Fix a stalled AI rollout
Copilot licences nobody uses, a chatbot that makes things up, a pilot that never reached production. We find out why and get it paying its way.
AI and agents
AI that works on your real data, and pays its way.
Most AI projects stall because the data underneath isn't ready. We start there, with clean models, secure access and clear ownership, then put AI to work where it saves real hours. Already invested in AI and not seeing the return? We fix that too.
Sound familiar?
If any of these are true, this is where we can help.
- We bought Copilot licences for everyone and hardly anyone uses them.
- Our chatbot makes things up, so nobody trusts what it says.
- We ran an AI pilot that looked great in the demo and never reached production.
- We're not sure what our staff are pasting into ChatGPT, or whether it's allowed.
- Everyone says we should be using AI, but nobody can tell us where it would actually help.
What we do
Demo on your data
A working session where we point AI at your own reports and documents, so you see what's real before you spend anything.
Readiness and security review
Data quality, permissions and Privacy Act 2020 obligations checked before AI goes live, plus prompt injection and data leakage testing on the AI you already run.
Custom agents
Agents that triage the inbox, chase overdue debtors or flag reconciliation exceptions, with a person approving anything that matters.
Custom skills
Your playbooks, review standards and processes written as reusable AI skills, so every agent works the way your team does.
Agentic engineering
The AI coding workflow we use ourselves, set up for your development team: agents, skills, review gates and guardrails.
Built in-house
Client data stays out of the model
Veil, our in-house tool, de-identifies client documents before they reach an AI model and restores the real values afterwards for authorised staff. It's the same thinking we bring to every AI project: useful, and safe by design.
More about VeilWhat your team uploads
Engagement letter for Jane Smith, IRD 123-456-789, re: Smith Family Trust
What the AI model sees
Engagement letter for CLIENT_0042, IRD TAXID_0007, re: TRUST_0003
How we work
We treat AI as an engineering project, not a magic trick. Every engagement starts with the data: what the AI will see, who is allowed to see it, and how good it is. Then we pick a narrow, measurable job, build it, and check the results against what a person would have done.
People stay in charge. Agents can draft, triage and flag, but anything that changes important records, sends money or reaches a customer goes through a person first. Every answer is traceable to its source, so your team can check it.
We follow New Zealand privacy law. Before any client or personal data reaches an AI model, we check the Privacy Act 2020 implications and, where needed, de-identify it first. See our guide to the Privacy Act 2020 and AI tools.
We build with AI ourselves
We use AI coding agents in our own delivery, each working to a written standard, with a person reviewing every change before it reaches a client’s systems. If your team is starting the same, read why your Copilot rollout isn’t paying off and AI agents or automated workflows?. We also build our own AI products. Veil finds names, account numbers and other identifying details in client files and swaps them for placeholder labels, so only cleaned text reaches the AI model. Authorised staff can restore the real values afterwards.
Further reading
All insightsLet's talk about what's slowing your business down.
Tell us about the report nobody trusts or the process that eats a day a week. We'll tell you straight whether we can help, and what it would take.