“Let’s use AI in the organization” is not a decision. Which unit, which task, which model — that is the decision.
Over five days we take seven units one at a time: marketing, sales, finance, HR, operations, legal and IT. For each we build the real task list, decide which model serves which task, which integration connects through which MCP server, and which decision has to stay with a human. The output is two documents: a per-unit tool stack and a 12-month rollout sequence.
Unit-Level Model and Solution Mapping — Analysis and Rollout Sequence
The problem underneath.
Enterprise AI decisions are usually made at the wrong level. Somebody says “let’s buy this model” or “let’s move to this platform” — but a model’s accuracy is not a property independent of the task. Legal contract review and marketing content generation do not need the same capability profile, do not carry the same cost, and do not have the same tolerance for error. A single organization-wide model choice buries those differences in an average.
This analysis works the other way round: task first, model second. In each unit we build the task list from the people doing the work, write the latency, cost, accuracy and confidentiality requirements against each task, and only then map a model to it. For some tasks a small local model is the right answer, for others a frontier model is, and for some the right answer is that the task should not be given to AI at all.
The same discipline applies to integration. Which system connects through which MCP server, which data never leaves the organization, and which decision stays with a human is recorded per unit. By the end of the fifth day you hold two documents: each unit’s own tool stack, and a 12-month plan showing the order they get built in — because starting in seven units at once is the most reliable way to finish in none.
If you want to see what the output of this analysis looks like, our unit-level MCP and tooling radar is a public example — not the one specific to your organization, but the same shape.
What the programme covers.
Unit-Level Task Inventory
Separately for marketing, sales, finance, HR, operations, legal and IT: the real task list as the people doing the work describe it, with volume and frequency.
Task-to-Model Mapping
Latency, cost, accuracy and confidentiality requirements are written against each task first; the cloud frontier / small / local model decision follows from them.
Integration-to-MCP Mapping
Which enterprise system connects through which MCP server; what can be reused from existing integrations and what genuinely has to be built.
Decisions That Stay With a Human
For each unit, the decisions that will not be handed to AI are written down explicitly — up front and with the reasoning, so they are not relitigated later.
Per-Unit Tool Stack
Each unit’s own stack: model, integration, interface and ownership. A structure that fits the unit’s work, rather than one stack imposed across the whole organization.
12-Month Rollout Sequence
What comes first and what comes after; dependencies, quick wins, and which step is waiting on which. Precisely so you do not start in seven units at once.
Who this is for — and who it isn’t.
Most vendors only answer the first half. The second half saves everybody a quarter.
A good fit if
- You have budget for AI but have not settled where to start at the unit level
- Different units have started using different tools and there is no organization-wide stack decision
- You suspect a single model or platform choice will not serve the whole organization
- You cannot settle the rollout order; every unit considers its own work the priority
Probably not if
- You have not run an organization-wide diagnostic yet — AI Awareness Analysis comes before this
- You want to focus on one problem in one unit; a seven-unit analysis would be overkill
- What you need is continuous comparative tracking of models rather than task-level mapping — AI Model Watch does that
Questions we get asked.
Is this the same as AI Model Watch and Technology Stack Analysis?
No — they answer different questions. Model Watch continuously compares models against the organization’s general needs; it is a six-month, ongoing programme. Unit-Level Mapping is one-off and narrower: it builds the task list for seven units and binds each task to a model and an integration. Most organizations do this first and move to watching afterwards.
Why exactly seven units?
Because most enterprise AI decisions concentrate on those seven surfaces: marketing, sales, finance, HR, operations, legal and IT. If your organization divides differently the list is adapted — what matters is not the number but that each unit’s own task profile is looked at separately.
Are five days enough for seven units?
Each unit gets a focused half-day session with preparation beforehand; the remaining time goes into the mapping and the rollout sequence. The aim is not to design every task in detail but to settle which task belongs to which model and which integration. Detailed design happens separately, on the first steps you choose.
Do you recommend a particular vendor?
No. The mapping falls out of the requirements written against the task: for some a cloud frontier model is right, for others a small model, for others a local model running inside the organization. And for some tasks the right answer is not to give them to AI at all — we write that down too.
Who uses the output?
The two documents go to two different tables. The per-unit tool stack is used by IT and the unit leads; the 12-month rollout sequence goes to the executive team and the board, for budget and prioritisation.
Often taken alongside.
AI Model Watch and Stack Intelligence
A 6-month analysis program comparing every cloud and local AI model against your organization's real needs — capability, cost and performance, in evidence-based tables.
AI Awareness Analysis
A 3-day diagnostic that establishes where your organisation stands on AI today — company-wide awareness sessions, needs gathering across units, IT and leadership strategy analysis, sector and company opportunity-risk analysis, and a detailed board report.
Agent and MCP Design Training and Health Check
A three-day-a-month programme that teaches a deterministic approach to AI agent design and audits the MCP installation already running in your organization — server inventory, authentication, permission scope and data access review; enterprise gateway, least privilege and telemetry design.
Let’s talk about Unit-Level Model and Solution Mapping.
Start with a free 90-minute assessment. We map where this fits, what it would touch and what it would change — then you decide.