Most founders I speak to already know the questions to ask before buying AI, at least the hard way, because they bought it once already. A chatbot that never got installed properly. A tool subscription nobody opened after week three. A build that worked in the demo and then quietly died in production. They are not against AI. They are against wasting another five figures finding out it did not stick.
The real questions are almost never technical. They are questions about your own business, and questions that make a vendor either sharpen up or start hedging; both outcomes are useful, because either way, you learn something before you sign the contract.
Below are the twelve I would ask if I were sitting on your side of the table, along with what a good answer sounds like and what a bad one is usually hiding.
Key Takeaways
- The best questions to ask before buying AI are about your process, not the technology, because process gaps are what kill most implementations.
- Ask what specific task the AI removes and how many hours a week it takes back. If nobody can answer in a number, the scope is not defined.
- MIT research found roughly 95% of AI pilots return nothing, and the failures share one trait: they started with a tool instead of a process.
- Buying specialists roughly doubles the odds a project reaches production compared with building it internally with no prior experience.
- Always ask who owns the accounts, the data, and the automations if the relationship ends. Vague answers here cost the most later.
- Development is about 30% of the work. Adoption, training and optimisation are the other 70%, so ask what happens after go-live.
- Start with one recurring task that already has volume. Small proven wins beat a full transformation plan that never gets past slide six.
Why Most AI Purchases Fail Before the Demo Ends
There is a well-cited MIT study showing that around 95% of enterprise AI pilots produce no measurable return. That number gets waved around as proof AI is overhyped. It is not. What it actually shows is a buying problem.
The 5% that work start with a process. The 95% that fail start with a tool. Somebody sees a demo, gets excited, buys the thing, and then goes looking for a problem inside their business that matches it. That is backwards, and it is expensive.
The same research showed something else worth knowing before you sign anything. Companies that bought from specialists got to production roughly twice as often as companies that tried to build it themselves with no prior experience. Not because internal teams are worse, but because they are doing it for the first time while also running the business.
So the first thing to sort out is not which vendor. It is whether you have defined the problem tightly enough that anyone could quote it. If you cannot describe the task in one sentence, including how often it happens and who currently does it, you are not ready to buy. You are ready to audit.
I have watched founders sign a proposal for something they could not describe two weeks later. The vendor was not dishonest. The scope was just never real in the first place.
Questions to Ask Before Buying AI For Your Own Business
Answer these four before you take a single sales call. They cost nothing, and they change what you buy.
1. What exact task am I trying to remove?
Not “improve efficiency”. Not “use AI better”. A named, recurring task. Answering inbound calls after 5 pm. Chasing quotes that have gone quiet for seven days. Re-engaging contacts in the database who have not been touched in six months.
If you have three or four candidates, that is fine. Write them all down. Ranking comes next.
2. How many hours a week does that task consume, and who does it?
Put a number on it. Ask the person who actually does the work, not your best guess. Founders are consistently wrong here, usually low. This number is the only honest way to size what a fix is worth to you.
3. What breaks today when that task is done badly?
Missed calls turn into booked appointments elsewhere. Slow quote follow-up turns into a competitor’s job. Harvard Business Review’s research on lead response time found firms that respond within an hour are around seven times more likely to have a meaningful conversation than those that wait even two. Speed is not a nice-to-have. It is the whole margin on some deals.
4. Am I willing to change how we work, or do I want the AI to fit our current mess?
This is the honest one. Automation on top of a broken process just makes the break happen faster. If the answer is that nothing about the current process can change, say that up front. A good partner will tell you what that constraint costs.
Questions to Ask Before Buying AI From a Vendor
Now the sales call. These eight separate people who build things from people who sell things.
5. Can you show me this running in a real business, not a demo environment?
Demos are staged. Live systems have edge cases, failures and messy data. Ask to see something with real names blurred out, real call transcripts, real logs. Anyone doing this work properly has that on hand within a minute.
6. What does this look like in month six, not week one?
Most AI purchases perform fine in week one. The failure mode is quiet decay. Nobody updates it, the process shifts, the outputs drift, and eight weeks later the team has gone back to the spreadsheet. Ask what keeps it alive.
7. Who owns the accounts, the data, and the automations if we part ways?
Ask this early and watch the reaction. You want your CRM in your name, your phone numbers portable, your data exportable, your automations documented. A partner should be comfortable saying you own all of it. If ownership is fuzzy, you are not buying a system. You are renting a hostage. Our AI automation contract checklist covers the clauses worth arguing over.
8. How much of my team’s time does this need, and when?
Every build needs your people. Access to systems, review of drafts, sign-off on how the AI speaks to customers. A vendor who says “none, we handle everything” has either not built this before or is planning to guess at your business. Ask for the hours, in writing, by week.
9. What happens when it gets something wrong?
Not if. When. AI systems make mistakes at the edges, and the question is what catches them. Human review on anything customer-facing at the start. Escalation rules. Alerts when confidence is low. If the answer is that it will not get things wrong, end the call.
10. What is the smallest version of this we could run first?
The best answer to almost every AI question is “start with one thing”. One task, live in weeks, measured properly. One of my clients started with nothing more than answering the phone reliably. Dr Claire’s practice was missing a large share of inbound calls despite two receptionists. Fixing that one gap lifted booked appointments by 44%. That is not a transformation programme. That is one task, done properly, compounding.
11. What does ongoing support actually include?
There is a version of this work that ends at handover, and a version where someone keeps training, tuning and extending the system as your business changes. They are different products at different prices. Find out which one you are buying. If you want to understand how these engagements are typically structured and costed, AI automation pricing breaks down the models.
12. What would you tell me not to buy?
My favourite question. Anyone who has done this for a while has opinions about what does not work yet, what is overpriced, and what a business your size should skip entirely. A vendor with no “no” in them is selling a catalogue, not advice.
The Answers That Should Make You Walk
Some responses are worth treating as a hard stop, no matter how good the rest of the pitch was.
Anything that starts with the technology instead of your process. If the first fifteen minutes are about models, platforms and integrations rather than what happens in your business on a Tuesday, they are selling their stack, not solving your problem.
Vague timelines with no milestones. “Six to eight weeks” with nothing to look at until week seven is a project you cannot steer. You want something live and measurable inside the first fortnight, even if it is small.
A price with no scope attached to it. Not because the number is wrong, but because a number without a defined deliverable can absorb any amount of change later. Get the scope written down. Get the exclusions written down too.
Guaranteed outcomes with no conditions. Real guarantees have conditions attached, because outcomes depend on both sides doing their part. Anyone promising a specific revenue number with nothing required of you is either inexperienced or hoping you will not read the terms.
And the big one: no plan for the 70%. Building the system is roughly 30% of the work. Adoption, training, tuning and change management are the rest. Businesses that have been burned before were usually burned right here. They got a working build that nobody used, and the tech was blamed for a process failure.
If you want a structure for checking whether your build is actually working once it is live, the 90-day AI automation checkpoint sets out what to measure.
What Good Looks Like
A good AI purchase is boring on paper. One named task, currently eating a known number of hours, handled by a system that has been proven elsewhere, running live within weeks, measured against a number you agreed before it started. It is not a platform, and it is not a transformation programme; it is one thing that used to need you, now handled without you, followed by the next one.
That sequence matters more than the tools. The businesses I see get real results from AI are not the ones who bought the most impressive thing. They are the ones who bought the right small thing first, saw it work, trusted the process, and then kept going. Two years in, they have a system that knows their business and a set of automations that quietly do the work, and the founder has their week back.
The businesses that stall are the ones who skipped the questions to ask before buying AI, bought the impressive thing instead, and could not describe what it was meant to fix. Now they are paying a monthly fee for something nobody opens.
Where to Start
If you are still working out whether your problem is a tool problem, a process problem, or a capacity problem disguised as a leads problem, that is worth a conversation before you spend anything.
I run a 30-minute Discovery Call where we go through the twelve questions above against your actual business. No slides. If the answer is that you are not ready to buy anything yet, I will tell you that and what to fix first.
Book a 30-minute Discovery Call
If you would rather get your own house in order first, start with AI adoption for SMEs. It covers the readiness work most founders skip.
Frequently Asked Questions
What questions should I ask an AI vendor before signing a contract?
Ask to see the system running in a real business, ask who owns the data and accounts if you part ways, ask how many hours your team needs to contribute and when, and ask what happens when the AI gets something wrong. Then ask what they would tell you not to buy. Vendors with real experience answer that last one easily.
How do I know if my business is ready for AI?
You are ready when you can name a specific recurring task, say how many hours a week it consumes, and name who currently does it. If you cannot, the next step is an audit of where your time actually goes, not a purchase. Readiness is about process clarity, not technical skill or team size.
How much should a small business spend on AI?
It depends entirely on what you are automating and whether you are buying software or a built system. A useful comparison is the cost of the problem: if a task eats ten hours a week of skilled time, price the fix against that, not against a software subscription. Get a scoped quote rather than a ballpark.
What is the biggest mistake businesses make when buying AI?
Buying the tool before defining the problem. Somebody sees a demo, gets excited, and then goes looking for a use case inside their business that fits it. That is the single most common reason implementations quietly die. Start with the task you want gone, then find the thing that removes it.
Do I own the AI system if a vendor builds it for me?
You should own your data, your accounts, your phone numbers and your customer records, and you should be able to export all of it. Documentation of the automations is reasonable to expect too. Some underlying platform licences will stay with the vendor, which is normal. Get the ownership split written into the agreement before work starts.
How long before I see results from an AI implementation?
For a single well-scoped task, expect something live within weeks and measurable results inside the first month. Larger builds take longer, but you should still see a working piece early rather than waiting months for one big reveal. If the plan has nothing to look at before week seven, ask for it to be broken into stages.
Should I buy AI software or have a system built for my business?
Off-the-shelf software works when your process is standard, and the tool covers it. A built system makes sense when the value sits in how your specific business operates, which is true for most founder-led firms with their own way of doing things. The deciding question is whether you would have to change your process to fit the tool.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.