There are too many ideas.
Every function can suggest a use case. Few organisations have a practical way to compare value, data fit, risk and the likelihood of adoption.
AI adoption + agent services
Put AI to work on one real problem.
Graymatter helps mid-sized organisations choose the right pilot, build a working prototype, and turn what works into a safe, supported way of working.
Founder-led · Fixed-price entry products · Vendor-independent · Working across Australia
PILOT CANDIDATE BOARD
IllustrativeProposal first draftHighHighMedRecommend
Service inbox triageHighMedMed
Candidate screeningHighMedHighNot first
FIRST TESTCan a sourced draft reduce preparation time without lowering quality?
Human approval stays in place.
Where many teams are now
You are probably not starting from zero.
AI has already arrived through individual tools, vendor features and staff experiments. The missing piece is usually a shared decision about where it belongs in the work.
The demos look easier than the work.
A generic assistant can impress in ten minutes. Real work has messy source material, hand-offs, edge cases, permissions and people who know when an answer is wrong.
People want progress and control.
Leaders want a result. Staff, IT and risk teams need to know what data is used, where judgement sits, how quality is tested and who can stop the system.
No one has spare delivery capacity.
The team that understands the workflow is already busy. The team that owns the systems cannot carry every experiment. Good candidates wait—or grow as unmanaged shadow AI.
43%
of Australian SMEs reported some AI adoption across Dec 2025–Feb 2026.
65%
of non-adopters cited distrust in AI decisions or a preference for human control.
The adoption problem is not just access to a model. It is relevance, confidence and a way to move from an experiment to real work.
National AI Centre source ↗Why Phil Gray
More than thirty years of making new technology work inside real organisations.
Phil’s background is not limited to AI tools. He has led enterprise innovation, organisational change and adoption in environments where risk, legacy systems and human behaviour all matter.
See Phil’s backgroundBuilt and led Westpac’s internal innovation accelerator, The Garage.
Former Global Head of Standard Chartered’s eXellerator innovation lab in Singapore.
More than three decades across business, people, change and transformation.
Designs and operates governed agents, review loops and working AI services—not only roadmaps.
A bounded first move
The first 90 days should answer four questions.
Each stage produces something your team can inspect and an explicit decision about whether to keep investing. A weak candidate can stop early. A strong one earns the next step.
What is worth testing?
Map real workflows. Rank candidates. Name the owner, users, data, risk and measurable outcome.
Opportunity MapCan we make it work?
Build a narrow prototype. Test representative tasks, difficult cases and the points where a person must decide.
Working PrototypeWill it work here?
Put it in front of real users. Check quality, workflow fit, controls and what changes around the technology.
Controlled PilotShould we adopt it?
Review the evidence. Document ownership, monitoring, support and the safe path when AI is unavailable.
Adopt / revise / retireFour ways to engage
Start where the uncertainty is.
You may need to choose a pilot, prove a chosen workflow, move a prototype into use, or keep a live service healthy. We do not force every organisation through the same program.
Opportunity Map
Find the workflows worth testing. We score value, feasibility, data, risk and adoption—then recommend one first move and an honest not-yet list.
Decision: what to test firstSee the scopeWorking Prototype
Make the chosen workflow tangible. We build, test and demonstrate a working version against agreed acceptance criteria before a production commitment.
Decision: pilot, revise or stopSee the scopePilot to Practice
Put a promising prototype into real work with users, controls, training, measures and a clear owner. The aim is adoption, not a clever demo.
Decision: adopt, extend or retireSee the scopeOngoing AI Partner
Keep what is working healthy and improve it. We monitor, tune, document, assess model changes and help choose the next bounded opportunity.
Decision: what earns attention nextSee the scopeAgents, in ordinary language
An AI agent is a worker with a brief, tools and boundaries.
It can take in a task, gather allowed information, complete several steps and prepare an outcome. That does not mean it should act without supervision.
For most first pilots, the useful design is bounded: clear inputs, a narrow job, approved tools, visible sources, defined stop conditions and a person approving consequential action.
Understand assistants, automations and agentsAGENT / PROPOSAL-PREPBOUNDARY: DRAFT ONLY
- 01Trigger
A team member requests a first draft.
- 02Approved context
Tender, content library and current account record.
- 03Tools
Search, compare, extract, draft and cite.
- 04Human gate
Owner checks evidence, judgement and commitments.
- 05Record
Sources, draft, changes and decision retained.
NO EXTERNAL SENDNO PRICE COMMITMENTSTOP ON MISSING SOURCE
Candidate workflows
Start where judgement is valuable and repetition is expensive.
These are examples, not claimed case studies. The right candidate depends on your systems, data, people and tolerance for error.
Customer & growth
Prepare sourced account briefs
Assemble proposal first drafts
Route and draft service responses
Operations
Triage exceptions for human review
Compare recurring reports
Coordinate actions across systems
Knowledge
Answer from approved source material
Find evidence across document sets
Keep procedures easier to use
Corporate teams
Prepare management review packs
Support onboarding workflows
Draft and check routine documents
Often not a first pilot: high-impact employment decisions, unchecked customer commitments, autonomous financial actions, or any workflow where a confident error cannot be safely caught.
Selected work patterns
Built in the messy middle between a demo and a business decision.
These are anonymised examples of systems Phil has designed and used. Identifying details and outcome claims have been removed; the point is to show the shape of the work and where control sits.
Explore the systems in detailEvidence-led growth intelligence
An agent system that gathers public signals, verifies sources, reconciles them with the current account record and prepares a ranked action brief. Relationship-sensitive action stays with the account owner.
Human review and calibration
A controlled queue for uncertain AI outputs. Reviewers see the evidence, make a bounded decision and create a durable feedback record that improves later recommendations.
Source-verified briefing
A research workflow that assembles facts from approved public and internal sources, preserves lineage, separates inference from evidence and produces an executive-ready brief.
AI service guardian
A read-only monitoring agent that checks service health, schedules, dependencies and exceptions, then reports green, amber or red with the evidence needed for a human response.
CONTROL IS PART OF THE BUILD
Move quickly enough to learn. Carefully enough to keep trust.
We adapt the Australian Government’s six essential AI practices to the size and risk of the job. The controls grow with the consequence—not with the hype.
See our practical control model- 01Decide who is accountable
- 02Understand impacts
- 03Measure and manage risks
- 04Share essential information
- 05Test and monitor
- 06Maintain human control
How Graymatter works
A small senior team, amplified by AI—not hidden behind it.
You work directly with founder Phil Gray. Phil owns the brief, client decisions, judgement, quality and the work that affects people.
Graymatter’s own governed agents help research, prototype, test, compare and document. Their work is bounded and reviewed. That makes delivery faster and is the same human–AI operating model we help clients build.
When a job needs specialist legal, privacy, security, data or change expertise, we say so and work with the right people.
About Phil and the delivery modelQuestions to ask early
No black box around the engagement.
Do we need an AI strategy first?
Usually you need enough direction to choose responsibly, not a long strategy exercise. The Opportunity Map creates a decision-ready first roadmap. A broader strategy may follow when working evidence gives it substance.
Are you tied to one model or platform?
No. Existing Microsoft, Google, CRM, service and data environments matter, but the workflow and constraints come first. We choose the simplest suitable approach and document the trade-offs.
Will an agent replace a role?
That is rarely a useful first design question. We start with the work: where time goes, where errors occur, what judgement matters and what a better workflow would enable. People remain accountable for consequential decisions.
What if the prototype is not good enough?
Then it has done an important job cheaply. We record why, recommend whether to reshape or stop, and leave you with the evidence. Graymatter does not need every prototype to become a production build.
Can you work with our IT and risk teams?
Yes. They should be involved early enough to shape data access, security, procurement, ownership and controls—not asked to approve a finished demo after the fact.
How does fixed pricing work?
We agree the boundary, deliverables, client inputs, decision gate and exclusions before a stage begins. There is no hourly meter. If new information changes the scope materially, we make the choice visible before doing additional work.
A practical first conversation
Bring us one workflow.
Tell us about a repeated, expensive, slow or important piece of work. We’ll help you decide whether it is a sensible place to start.
Start with the workflow