its evidence | Run the phases, produce the analysis, show the reasoning | Supply access to the people and data the analysis depends on |
| Gate decisions | Recommend, with the evidence attached | Decide — proceed, adjust or stop is a business call made by you at every gate |
| Data and its governance | Assess quality and readiness; state what governance permits and blocks | Own the data, its permissions, and any approvals your regulator or policies require |
| What happens after | Hand over working artefacts and an honest findings document | Own the production decision and its budget — we bid for the build like anyone else, on the same document everyone can read |
What you receive, and when
Reporting exists so you can judge the service without asking for a meeting.
- The weekly written status, short enough to read and specific enough to forward.
- The gate documents: evidence, options, recommendation and recorded decision — the paper trail that makes the final choice defensible to a board or a regulator.
- The handover pack at the end of every engagement, including the stopped ones: what was tested, what the numbers were, and what would need to change for the answer to change.
Questions we are asked
What does an engagement cost and how long does it run?
Each phase is scoped and priced separately, and the gates exist so you never commit to the whole arc up front. Discovery is measured in weeks; what follows depends on what it finds. We would rather give you a dated plan at gate 1 than a number before framing.
Do we need our data sorted out first?
No — finding that out is what discovery is for. 'Fix the data first' is a common and useful gate-2 outcome, and it arrives with a specific list rather than a vague accusation. Postponing the engagement until the data is perfect usually means postponing it forever.
Are you selling a particular model or platform?
No. As an integrator we work across the platforms, and the feasibility phase tests options against your data rather than defaulting to a favourite. Where a managed platform is the honest recommendation, the run-cost estimate says so with numbers, and the reasoning is in the gate document for anyone to challenge.
What if the pilot fails?
Then the engagement worked: it bought certainty at pilot price instead of production price. The handover pack documents what was tested and why it missed, which is exactly the document that stops the same idea being re-pitched to you in a year with the same flaws.
How is this different from the AI capability pages?
Those pages describe what we build — RAG systems, agents, vision, analytics platforms. This page describes the engagement that decides whether building any of them is justified. Arriving here first is the cheaper order.
Who from our side needs to be involved?
A sponsor with authority to make the gate decisions, the owners of the processes under examination, and whoever actually knows the data — often not the person the org chart suggests. The weekly rhythm is built to use their hours sparingly and visibly.
Continue reading
Consulting
The delivery model in general: how managed engagements work across every domain.
Enterprise AI & RAG
Retrieval-augmented generation over your own documents, with permission filtering at retrieval time, citations and an evaluation set your experts agree.
Start with an assessment
The fastest way to a useful answer is a short, scoped look at what you already have.