A business doesn't need an AI strategy. It needs a few specific jobs done faster, in tools it can afford, with someone to fix them when they break. The warning signs below are about telling those two things apart.
We sell AI consulting, so read this with that in mind. Every test here applies to us too. At the end we say how we try to pass them.
1. The deliverable is a deck, not a plan
What it looks like: lots of talk about "transformation" and "AI maturity", ending in slides with themes and principles. No named tools, no prices, no order of work.
Why it happens: a strategy deck is quicker to write than a build plan, and it can't be wrong in any way you can check.
Ask: "Can I see a sample of the final deliverable?" A good answer is a real document for a made-up business, with tools, costs and next steps you could act on without them.
2. They automate the process you already have
What it looks like: nobody asks why a step exists. The messy spreadsheet gets a bot bolted on instead of being replaced.
Why it matters: this is an old mistake. In 1990 Michael Hammer argued in the Harvard Business Review that companies were using technology to speed up old ways of working instead of rethinking them. His article was titled "Reengineering Work: Don't Automate, Obliterate". AI makes the same mistake easier, because it can automate almost anything, including steps that shouldn't exist.
Ask: "What would you remove or change before automating anything?"
3. The demo is the product
What it looks like: an impressive demo on clean sample data. Nobody has tried it on your real inbox, your odd job codes or your customers' spelling.
Why it matters: the gap between a demo and something that works every day is where most of the effort goes. A demo doesn't show what happens when the AI gets something wrong, who notices, and how it's fixed.
Ask: "What happens in week two when it gets something wrong? Who fixes it, and what does that cost?"
4. Everything is an "agent"
What it looks like: a chatbot, a set of rules or a scheduled email gets called an "AI agent".
Why it matters: Gartner has a name for this, "agent washing": rebranding existing AI assistants, robotic process automation and chatbots as agentic AI without real agentic capability. In June 2025 Gartner estimated that only about 130 of the thousands of vendors claiming agentic AI were the real thing, and predicted that over 40% of agentic AI projects would be cancelled by the end of 2027 because of rising costs, unclear business value or weak risk controls.
Plain automation is often exactly what a business needs. The problem isn't the automation. It's paying agent prices, or accepting agent-sized risks, for something that isn't one.
Ask: "What decisions does it make on its own, and what happens when it's wrong?"
5. Hourly billing with no end point
What it looks like: a day rate, a "discovery phase" with no fixed length, and a scope that grows.
Why it matters: hourly billing rewards the consultant for taking longer. AI work is hard to estimate, which is exactly why the risk should sit with the person who knows how long it takes.
Ask: "What is the fixed price, and what's out of scope?"
6. Silence on data
What it looks like: no mention of where your customer data goes, whether the vendor trains on it, or the Privacy Act.
Why it matters: you stay responsible for personal information you put into AI tools. Since 1 May 2026, the Privacy Act's new IPP3A rule also covers information about people you collect from other sources, which many AI tools do. Our IPP3A guide explains it.
Ask: "Where will our customer data be stored, and does the vendor train on it?"
A bonus one: the one-tool consultant
Some consultants are resellers or partners for a single platform, and every problem turns out to need that platform. That isn't always bad, but you should know.
Ask: "Do you get commission or partner benefits from any tool you recommend?"
Watch the promises too
"Save 10 hours a week" is easy to say. In New Zealand, section 12A of the Fair Trading Act makes it an offence to make an unsubstantiated representation. The Commerce Commission puts it simply: a claim is unsubstantiated if the business making it doesn't have reasonable grounds for it when it's made, whether or not it turns out to be true. So ask what any hours-saved claim is based on.
Overseas regulators have started acting on inflated AI claims. In March 2024 the US Securities and Exchange Commission charged two investment advisers with making false and misleading statements about their use of AI. The penalties totalled $400,000. In September 2024 the US Federal Trade Commission announced "Operation AI Comply", a set of cases against businesses using AI claims to mislead people.
Green flags
- A sample deliverable you can read before you pay.
- Fixed prices, shown + GST, with what's out of scope written down.
- They want to fix the process before automating it.
- They recommend tools from more than one vendor, and say when the answer is a $30 app or a setting you already have.
- They talk about who maintains it after handover, and what that costs.
- They raise data and privacy before you do.
Ten questions to ask any AI consultant
- Can I see a sample of the final deliverable?
- What will you need from my team, and for how long?
- What would you change or remove before automating anything?
- What is the fixed price, and what's out of scope?
- Which tools would you use, and what do they cost per month?
- Do you get commission or partner benefits from any of them?
- Where will our data be stored, and does the vendor train on it?
- What decisions will the system make without a person?
- What happens when it gets something wrong after handover?
- What is any time-saving claim based on?
Government funding and consultants
The Government's AI Advisory Pilot co-funds an AI roadmap and implementation for eligible businesses, through the Regional Business Partner Network. Only approved providers can deliver funded work. If a consultant mentions funding, ask whether they are an approved provider for the pilot and confirm it with your Regional Business Partner. For the record: Even Odds is not an approved provider. Our funding page explains the pilot.
Ask us the same questions
Here's how we try to pass our own test. The AI Opportunity Report is $495 + GST, fixed. It ends in a ranked list of opportunities and a build plan with fixed prices, not a deck. We recommend the process fix first and the tool second, we don't take commission on tools, and every report notes where each recommended tool keeps your data. If you book an AI Implementation within 60 days, the report fee is credited in full. Hold us to all of that.
Sources
Checked on 8 October 2026. Laws and guidance change, so check the source before relying on it.
- Michael Hammer, "Reengineering Work: Don't Automate, Obliterate", Harvard Business Review, July 1990
- W.Media: coverage of Gartner's 25 June 2025 prediction on agentic AI projects and "agent washing"
- Commerce Commission: explanation of unsubstantiated representations under s 12A of the Fair Trading Act (2023 media release)
- US SEC: charges against two investment advisers for misleading statements about AI (18 March 2024)
- US FTC: Operation AI Comply (25 September 2024)
- business.govt.nz: AI Advisory Pilot Programme