Services/AI and automation
AI that does the admin, with a person checking the parts that matter.
Document processing, enquiry handling, approvals and reporting, built into the tools you already use. I start from your process, not from a demo of what AI can do.
Sound familiar?
Someone retypes PDFs into the CRM
Invoices, applications, signed forms. The data is right there, and a person keys it in anyway.
Enquiries wait for a human to triage them
Hot leads sit in the same inbox as spam and supplier newsletters until someone gets to it.
Monday is report-building day
Six exports, one spreadsheet, two hours. Every week. The same numbers in the same cells.
You tried ChatGPT and it didn’t stick
Chat isn’t a workflow. The useful version runs in the background, inside your systems, and only asks a person when it’s unsure.
What I build
Each of these plugs into the systems you already run. Nothing here needs you to move to a new platform.
- Document extractionPDFs, scans and emails read into structured records in your CRM, accounting or practice system, with a review queue for anything uncertain.
- Enquiry classification and routingInbound email and forms sorted by intent and urgency, sent to the right person with the context already attached.
- AI agents with a human in the loopMulti-step tasks like drafting a quote or preparing an application, done by an agent and approved by a person before anything leaves the building.
- Workflow automationOnboarding, approvals, notifications and follow-ups that happen because a record changed, not because someone remembered.
- Automated reporting and reconciliationThe Monday report built from the source systems and in your inbox before you are.
- Internal assistants on your own documentsStaff asking questions of your policies, procedures and past work, with answers that cite where they came from.
How I approach it
1
Human in the loop by default
Anything that touches money, clients or compliance gets a review step until the system has earned the right to skip it. You decide when that is.
2
Fail loudly, never silently
Logs, retries, confidence thresholds and alerts. When the model isn’t sure, it says so and a person gets a notification, not a wrong record.
3
Clear about where data goes
Which provider sees what, what gets stored and for how long, written into the scope before anything is built. Client data stays in Australian regions wherever the provider offers one, and nothing is used to train a model.
Related work
All case studiesQuestions I get asked
Do I need to replace my CRM or accounting system?
No. Almost all of this sits around the systems you already have and writes into them. If a system genuinely needs replacing, I’ll say so, but that’s a separate conversation.
What does it cost to run?
Model usage is usually small next to the admin it replaces, but it’s not zero. I estimate the monthly running cost in the scope so there are no surprises on the first bill.
What happens when the AI gets it wrong?
That’s what the review step is for. Nothing irreversible happens without a person until you decide the system has earned it, and even then every action is logged so it can be traced and undone.
Which part of your week should a system be doing?
Describe the manual process and I’ll tell you what I’d automate first, and what I’d leave alone.