Skip to main content

From Copilot to the agentic mid-market

Five levels for moving AI from personal productivity to accountable business capability, for mid-market organisations in Australia and New Zealand.

White paper · 41 pages · PDF · Free: the first 5 pages here, the rest with your work email

  • Decide what AI may see: the sources it draws on, who owns each one and how current it must be.
  • Decide what AI may do: the highest action each agent may take, with a control for every step up.
  • Decide who answers for it: a named business owner, a record of what it did and a way to stop it.

About 45 minutes in full. For the short version, read pages 2, 4, 12 and 34.

~50%less onboarding administration at a multi-site aged care provider, from fixing the integration beneath its paperwork (page 4).

White paper · Edition 1.1

From Copilot to the agentic mid-market

The DataMust governed data and AI maturity model

Five levels for moving AI from personal productivity to accountable business capability, for mid-market organisations in Australia and New Zealand.

Jordan Whiting

Founder, DataMust

November 2026

Australia and New Zealand

After Copilot comes governed capability

Many organisations have given their people AI assistants such as Microsoft 365 Copilot. Far fewer have given AI the trusted data, governed access and accountability it needs to change how work gets done. Three leadership decisions close that gap.

Decide what AI may see

Name the sources AI may draw on, who owns each one and how current it must be. Answers are only as good as their provenance.

Chapters 2 and 4

Decide what AI may do

Set the highest action each agent may take, from reading to acting within limits. Every step up the ladder needs a matching control.

Chapters 3 and 4

Decide who answers for it

Give every AI capability a named business owner, a record of what it did and a way to stop it, before it goes live.

Chapters 3 and 5

~50%

less onboarding administration at a multi-site aged care provider, from fixing the integration beneath its paperwork. From our work, page 4

At a glance The five levels of governed AI in full on page 12

Assist

Connect

Specialise

Orchestrate

Scale

As AI gains more context and agency, the organisation must gain more assurance.

Where to start

Score one named process on five dimensions (page 34), look first at its weakest score, and plan the next 90 days (page 36). Or bring the process to a free 30-minute consultation and score it with us.

Five chapters from assistant to agent

01

Why AI stalls after the pilot

The gap between personal productivity and process change, and the five frictions behind it.

Exhibits 1 to 4

Page 5

02

Five levels of governed AI

The maturity model, with Assist and Connect in practice.

Exhibits 5 to 10

Page 10

03

Giving agents authority safely

The authority ladder, charters, orchestration, recovery, scale and ANZ obligations.

Exhibits 11 to 17

Page 17

04

The Microsoft foundation

Which capabilities do which job, and when Microsoft Fabric earns its place.

Exhibits 18 to 21

Page 25

05

Your next 90 days

Practice, measures, a self-assessment, a roadmap and an executive checklist.

Exhibits 22 to 28

Page 30

Written for

Leadership teams in mid-market organisations of about 50 to 1,000 people that run on Microsoft: chief executives, finance, technology and operations leaders, data leads, and risk and compliance.

Also: a client case on page 4, and About DataMust, reference and sources from page 38.

About 45 minutes in full. For the short version, read pages 2, 4, 12 and 34.

How to read the exhibits

Evidence Published research, cited on the page From our work DataMust client work, anonymised, as published Illustrative A worked example, not a result Unlabelled A DataMust framework: our model, not an industry benchmark

The fix sat beneath the paperwork

A multi-site Australian aged care provider wanted faster resident admissions. An obvious answer would have been electronic signatures. The useful answer was underneath them.

~50%

less onboarding administration

Approximate figure as published

12 weeks. The provider's first Fabric implementation, now the foundation for planned finance, HR and AI work.

DataMust case study, Aged care resident onboarding, 2026. Client anonymised. Read the case study

Where it started

Admission packs were printed and posted. The same resident details were keyed in again at every stage, and postal delays slowed admissions, and bed fill with them.

What we found

Electronic signatures alone would have removed the paper but kept the manual data entry. The real cost was the retyping between systems, not the paper itself.

What we decided

Fix the integration first. Microsoft Fabric became a hub between the systems, and a digital workflow on the Power Platform replaced the paper packs, so resident details no longer had to be keyed in at every stage.

How the admission flow changed

Before

Printed pack

Posted out

Signed on paper

Posted back

Keyed into admissions

By hand

Keyed into clinical

By hand, again

After

Digital workflow

Power Apps, Power Automate, DocuSign

Microsoft Fabric hub

Details entered once

Admissions system

Updated from the hub

Clinical system

Updated from the hub

Simplified from the published case study. System names generalised.

Digital transformation does not begin with AI.

The lesson for leaders looking at agents: find the step where people retype what a system already knows, and fix the data beneath it first.

01

Why AI stalls after the pilot

AI assistants spread quickly because they ask little of the organisation. Changes to how work gets done spread slowly, because they ask a great deal. The gap sits in data, access and accountability, and it widens as AI is asked to do more.

Illustrative. The shape of the gap, not measured data.

The adoption gap

Page 6

The reconciliation trap

Page 7

What research says

Page 8

Five frictions

Page 9

Pages 6 to 41

Read the rest of the guide

Enter your work email to open pages 6 to 41 here and download the PDF.

We use your email to follow up about this guide. Privacy policy