What we mean by "process automation"
Most AI projects start by automating a single task. Process automation is bigger: it looks at an entire business process, quote-to-cash, claims-to-payment, hire-to-onboard, complaint-to-resolution, and asks where each manual touch happens, why, and which of those touches a modern AI workflow can absorb.
For most New Zealand SMBs the answer is "more than you'd think". Reading documents, classifying enquiries, drafting standard responses, looking up customer history, summarising meetings, generating reports, validating data, escalating outliers, most of this is now in scope for AI to do reliably, with a person checking the high-stakes points.
Processes we commonly automate
Invoice and AP automation
Inbound invoice → extract line items → match to PO → validate against approval rules → push to Xero/MYOB → flag exceptions for human review.
Quote-to-cash
Enquiry → qualified intent → priced quote → proposal sent → followed up → won/lost recorded → invoice raised. AI absorbs the read-and-judge steps.
Insurance and warranty claims triage
Claim form submitted → policy checked → required docs validated → routed to right adjuster with summary attached → audit log captured.
Compliance and incident reporting
Raw incident data → structured report drafted → reviewed by manager → submitted to regulator/internal, with full versioning of every edit.
Customer onboarding
Sign-up → identity check → document collection → CRM record created → welcome email sequence triggered → handoff to account manager.
Operational reporting
Daily/weekly/monthly summaries written from raw data, sales, ops, support, financial, delivered to the right inboxes on the right cadence.
Document review
Contracts, SoWs, RFPs, supplier agreements, extract key terms, flag deviations from your standard, summarise risks.
Inbox triage at scale
Shared support / sales / claims mailboxes, every message classified, prioritised, summarised, drafted-reply-attached, before a human even opens it.
How we measure success
In discovery we baseline the current process. Then we report against those numbers post-launch. Typical metrics:
Ranges reflect typical results across SMB process automation projects globally; we baseline against your actuals in discovery and report against those numbers, not these.
Designing for regulated industries
Insurance, financial services, healthcare and legal all need more than just speed. They need provable controls. Every process we build for regulated NZ businesses includes:
- Immutable audit logs. Every model invocation, prompt version, output, decision and downstream action, timestamped, signed, exportable.
- PII redaction. Sensitive data is identified and masked before prompts hit the LLM. Original data never leaves your boundary.
- Data residency. Default AWS Sydney; on-prem or NZ-only options where required by sector regulation.
- Human-in-the-loop on high-stakes decisions. Configurable thresholds, by value, by customer segment, by confidence, that route to a person before action.
- Versioned prompts and models. When a prompt or model changes, the change is logged and the prior version archived so any past decision can be re-traced.
- Fail-closed defaults. When the workflow is uncertain, it stops and asks for human review, never assumes.
Frequently asked questions
What's the difference between process automation and workflow automation?
Workflow automation chains specific steps together. Process automation looks at an entire business process end-to-end, for example, the complete quote-to-cash cycle, or the full claims-handling lifecycle, and identifies where AI can remove manual touchpoints across the whole thing. Most process automation projects involve building several connected workflows.
Which processes pay back fastest?
The processes with the highest manual touch and the most repetition: invoice processing, document review, claims triage, quote generation, customer onboarding, compliance reporting, and incident report writing. Anything where someone reads unstructured input, makes a judgement, and writes structured output.
How do you measure success?
We baseline current process metrics in discovery, time per case, error rate, throughput, cost per transaction, and report against those after launch.
What about audit and compliance?
Every step of every process is logged: input data, model used, prompt version, output, decision, downstream action, and human reviewer. Logs are immutable, timestamped, and exportable for audit. We design with regulated industries in mind.
Do we need to commit to a long contract?
No. Discovery is free, the build is a fixed-scope project, and ongoing support is a month-to-month retainer with no minimum term.
Ready to scope a process automation project?
Book a free 30-minute strategy call. Bring the process you want to look at and we'll talk through what's automatable, what isn't, and what realistic ROI looks like.
Get in touch