Straight answers to the questions people actually ask

Pilots are fixed-price and sized to scope. Smaller automations (a single workflow with one or two integrations) sit at the lower end of a typical mid five-figure investment. Voice agents, full mobile apps, or anything touching multiple enterprise systems run higher. The operations audit (a short, fixed-price engagement) produces a precise quote before you commit to any build. We don't publish exact figures because every engagement is genuinely different; the discovery call gives you indicative numbers in 30 minutes.

You do. Every line of code we write for you is yours, transferred under our standard engagement agreement on full payment. No ongoing licensing, no hostage architectures, no proprietary frameworks you can't extend without us. The only exception is general-purpose tooling we've open-sourced or use across clients, which stays under its existing licence.

Wherever you want it to. We deploy in the cloud region you choose: AWS or Azure Sydney for NZ businesses that need data resident in the AU/NZ region; Google Cloud Auckland or Sydney; or your existing hyperscaler if you have one. We tell you exactly which AI providers process which data, what they retain, and for how long. If you have specific compliance requirements (ISO, HIPAA-equivalent, government), we work to those.

It will, sometimes. We design every system with that assumption baked in. For high-stakes workflows we add human review checkpoints; for lower-stakes ones we add confidence thresholds and graceful fallbacks. The pilot phase explicitly tests failure modes before you scale. Anyone who tells you their AI has 100% accuracy is selling something.

Whichever fits the job. We're model-agnostic and work fluently with OpenAI (GPT family), Anthropic (Claude), Google (Gemini), and open-source models like Llama where they fit. Cost, latency, accuracy on your specific task, and where the data needs to live all matter. We benchmark candidate models during the audit so you understand the trade-offs.

Not under our default architecture. We use the commercial APIs of OpenAI, Anthropic, and Google, which are explicit about not training on customer inputs. We can also deploy models locally or in dedicated cloud instances if your data sensitivity requires it. We never share client data with anyone for training purposes, and we never use your data to train our own internal models.

Discovery is 30 minutes. The operations audit runs 1-2 weeks. A pilot build runs 4-8 weeks depending on scope. From first call to working software in production: usually 6-10 weeks. After that, ongoing improvements and new workflows happen continuously under a retainer or as discrete projects.

An agent is software that can make decisions and take actions on its own (e.g. a voice agent that handles an inbound call, deciding what to ask, when to escalate, when to book a meeting).

An automation is a deterministic pipeline that runs a fixed sequence of steps when triggered (e.g. a form submission triggers an email, a CRM record, and a Slack notification).

An integration is the plumbing between systems (e.g. connecting your CRM to your accounting software so they share data).

Most real solutions combine all three. Where you draw the lines matters less than picking the right pattern for the job.

Yes, but most of our work is with NZ businesses. Our default cloud regions and data sovereignty defaults are NZ/AU oriented. If you're based elsewhere we can still work with you; we just default the conversation to NZ business hours and tooling.

If your team has shipped two or more production AI workflows and feels confident, you probably don't need us for ongoing build. You might still want us for a specific capability (voice agents have particular pitfalls; multi-agent orchestration is its own discipline) or as a sounding board on architecture. Our training courses exist exactly for this case: we work alongside your team on your real workflows, transferring the skills as we go, so the capability stays in-house.

Then we'll tell you, on the discovery call, and you'll save the cost of an audit. Honestly, AI is a poor fit for plenty of workflows: highly creative work where consistency matters, low-volume tasks where the ROI doesn't clear the build cost, and processes where the existing manual approach is genuinely working well. If we don't see the leverage, we won't pretend it's there.

The trial is a demonstration suite, not a SaaS product you can subscribe to. The components inside it (the field-capture voice forms, the meetings transcription, the sales coach) are pattern templates we customise per client during a pilot build. If one of them is close to what you want, that pilot will likely run faster and cheaper.

Honestly: not yet, well. Mainstream TTS engines (ElevenLabs, OpenAI, Google) all mispronounce te reo Māori vocabulary. The technology will be there in 12-24 months but isn't today. We do voice agents in English (NZ accent) and other major languages well; if te reo Māori is mission-critical for your use case, talk to us and we'll be straight about what's currently possible.

We sign mutual NDAs at the discovery stage. We work to ISO 27001 patterns for data handling. All of our infrastructure and the major AI providers we use carry SOC 2, ISO 27001, and (for relevant ones) HIPAA-equivalent certifications. For specific regulated industries (financial services, health, government) we shape the architecture to meet your regulator's requirements; that conversation happens during the audit.

Genuinely fine. The audit document is yours, and several past clients have used it as the basis for an internal build or a competitive RFP with another vendor. We'd rather you make the right call for your business than feel locked in. Our hit rate on follow-on engagements is high enough that we don't need to play games to keep clients.

Yes. We won't share them publicly, but on a discovery call we'll connect you with two or three past clients in similar industries who've agreed to speak with prospects. Most of them are happy to be candid about what worked and what didn't.