Every consultancy in Australia added “AI” to its homepage in the last two years. Picking a genuine AI development partner out of that noise comes down to asking questions that can’t be answered with marketing. Here are the ten we’d ask if we were buying.
1. “Show me something you built that’s still running.”
Demos are easy; production systems that have survived contact with real users for six months are not. Ask for a system in production, what broke in the first month, and how it was fixed.
2. “Who owns the IP — including the prompts and training data?”
The only acceptable answer is you do. Some providers retain ownership of “their” components, which quietly turns into vendor lock-in. Get IP assignment in the contract, covering code, prompts, fine-tuned models and documentation.
3. “Where does my data go?”
A competent partner can tell you exactly which providers see your data, under what agreements, whether it’s used for model training (it shouldn’t be), and what stays onshore. If the answer is vague, the architecture is too.
4. “What happens when the model is wrong?”
AI systems fail differently from normal software — confidently and plausibly. Listen for review loops, confidence thresholds, human-in-the-loop design and fallbacks. Anyone who claims their system doesn’t make mistakes hasn’t shipped one.
5. “What’s the smallest version of this you’d build first?”
Good partners shrink scope; bad ones expand it. A partner who proposes a 4–8 week MVP before a big build is protecting your money, not their pipeline.
6. “What does maintenance look like — and cost?”
Models get deprecated, APIs change, your data drifts. A system without a maintenance plan is a future emergency. Expect a clear answer on monitoring, updates and what’s included in ongoing support.
7. “Which of my existing systems will this integrate with?”
The value of AI automation usually lives in the integration — the AI reading from and writing to your CRM, job system or accounting platform. A partner who hasn’t asked about your stack by the second meeting is building in a vacuum.
8. “Who will actually do the work?”
In larger firms, the people who pitch are not the people who build. Meet the engineers. Ask how many other clients they’re spread across.
9. “How will you measure whether this worked?”
Hours saved, error rates, turnaround time — agreed before the build starts. A partner reluctant to commit to metrics is telling you something.
10. “What would you NOT automate in my business?”
Our favourite filter. A genuine expert has opinions about where AI is the wrong tool — high-stakes judgement calls, thin-data processes, things a $200 SaaS tool already solves. “Everything can be automated” is a sales answer, not an engineering one.
Run the test
Take these ten to every shortlisted provider — including us. Book a consultation and ask away; if we’re not the right fit for your problem, we’ll say so and point you somewhere better.