Custom Agents for Processes No Product Covers
Microsoft, Google, and OpenAI all sell agent platforms now. What a small business is missing is not a platform: it is the person who designs, builds, tests, and looks after the agent. That is what we do, on whatever stack fits your business.
Sound Familiar?
The same problems come up again and again in New Zealand small businesses.
An agent platform is not an agent
Copilot Studio and tools like it are genuinely capable, and they are platforms: someone still has to map the process, build the agent, test the edge cases, and own it when it misbehaves. That someone is rarely on a small business's payroll.
Your process spans five systems and one human bridge
Email to spreadsheet to job system to accounting to courier portal, held together by one person retyping the same details four times. When they are away, the business slows down.
Your industry software has no AI in sight
Niche NZ industry systems and long-serving legacy software sit outside the tidy connector catalogues. Reaching them safely is real engineering work, and no platform does it for you out of the box.
The workflow lives in one person's head
The rules, exceptions, and 'oh, but for that customer we always...' knowledge is undocumented. Every departure or holiday is a risk the business quietly carries.
What the AI Agent Does
Built around your business, not a one-size-fits-all product.
Wired into your actual systems
Modern APIs, databases, file drops, even screen-level operation of legacy software. If a person can operate it, an agent almost always can too.
Executes processes end to end
Not suggestions, actions: the agent reads the input, applies your rules, updates every system involved, and reports what it did. Multi-step, multi-system, done.
Your rules, with approval gates
Routine cases flow straight through, exceptions and high-value cases stop for a human yes. Every action is logged, so you can always see exactly what happened and why.
Documents the process as it works
The agent's rules and memory live in plain files you can read. The workflow that used to exist only in someone's head becomes an asset the business owns.
How It Works
A steady, low-risk path from first chat to an agent you trust.
- 1
Discovery call
A free 30-minute conversation about the process that hurts most, the systems involved, and whether an agent can genuinely take it on.
- 2
Map the process honestly
We sit with the person who does the work and capture the real rules, including the exceptions. If the automation does not stack up, we say so here.
- 3
Build and shadow
The agent runs alongside the human first, doing the work in parallel while every action gets checked. Nothing goes live until it earns it.
- 4
Ongoing care
We monitor the agent, adapt it when systems or rules change, and stay the phone call away that custom automation needs.
A building supplies merchant with a 20-year-old inventory system
Orders arrived by email and phone as PDFs and scribbled notes, and one administrator keyed them into the legacy system, checked pricing, and sent confirmations. Orders queued when she was busy and stopped entirely when she was on leave.
An agent now reads each incoming order, validates products and pricing against the current price list, enters it into the legacy system, and drafts the confirmation. Clean orders flow through in minutes; anything unusual waits in her approval queue.
Same-day order entry became the norm instead of the goal, the single point of failure is gone, and the administrator now spends her time on customers rather than keying.
Illustrative example based on typical results. Your numbers depend on your business, which is exactly what the discovery call works out.
What It Costs
Honest, indicative numbers so you can decide whether the conversation is worth having. No pressure, no hidden extras.
- Discovery and honest feasibility assessment
- Process mapping with the people who do the work
- Agent build, integration, and shadow running
- Approval gates and full action logging
- Documentation of the process as a business asset
Covers hosting, monitoring, AI usage, updates when your business changes, and a real person to call when you need one.
What moves the price:- Number of systems and how hard they are to reach
- Legacy or niche software integration effort
- Volume, and how many exception paths the process has
Indicative ranges, excluding GST. Custom work varies more than anything else we do, so the free discovery call matters most here: you leave it with an honest yes, no, or number.
Common Questions
Isn't Copilot an agent now?
Yes. Copilot Studio has over a thousand connectors, Xero and Shopify included, and computer-using agents for older software. It is a real platform, and that is exactly the point: platforms need builders. Mapping your workflow, handling the edge cases, testing, and ongoing care is the actual work, whichever stack it runs on. We do that work, pick the right platform for the job (sometimes that is Copilot Studio), and stay responsible for the result.
Our main system is ancient. Is that a dealbreaker?
Usually the opposite: old systems are where the manual work piles up, so they are where agents pay off most. If it has a database, file exports, or even just a screen a person operates, there is almost always a way in. The discovery call will give you a straight answer.
What happens when the agent gets something wrong?
The design assumes it will, occasionally: routine actions are logged and reversible, anything consequential waits behind an approval gate, and the shadow-running period exists so mistakes surface while a human is still doing the job in parallel.
How long does a build like this take?
Simple single-system agents take a few weeks. Multi-system builds with legacy integration typically run six to twelve weeks including shadow running. You see it working on real cases well before the handover.
Ready to Put AI Agents to Work?
Tell us about your business and we'll show you how AI agents can save you time, money, and headaches.
Get in Touch
Phone
+64 204 105 1820What We Can Help With
Our Promise
We believe great technology should be accessible to everyone, not just the tech-savvy: