Why the partner matters more than the model
Every AI consultancy can demo something impressive in a sales call. The gap between a slick demo and a system that runs reliably in your business — with your data, your compliance obligations, and your edge cases — is where most projects quietly fail. In a market moving as fast as Sydney’s, the differentiator isn’t which model a firm uses; it’s whether they can ship, operate, and hand over something that keeps working after the invoice is paid.
The questions below are designed to surface that difference quickly. A strong partner answers them concretely and without defensiveness; a weak one reaches for jargon. Use them in your first conversation — the answers tell you more than any case-study deck.
The 12 questions to ask before you sign
- Who actually writes the code? Ask whether senior engineers deliver the work, or whether it is handed to juniors or offshored after the pitch.
- Where will our data be processed and stored? For Australian businesses, insist on knowing whether personal information stays onshore and how the Privacy Act is respected.
- What happens when the AI is wrong? A serious partner designs for failure — human review, confidence thresholds, and the ability to reverse an automated action.
- How do you measure success? Look for outcomes tied to your metrics (hours saved, error rates, revenue), not model accuracy in a vacuum.
- Can you show a system you built that is still running? A live, in-production reference beats a prototype every time.
- How will this integrate with our existing tools? The value is in connecting your CRM, ERP, and email — not a standalone app you log into separately.
- What does handover look like? You should own the code, the documentation, and the ability to run it without the vendor.
- How do you handle model and data drift over time? Ask how they will know when performance degrades and what the plan is to fix it.
- What is your approach to security and access control? Who can see what, how are secrets managed, and how is access audited?
- What are the ongoing costs — realistically? Understand inference, hosting, and maintenance costs before you commit, not after.
- Who owns the intellectual property and the data you help create? Get it in writing.
- What happens if we want to leave? A confident partner makes exit easy; lock-in is a red flag, not a feature.
Red flags that should give you pause
- Guaranteed outcomes with no discovery — nobody can promise ROI before understanding your data and workflows.
- A demo that only ever runs on the vendor’s data, never a sample of yours.
- Vague answers about data residency, security, or what happens when the model fails.
- Pricing that hides ongoing inference and maintenance costs behind a low upfront number.
Blog
Insights, frameworks, and strategies from the Algorythmos team on AI, security, and data innovation.