How to Choose an AI Consultancy (And What to Watch Out For)

Choosing an AI consultancy comes down to four things: do they invest time understanding your problem before proposing solutions, can they explain their technical approach in plain language, do they have a track record of delivering working systems (not just prototypes), and will they support you after launch? The right consultancy acts as a partner, not a vendor. The wrong one burns your budget on impressive demos that never make it to production.

The AI consultancy market has exploded. Every web agency, IT consultancy, and freelance developer now offers "AI services." Some are excellent. Many are not. And the difference between the two can cost you tens of thousands of dollars and months of wasted time.

This guide will help you separate the genuine experts from the opportunists, ask the right questions, and choose an engagement model that protects your interests.

What Does a Good AI Consultancy Actually Do?

A credible AI consultancy follows a structured process. Here is what that looks like:

1. Discovery and Scoping

Before writing a single line of code, they should spend time understanding your business, your processes, your data, and your actual problem. This is not a free "strategy call" that turns into a sales pitch. It is paid work that produces a tangible output: a clear scope document, technical approach, timeline, and budget.

What to look for: They ask more questions than they answer. They push back on assumptions. They might tell you AI is not the right solution for some of your problems.

2. Building and Iterating

Good consultancies build incrementally. They ship a minimum viable product, get it in front of real users, collect feedback, and iterate. They do not disappear for three months and emerge with a finished product that does not match what you needed.

What to look for: Regular check-ins (weekly or fortnightly). Working demos at each stage. Willingness to change course based on what they learn.

3. Testing and Validation

AI systems need thorough testing beyond standard software QA. This includes testing for hallucinations, edge cases, bias, and performance under load. A good consultancy has a defined testing methodology, not just "we'll try it and see."

4. Deployment and Handover

The system needs to work in your environment, with your data, used by your people. Good consultancies handle deployment, provide documentation, and train your team.

5. Post-Launch Support

AI systems are not "set and forget." Models drift, data changes, edge cases appear, and users discover new needs. Your consultancy should offer ongoing support, monitoring, and maintenance as standard.

What Are the Red Flags When Evaluating AI Consultancies?

Watch out for these warning signs. Any one of them should give you pause. Multiple should send you running.

Red Flag	What It Usually Means
No discovery phase	They will build what they want to build, not what you need
Promising "AGI" or "superintelligence"	They are selling hype, not practical solutions
Cannot explain their technical approach simply	They may not fully understand it themselves
No post-launch support offering	They will disappear when problems arise
Guaranteed specific accuracy percentages upfront	AI performance depends on your data; honest consultancies caveat this
All projects use the same technical stack	They have a hammer and everything looks like a nail
No references or case studies	They may not have delivered real projects
Quoting before understanding your problem	The price is arbitrary and the scope will blow out

What Questions Should You Ask an AI Consultancy?

Use this checklist in your initial conversations:

About Their Process

• What does your discovery phase look like, and what does it produce?
• How do you scope AI projects when outcomes are inherently uncertain?
• What happens when the project reveals that the original approach will not work?
• How do you handle scope changes?

About Their Technical Capability

• Which AI models and frameworks do you typically work with, and why?
• Can you walk me through the architecture of a recent project (at a high level)?
• How do you test AI systems for accuracy, hallucination, and bias?
• What is your approach to data security and privacy? (If you are in Australia, ask specifically about data sovereignty.)

About Delivery and Support

• What does your post-launch support look like?
• How do you handle ongoing model updates and maintenance?
• Will we own the code and intellectual property?
• What happens if we want to bring the work in-house later?

About Track Record

• Can you share references from similar projects?
• What is a project that did not go as planned, and how did you handle it?
• What percentage of your projects make it from prototype to production?

That last question is particularly telling. The industry-wide rate of AI projects making it to production is around 30-50%. A consultancy that claims 100% is either cherry-picking or not being honest.

Which Engagement Model Should You Choose?

Model	How It Works	Best For	Risk Profile
Fixed Price	Defined scope, defined price, defined timeline	Well-defined projects with clear requirements	Low risk for client; consultancy bears overrun risk
Time & Materials	Pay for hours worked at agreed rates	Exploratory projects, evolving requirements	Higher risk for client; need active budget management
Retainer	Monthly fee for ongoing access and support	Long-term partnerships, ongoing AI needs	Moderate; predictable cost, flexible scope
Hybrid (Fixed Discovery + T&M Build)	Fixed-price scoping phase, then T&M for development	Most AI projects (recommended starting point)	Balanced; de-risks the unknowns

Our recommendation for most first-time AI projects: start with a fixed-price discovery phase ($3,000-$8,000 for most projects). This gives you a clear scope, technical approach, and budget estimate before you commit to the full build. If the consultancy is not the right fit, you walk away with a valuable scoping document that any competent team could execute on. For context on what that build might cost, see our guide to AI project costs.

How Should You Evaluate AI Consultancy Proposals?

When comparing proposals, look beyond the price tag:

• Specificity: Does the proposal reference your specific business context, or could it be sent to anyone?
• Honesty about limitations: Does it acknowledge what AI cannot do for your use case?
• Clear deliverables: Can you point to specific, measurable outputs at each stage?
• Risk mitigation: How does the proposal handle the scenario where the approach does not work as expected?
• Post-launch plan: Is ongoing support included, or is it an expensive add-on?

A Note on Transparency

We are an AI consultancy ourselves, so we have an obvious interest here. We are not going to pretend otherwise. What we will say is that the questions and red flags in this guide are the same ones we would want you to ask us. We believe that organisations who ask hard questions end up with better outcomes and better consultancy relationships, including with us.

If you would like to understand how we approach AI projects, you are welcome to explore our strategy services or read our honest take on which AI models work best for different business use cases.