
AI services, from first workflow to production.
Advice when you need a plan, engineering when you need it built, and training so your team can run with it. Every engagement is measured against a number you already track.
Short answer
Levy Street provides AI consulting, custom AI development, AI agents, workflow automation, integrations, evaluation, governance and team training for New Zealand organisations. Most clients start with a scoped workflow assessment, then either buy an existing tool, build a narrow production system, or embed our engineers with their operators until the system runs reliably.
Advice and planning
Decide where AI is worth using before anyone writes code.
AI consulting
Independent advice on where AI pays off in your business, from people who build and run production systems.
ReadServiceAI strategy
A ranked, costed plan for where AI goes first in your organisation, written by engineers who will have to build it.
ReadServiceAI readiness assessment
A scored, evidence-based view of whether your organisation can use AI safely now, and what to fix first.
ReadServiceAI implementation roadmap
A dated delivery sequence for AI, where each step starts only when its data, systems and people are ready.
ReadServiceAI model selection
Which model should run each step of your system, chosen by testing on your own work, with a test you can rerun.
ReadServiceAI governance
A working AI policy, a register of every AI system, and review routines that match how your tools actually behave.
ReadServiceExecutive AI workshops
Decision workshops for leadership teams and boards: a ranked shortlist, owners, risk limits and a business case outline.
ReadBuild
Production systems wired into the software your team already uses.
Custom AI development
Production AI systems built around your data, rules and software, with acceptance criteria agreed before any code.
ReadServiceAI agent development
Agents with a short list of tools, permissions per action and human approval before anything irreversible.
ReadServiceAI workflow automation
Automations that read, sort and draft the messy parts of a process, with every exception routed to a person.
ReadServiceAI integration
Connect AI to Xero, Microsoft 365, your CRM or job system with scoped access and a named owner for every connection.
ReadServiceAI document processing
Invoices, forms and contracts read, checked and entered into your systems, with uncertain fields sent to a person.
ReadServiceRAG development
AI that answers from your own documents, with citations, a relevance benchmark and permissions that match the source.
ReadServiceAI chatbot development
Chatbots built for one job, answering from your approved content, with written rules for when a person takes over.
ReadServiceAI voice agents
Phone agents that answer, book and route calls, tested on NZ accents and place names, with a person one step away.
ReadServiceAI customer support
Ticket routing, answers drafted from your own help content, and escalation rules your customers can trust.
ReadServiceAI reporting
Reports where AI drafts the words, code computes the numbers, and every figure traces back to its source.
ReadServiceAI proof of concept
A narrow build on your own data, judged against go/no-go numbers agreed up front, with the cost of production priced in.
ReadRun and improve
Keep systems accurate, safe and affordable once real people depend on them.
Forward deployed engineers
Levy Street engineers embedded with your operators until an AI system runs in production and your team can own it.
ReadServiceLLM evaluation
Test sets built from your own work, scoring you can audit, and regression checks before every model or prompt change.
ReadServiceAI security review
A scoped, hands-on test of one AI system: prompt injection, data leakage, tool permissions and runaway costs.
ReadServiceAI system rescue
Diagnose and fix AI systems that stalled, drifted or got expensive, starting with where the failures actually come from.
ReadServiceManaged AI operations
We run your AI systems after launch: monitoring, incidents, tested releases and vendor changes, under a written scope.
ReadServiceAI training
Hands-on AI training built on your own documents and tools, with free follow-up lessons for every participant.
ReadServiceMicrosoft Copilot training
Hands-on Microsoft Copilot training on your own files, starting with the settings that decide what Copilot can see.
ReadServiceChatGPT training
Hands-on ChatGPT training for teams on ChatGPT Business, from workspace settings to shared GPTs and checked analysis.
ReadServiceClaude training
Hands-on Claude training on Claude Team, from sharing settings to projects and artifacts on your own documents.
ReadQuestions people ask
Where should a New Zealand business start with AI?
Start with one workflow that is repetitive, measurable and costly when it is slow: inbox triage, document handling, reporting or research are common first picks. Map it with the people who do it, decide what AI may and may not do, and measure the result before expanding. A workflow assessment does exactly this.
Do you only build custom software?
No. Often the right answer is an off-the-shelf tool such as ChatGPT, Claude or Microsoft Copilot, configured well and backed by training. We recommend a custom build only when the workflow depends on your own data, rules or systems in a way a generic product cannot handle. See custom AI versus off-the-shelf tools.
Do you work outside Wellington?
Yes. We are based in Wellington and work with organisations across New Zealand, including Auckland and Christchurch. Discovery is usually on site; build work runs remotely with regular in-person sessions.
Have something to build?
Tell us about the workflow. We will map where AI can change speed, margin or decision quality, and what it would take to run it in production.