Why most NZ SMB AI projects fail (and the fix)

Most small NZ businesses that try AI and quietly drop it don't have a tech problem. The model was fine. What broke is the workflow, and one rule catches most of it: if a human still has to copy work from one tool to another, your AI isn't working yet.

The demo that never became a workflow

Someone on the team, often the owner, finds that ChatGPT or Claude can draft a quote, summarise a meeting, or sort a month of bank lines. It works the first time, and the obvious thought is: we should do this every day.

Three weeks later nobody is. The demo worked. The workflow never showed up.

A demo is one person at a screen asking a question and getting a good answer. A workflow is the same task running every Tuesday at 2pm whether or not the person who set it up is in, with a clear input, a clear output, and a clear handover to whatever happens next. The gap between them isn't technical. It's who runs it, when, with what input, where the output goes, who checks it, and what happens when it breaks. None of that is hard. It just has to be decided.

The human is the integration

This is the one the rule catches head-on, and it's the most common reason an "automation" saves no real time.

Someone sets up ChatGPT to draft customer follow-ups. It generates the draft. A person copies it into Outlook, updates the CRM to mark the customer as followed up, then ticks the box in the spreadsheet that tracks the week. That isn't automation. That's a person doing four tool transitions, one of which is "ask for a draft." The AI saved a few minutes of typing. The rest is still hand work.

The real question was never "did the AI write it?" It's "did the work move from one tool to the next without a person carrying it across?" If a person is the bridge between two systems, you haven't automated the expensive part. You've just changed which keys they press.

That bridge is what integration tools exist to remove. Zapier, Make, and n8n wire one app's output straight into the next. Claude managed agents go further: they hold context across a whole job, reach your authenticated tools through MCP connectors, and run on a schedule with every step traced. The leverage was never in the drafting. It's in the connections.

The "AI strategy" nobody owns

Same problem, bigger. A business decides to "do AI." There's a workshop, a list of opportunities, real enthusiasm. Six months on, almost nothing is in production.

The missing piece is ownership. "We're going to use AI for customer service" is not an outcome. "By the end of next month our after-hours enquiries get a triaged auto-reply that captures contact details and lands in Sarah's inbox by 8am" is an outcome. One person, one workflow, one measurable result, one deadline. If the project doesn't reduce to that shape, it doesn't ship.

This isn't really an AI problem. It's an execution problem that hits AI projects hard, because the tech is new enough that everyone wants a say in the decisions and nobody wants to be the person on the hook for it going live.

Automations drift, and nobody's watching

This one bites the teams that have already had some success. They set up a couple of automations, the automations run, everyone stops thinking about them. Then a tool changes, or a process changes, and the thing has been quietly producing wrong output for weeks. Sometimes it reaches a customer. Sometimes a report shows the wrong number all quarter and a decision gets made on it.

Automations need a heartbeat. Not a dashboard, just a simple "did this run, did the output look sane" check that someone actually reads. Weekly is usually enough. Treat the automation like a staff member: it has a job, it has a manager, and someone notices when it doesn't show up.

Chase the connections, not the drafting

The sequence I run to fix one of these is the same every time.

1. Map the workflow as it actually runs. Every click, every tool, every handover, unsanitised. The first version is always uglier than people expect. That's the point.
2. Mark every human-copy step. Every place a person moves information between systems. Those are the leverage points.
3. Close the most frequent one first. Volume times annoyance is the priority order. Sometimes that's a Zapier or Make connection, sometimes swapping a tool for one that connects properly, sometimes a small Claude agent watching one inbox and posting the result where it needs to go.
4. Add the heartbeat, then take the next one. Whoever owns it gets a weekly did-it-run check. Then back to step three for the next-most-frequent gap.

Over a few months that isn't a transformation. It's a quietly more reliable business where the team spends less time being a bridge between systems and more on the work that needs them.

Small NZ teams have the edge

Small teams have a real advantage here that big organisations don't. You can map your workflows on one page, change a process by walking across the office, decide on Monday and have it running by Friday. Almost none of that is true at 1,200 people.

So the teams I see getting value from AI aren't the ones with the biggest budgets or the fanciest tools. They're the ones that took their actual workflows seriously and closed the human-copy gaps one at a time. That work is more available to a twelve-person business in Nelson than to a large one anywhere.

If your AI hasn't stuck yet, my bet is the tech didn't fail you. The workflow shape just never got finished.

If you want a hand with it, mapping the workflows and picking the first human-copy gap worth closing, that's the work I do. The framework for picking the first automation is the natural next step, and if you're heading toward work that runs without you, getting started with Claude managed agents is where that goes. Or just book a conversation and we'll work through it.