Services
Artificial intelligence, put to work.
Most organisations do not have an AI problem. They have a prioritisation problem, an integration problem and a governance problem. AI has made all three urgent.
Symptoms, not specifications
Where this usually starts
Nobody books this conversation because they want AI. They book it because of one of these.
- A pilot worked in a demo and has been stuck at pilot for months.
- Staff are already pasting client information into chat tools nobody has approved.
- A process depends on someone reading documents, and that someone is the bottleneck.
- The board has asked for an AI position and nobody wants to invent one.
- You have a licence estate you are paying for and cannot show the value of.
Where we help
What this covers
Opportunity assessment and use-case selection
We map your processes against four tests: volume, variability, the value of a correct decision, and the cost of a wrong one. The output is a scored shortlist that survives a board conversation, not a wishlist of everything AI could theoretically touch.
Business case and ROI modelling
Build effort, licensing and run costs modelled against measurable reductions in cycle time, rework and error rates. Assumptions are stated openly so your finance team can challenge them.
Architecture and platform selection
Model choice, hosting and data residency, retrieval strategy, integration pattern and the build-versus-buy decision, assessed against the stack you already run. We document the trade-offs, including the ones that argue against building.
AI policy, governance and enablement
An acceptable-use policy staff will actually follow, a risk register aligned to the Privacy Act and the Australian Privacy Principles, and role-based enablement.
How we build it
The gate is the product.
The model reading a document is the easy half. What makes it safe to run unattended is what happens next: a confidence threshold, a person on the low-confidence path, a written record of every decision, and a correction loop that makes next month better than this one.
Intelligent automation and AI agents
Traditional automation follows a fixed path and breaks the moment a format changes. Intelligent automation combines language models with orchestration, integration and enterprise controls, so processes that depend on reading documents, interpreting context and applying judgement can run end to end.
Cognitive document processing
Structure extracted from contracts, invoices, applications, statements and email threads, handling the variation that defeats template-based OCR.
Agent orchestration
Multi-step agents that plan, call tools, query your systems and complete work, coordinated through MCP and API integration rather than brittle screen-level automation.
Guardrails and human review
Confidence thresholds, approval gates and full decision logging. Low-confidence cases route to a person, and the resolution feeds back into the system.
Enterprise integration
Deployed into HubSpot, Salesforce, Microsoft 365, SharePoint, Google Workspace, BigQuery and custom systems through secured APIs.
Why structured agents beat ad-hoc chat
Anyone can get a good answer out of a chat window. The hard part is getting the same quality of answer from five different people, every time, six months from now. A structured agent carries the standard with it: the same context, the same checks, the same output format.
On your side of the table
What you actually receive
Every engagement ends in artefacts you keep, not a deck you file. The list is the same whether or not you appoint us for the build that follows.
- A scored use-case shortlist, with the ones we recommend against and why
- A business case your finance team can pull apart, assumptions stated
- A working proof of value running against a realistic sample of your data
- An evaluation set, so quality is measured rather than asserted
- Acceptable-use policy, risk register and role-based enablement material
- Decision logs and monitoring on every automated step in production
Delivery lifecycle
How an engagement runs
Discover
Process mapping, data readiness and ROI ranking of automation candidates.
Design
Architecture, guardrails, escalation paths, success criteria and governance.
Build & tune
Development, integration and evaluation against a labelled test set.
Deploy & operate
Phased rollout, monitoring, tuning and a named support path.
Realistic timelines
A proof of value is typically live in four to six weeks. Full production deployment usually runs three to six months, driven by integration depth and data quality far more than by the AI itself.
AI Usage Audit
You have already bought AI. This is about whether it is actually working.
Most Australian businesses now hold licences for Copilot, ChatGPT, Claude or Gemini, or an AI feature bundled into their CRM. It works until a token expires, a permission turns out broader than anyone realised, or three teams solve the same problem three different ways.
The six domains we examine
Platform & licensing
Which tools are in use, by whom, on which plan. Overlap, shelfware, and the licence tier quietly capping what your team can do.
Workspace configuration
Admin settings, retention and training-data controls, knowledge structures, model defaults, and who holds administrative rights.
Connectors & integrations
MCP servers, API keys, Microsoft 365, SharePoint and CRM connections. Scope of access, token lifecycle and failure modes.
Prompt & agent estate
What has been built, by whom, and whether it is maintained, duplicated across teams, or abandoned after its author moved on.
Security & data handling
Where data goes, what is retained, who can see what, and how that sits against your Privacy Act and Australian Privacy Principles obligations.
Adoption & capability
Real usage measured against licences held, where value is landing, and the skill gaps holding the rest of the organisation back.
Your deliverables
- Findings report: every issue with evidence, severity and business impact
- Maturity scorecard: all six domains rated against a defined target state
- Remediation roadmap: quick wins, structural fixes and longer-term investment
- Reference configuration: the recommended setup, to settings level
- Enablement session: a working handover with your admins and power users
You likely need this if
- Your MCP or connector setup has no clear owner
- Access tokens expire and break live workflows
- Staff use personal AI accounts for company work
- No one can say where prompts and uploads are stored
- Adoption is uneven and the return is not evidenced
- Three teams have built the same assistant separately
Typically two to three weeks. Delivered remotely, or on site where that is useful, and run as a standalone engagement. There is no obligation to appoint us for the remediation we recommend.
Frequently asked
Questions buyers actually ask
What does an AI usage audit cover?
Six domains: platform & licensing, workspace configuration, connectors & integrations, prompt & agent estate, security & data handling, and adoption & capability. You receive a findings report, a maturity scorecard, a remediation roadmap, a reference configuration and an enablement session, typically over two to three weeks, with no obligation to appoint us for the remediation we recommend.
How long does an AI proof of value take?
Typically four to six weeks. Full production deployment usually runs three to six months, driven more by integration depth and data quality than by the AI itself.
Do you work with a specific AI vendor?
No. We recommend Anthropic Claude, OpenAI, Microsoft Copilot or Google Gemini based on what fits your stack, budget and risk appetite, not a partner agreement.
Further reading
We've written this argument out at length.
Some of the organisations we work with
Start with a conversation, not a proposal.
Tell us the problem. We’ll tell you honestly whether it’s worth solving, and what solving it would take.