Model
Which model does which work, pinned by role and workload; a documented path to swap it.
Adopting AI
Adopt AI the way you adopt any material capability: staged, governed, measured.
How we structure every engagement, in one page. Forward it to your board; it reads in five minutes.
The state of adoption
Where adoption actually stands.
Adoption is no longer the question. 88% of organisations use AI in at least one function; 5.5% report real financial returns from it.1 Close to three quarters plan to run AI agents within two years; 21% have a mature governance model for them.2 Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027.3
The organisations capturing value share a pattern: they adopt in stages, govern from the first stage, and measure from a baseline. The six frameworks below are how we build that pattern into every engagement.
The six-stage evolution
Adoption advances in six stages.
AI capability in an organisation advances through six recognisable stages.
- 01Pre-AIWork runs on human coordination alone.
- 02AutomationScripted, rule-based flows take the repetitive load.
- 03CopilotAI assists individuals inside their existing tools.
- 04OrchestrationAI coordinates work across systems, with humans directing.
- 05AgentsAI carries whole workflows; humans hold the gates.
- 06AutonomyAI operates the function; human judgement concentrates at money, legal, and risk decisions.
Most organisations sit at Copilot. The climb from there is a governance climb, not a model climb: the models are already capable of the later stages; what advances an organisation is the control structure around them. The advance is in capability; the control is constant.
We stage every engagement against this ladder: find where you stand, move one stage at a time, and govern each stage the same way.
The governance spine
Four columns a board should require.
Four columns govern every deployment we run. A board should require the same four of any AI programme.
Cost
Spend visible per user and per task; caps set before scale, reviewed as usage grows.
Security
Data residency, retention, and access defined in writing; every agent action logged and auditable.
Guardrails
What agents may touch, where humans gate, and what evidence each gate requires.
The economics of tokens
The same task, very different bills.
Two AI plans with similar capability can produce very different bills: on published prices, the cost to complete the same task can vary by an order of magnitude across models and plans. Seat fees and token consumption are different bills, and the second one scales with use.
Model choice is a financial control. We treat it as one: profile the workload, pin the default model per task, set per-user spend caps before scale, and instrument cost per task so the unit economics of every workflow stay visible.
The harness
The asset that survives the model.
The model is the horsepower. The harness turns it into governed work: the skills that encode how your organisation does things, the access that connects AI to your systems on your terms, and the memory that carries context between sessions.
The harness is the durable asset in an AI programme. Models improve quarterly and will be swapped; the harness survives the swap and compounds. Every engagement we deliver builds the client's harness as a first-class deliverable: documented, owned by you, portable across models.
Sovereignty and continuity
Your model is a supplier.
Boards manage concentration risk in every material supplier. The model behind your AI programme is a material supplier: it can be repriced, deprecated, or withdrawn, and under CPS 230 the dependency itself is a regulated matter for APRA-regulated entities.
The hedge is architectural. A model-agnostic harness lets you swap the model while the skills, access, and memory continue working, and lets critical workflows fall back to a model running inside your own perimeter. Adopted early, the hedge costs little; retrofitted late, it costs a programme.
Value realisation
Value shows up where it is measured.
Programmes that measure can show value. The method is four steps, and it runs inside every engagement.
- 01
Baseline
Before building, measure how the work runs today: cycle time, cost, error rate, volume.
- 02
Instrument
The delivery itself records throughput and cost per task as it runs.
- 03
Attribute
Credit the change conservatively, in ranges, on your own data.
- 04
Report
A recurring value statement against the baseline, in the numbers your board already uses.
The standard to hold any provider to, ourselves included: if the baseline is not measured before the build starts, the value conversation is already lost.
In practice
How this shows up in an engagement.
The six frameworks are how every pod runs. A pod (1–3 senior experts plus orchestrated agent capacity, sized to the engagement) stages the work against the evolution, runs it on the governance spine, instruments cost per task, builds your harness as a deliverable, and reports value against the baseline.
Begin
Book a scoping conversation.
A free working conversation on where you stand across the six stages, and the first move that fits.
Sources
- McKinsey & Company, The State of AI in 2025: Agents, innovation, and transformation, November 2025 (global survey, n=1,993).
- Deloitte, State of AI in the Enterprise, 9th edition, January 2026 (n=3,235, 24 countries).
- Gartner press release, 25 June 2025.
Regulatory references: Privacy and Other Legislation Amendment Act, automated-decision disclosure obligations, effective 10 December 2026; APRA Prudential Standard CPS 230, in force 1 July 2025; ASIC Report 798.