Tool-first
A deployment question
Which AI should we buy?
The shape of the work is taken as given. Technology is layered on top, and use cases are often selected from generic lists or vendor claims.
AI workflow mapping for organisations
Independent lab · Wellington, New Zealand
Deployed
Intelligence.
From AI deployment to intelligence allocation.
Map where AI could carry work, where human review is required, and where human judgment, presence or decision authority remains decisive.
Decision support for workflow and AI planning—not a job-replacement model.
A plain-language definition
Deployed Intelligence is an AI workflow-mapping methodology. It breaks organisational roles into tasks and steps, assesses the characteristics of each step, and applies a specified technology and policy profile to identify an appropriate working arrangement.
§ 01 · The planning problem
Plan the work,
not just the tool.
AI decisions often begin with models, vendors and licences. But the useful question sits inside the workflow: which parts of the work could a particular system carry, which require review, and which should remain under human control?
Work-first
An allocation question
Which working arrangement fits each part of the work?
Human capability and machine capability appear on the same workflow map—planned together without pretending they are interchangeable.
The aim is not to maximise machine participation. It is to design the right arrangement for each part of the work.
§ 02 · Start with one step
Inside one task,
the answer changes.
A job contains tasks, and each task contains steps. The appropriate relationship between a person and an AI system can change from one step to the next—even within the same few minutes of work.
Resource Consents Planner · Assess the environmental effects of a proposal
Human required
“Weigh how much a proposed building would affect each neighbour.”
What the analysis says
Value judgment: 7/10. Reasonable planners could
weigh the same facts differently. The step involves more than
retrieving a rule or calculating a measurement.
What decides the disposition
The judgment is the core work. A system may gather evidence,
compare plans and prepare analysis, but a person should form and
own the final assessment.
The steps around it
Pulling plans, mapping the site against the district plan and
tabulating measured effects may be
AI carries, human approves.
Same task, different working arrangement.
Five working arrangements
These describe participation in a step—not who is accountable for the organisation as a whole.
AI carries
The system performs the primary work. People govern and monitor the workflow rather than reviewing every instance.
AI carries, human approves
The system performs the primary work; a person checks the result before it has an effect.
Human leads, AI assists
A person performs the core step while the system drafts, retrieves, analyses or checks within it.
Human required
A person must perform or own the core decision because judgment, relationship, presence or accountability is essential.
Beyond this profile
The technology is not the right instrument for the step. This is a scope statement, not a judgment about the work’s value.
1
step
Weigh the effect on each neighbour
35
steps
One environmental-effects task
17
tasks
One resource-consents role
12
roles
One composite organisation
§ 03 · How the map is built
A provisional map.
Then evidence and review.
Language models help create and assess the map, but they do not make the methodology self-validating. Inputs, outputs, assumptions and practitioner corrections remain visible throughout.
Define the work
Collect role titles, organisational context, job descriptions and practitioner knowledge. Record what the analysis can and cannot infer from the available inputs.
Generate tasks and steps
Build a provisional representation of the role: common tasks, then the steps through which each task is usually performed.
Check it with practitioners
Ask people who know the work what belongs, what is missing, where the wording is wrong, and where the workflow branches or varies.
Score the work
Assess every step against the same versioned work dimensions, including consequence, value judgment, relationship sensitivity, context and procedural stability.
Apply technology and policy
Apply a specified technology capability envelope and a separately stated deployment policy. Keep “can it?” apart from “should it?”
Read the full AI workflow-mapping methodology →
Definitions, configurations, evidence status and limitations
§ 04 · The profile matters
Not “AI” in the abstract.
A specified system.
Every disposition depends on what technology is being assessed, what information and tools it can access, and what level of autonomy the organisation is willing to permit.
Can this system perform the step?
An empirical claim about a specified technology, setup and standard of performance. It should be tested and revised as systems change.
How much human control should remain?
A governance choice shaped by consequence, values, accountability, professional obligations, relationships and organisational risk appetite.
Human required because the stakes are grave
Building Consents OfficerAssess fire safety, including escape routes, alarms and fire separations.
Consequence: high
Value judgment: low
The assessment is largely rule-bound, but the downside of an error may involve serious safety consequences. A system may carry out checks and prepare evidence; a person retains authority over the call.
Human required because the judgment is the work
Resource Consents PlannerForm and defend a recommendation where the effects appear unacceptable.
Consequence: high
Value judgment: high
Reasonable professionals may weigh the same considerations differently. Who makes and defends the judgment is part of its legitimacy, not merely a quality-control step.
§ 05 · One role
The human core appears
inside the workflow.
Eight selected tasks from the 17-task Resource Consents Planner map. The complete analysis is available in the council case study.
AI carries
AI carries, human approves
Human leads, AI assists
Human required
Beyond profile
01
Check a lodged application for statutory completeness
96%AI can carry
02
Determine consent triggers and activity status
96%AI can carry
03
Assess the environmental effects of a proposal
57%AI can carry
04
Prepare a notification recommendation
82%AI can carry
05
Prepare a section 42A hearing report
72%AI can carry
06
Present planning evidence at a hearing
36%AI can carry
07
Issue the resource consent decision package
97%AI can carry
08
Monitor compliance with granted consent conditions
86%AI can carry
Explore the full council case study
All 12 roles, 229 tasks and 6,101 pipeline-generated steps.
§ 06 · Organisation view
One organisation,
mapped as a mix.
Roles are sorted by the share of steps the AI can carry, either directly or subject to human approval. That headline is only an entry point: the full distribution shows where assistance, human authority and physical scope still matter.
AI carries
AI carries, human approves
Human leads, AI assists
Human required
Beyond profile
Accounts Payable Officer
Finance · 359 steps94%AI can carry
Resource Consents Planner
Regulatory Services · 502 steps82%AI can carry
Building Consents Officer
Regulatory Services · 307 steps72%AI can carry
Communications Advisor
Strategy and Policy · 615 steps67%AI can carry
Customer Services Officer
Customer and Community · 379 steps61%AI can carry
Policy Advisor
Strategy and Policy · 787 steps59%AI can carry
Democracy Advisor
Governance · 459 steps58%AI can carry
Customer Services Team Leader
Customer and Community · 516 steps54%AI can carry
HR Advisor
People and Capability · 661 steps54%AI can carry
Community Librarian
Customer and Community · 452 steps33%AI can carry
Water Services Operator
Infrastructure and Operations · 632 steps28%AI can carry
Parks and Open Spaces Worker
Infrastructure and Operations · 432 steps23%AI can carry
What this map does—and does not—say
It shows how work is classified under one stated technology and policy profile. It does not show implementation cost, readiness, data quality, employee acceptance or actual system performance in deployment.
Moving a step into a dependable workflow may require integration, permissions, testing, assurance, process redesign, training and change management. The map identifies where to investigate; it does not remove the need for that work.
§ 07 · What an engagement produces
A shared evidence base
for better AI decisions.
Start with one role, one function or a wider organisational map. The objective is not a generic list of AI use cases, but a structured view of your own work.
A map of the work
Roles represented as tasks and steps, with practitioner corrections, missing work and workflow variation recorded.
A technology-specific opportunity view
The same work assessed against a versioned technology profile rather than against “AI” as a single, timeless category.
A control and risk view
Where review, professional judgment, human decision authority, relationship work or physical presence remains necessary.
An investment agenda
Which opportunities depend on retrieval, system access, workflow redesign, assurance, better data or organisational change.
A facilitated conversation
A common map for operational leaders, practitioners, digital teams, HR and risk owners to examine together.
Begin with a bounded pilot.
A role or function can provide a practical first test of the methodology before considering a broader map.
§ 08 · Evidence and limits
Inspectable claims,
not a magic score.
A language model participates in the analysis, so reliability, construct validity and external validation are central questions—not footnotes.
Versioned and traceable
Scores and dispositions retain the model configuration, rubric version, technology profile and named rule that produced them.
Tested for consistency
Internal experiments examine model choice, prompt structure, batching, context, score stability and whether important metric boundaries remain distinct.
Checked against practitioners
Practitioner research is being developed to test whether role maps, task decompositions and selected analytical judgments match the work as experienced.
Policy is labelled as policy
Capability claims, risk appetite and human-authority requirements are kept separate so organisational preferences are not presented as facts about technology.
Uncertainty remains visible
Borderline decisions, excluded metrics, profile assumptions and thresholds still awaiting calibration are recorded rather than hidden.
§ 09 · Frequently asked questions
The short
answers.
What is Deployed Intelligence?
Deployed Intelligence is a methodology for mapping how human and machine capability could be allocated across an organisation’s workflows. It analyses work at task and step level rather than assigning one automation score to an entire job.
How does AI workflow mapping work?
A role is mapped into common tasks, each task is decomposed into steps, and each step is assessed against a consistent set of work characteristics. A versioned deployment profile then considers the capabilities of a specified technology and the level of human control the organisation requires.
Does this predict which jobs will be replaced?
No. Jobs contain different kinds of work, and the appropriate arrangement often changes from one step to the next. The methodology maps possible working arrangements rather than treating a job as wholly automatable or non-automatable.
What is a deployment profile?
A deployment profile describes a specific technology and setup: its tools, access to information, perception limits and other capabilities. It also records the deployment policy used to determine human review and decision authority.
Is the output generated entirely by AI?
Language models assist with task generation, decomposition and scoring, but the methodology uses versioned definitions, fixed configurations, explicit decision rules and practitioner review. Every result retains its provenance and can be inspected.
What does an organisation receive?
Depending on scope, an engagement can produce task and step maps, technology-specific working-arrangement profiles, opportunity and risk views, integration priorities and facilitated review with people who know the work.
§ 10 · About the lab
Independent,
Wellington-based.
Deployed Intelligence Labs Limited
Independent research and advisory lab
- Based
- Wellington · New Zealand
- Focus
- AI workflow mapping
- Work
- Research, advisory and mapping
- Contact
- hello@deployed.nz
Deployed Intelligence develops a task-based method for understanding where AI may fit in organisational work—and where human capability remains decisive.
The work draws on senior experience across New Zealand government and the private sector, with a particular focus on how large organisations organise work, make decisions and adopt new technology.
The methodology is in active validation. Current work includes production-pipeline testing, practitioner research and applied organisational mapping.
The broader lab also develops applied information products, including The Brief. More about the lab and its wider work is available at deployed.nz.
§ 11 · Next step
Start with a
conversation.
Get in touch if you are considering a role or function to map, developing an AI workforce strategy, researching adjacent questions, or exploring a collaboration.
Thanks, there.
Your message is on its way. You will receive a direct response, usually within one or two working days.