Most founders send the proposal to their accountant, but accountant questions on an AI automation contract should go beyond asking, “Can we afford this?” That is the wrong question, and it is the reason so many of these deals go sideways six months in. The questions that matter are about cash timing, ownership, and what you are left holding if the relationship ends.
I have sat on both sides of this. I sell AI builds, and I have also cleaned up after other people’s contracts. The pattern is consistent. The businesses that got burned did not overpay. They signed something vague and discovered, much later, that they owned none of it.
Here is what to put in front of your accountant before you sign.
Key Takeaways
- Accountant questions on an AI automation contract should start with cash timing, not price: when do you pay, and what triggers the next instalment?
- Ask what happens to the system if the vendor disappears. If the answer is that nothing works without them, you are renting, not owning.
- A build fee plus an ongoing support fee is normal. A build fee that leaves you with nothing you can keep is not.
- Get the data clauses checked: where customer data sits, who can read it, and what the Privacy Act 2020 requires of you as the holder.
- Most AI projects fail on process, not technology. MIT put the failure rate near 95 per cent, and almost none of that is a software problem.
- Ask for the exit terms in writing: export formats, notice period, and who owns the prompts and context files after termination.
- Treat the first invoice as a test of the relationship. Vague scope on paper turns into vague delivery in practice.
Why Your Accountant Should Read the AI Automation Contract First
Your accountant is the only person in your orbit who reads contracts for a living and has no emotional stake in the project. You are excited. The vendor is excited. Your ops manager has already told the team it is coming. Nobody in that group is going to slow down and ask what happens in month nine.
An accountant will, because they have watched clients capitalise something that turned out to be a subscription, and they have watched software spend quietly become the third-biggest line on the P&L without anyone approving it.
There is also a tax and treatment question that most founders never raise. Is this a one-off build that sits on the balance sheet, an operating expense, or a mix of both? The answer changes depending on what you actually receive. If you get files, scripts and configurations that live in your own accounts, that is an asset. If you get access to something the vendor hosts and controls, that is a subscription no matter what the invoice calls it. Your accountant needs to know which one they are recording before the first payment goes out.
The last reason is simpler. An AI automation contract is usually the first technology purchase a founder makes without a clear benchmark. You know roughly what a van costs, what a staff member costs, what a website costs. You have no reference price for an AI build, which means you cannot tell a fair deal from a bad one on instinct. Your accountant cannot price it either, but they can tell you whether the structure of the deal is normal, and structure is where the damage usually lives.
The Money Questions: What Are You Actually Buying?
Start with cash timing, not total cost. Ask the vendor to write down the full payment schedule for the first twelve months, including every fee that is not the headline number.
The questions to work through:
What triggers each payment? A deposit on signing is standard. A second payment on a defined milestone is standard. A payment schedule tied to dates rather than delivered work is a problem, because it pays the vendor for time rather than outcomes. If the milestones are described as “phase two commences”, push back and ask what you will be able to do at the end of phase two that you cannot do today.
What is included in the ongoing fee, and what is billed separately? Nearly every AI build has an ongoing component. That is normal and usually correct, because these systems need training and adjustment as the business changes. What matters is knowing where the line sits. Ask directly: if we want a change in month four, is that included, or is it quoted separately at an hourly rate? Get the hourly rate in writing now, while you have negotiating position, not later when you need the change.
What are the third-party costs on top? Almost every build runs on tools the vendor does not own. Model usage, phone numbers, SMS credits, hosting. These are real costs that land on your card, not theirs. A good vendor will tell you the expected monthly range without being asked. If they cannot give you a range, they have not built enough of these to know, which is information in itself.
What does a return actually look like, in your numbers? This is where your accountant earns their fee. Take the vendor’s claim and convert it into your own maths. If the system is meant to recover dormant leads, ask what your average deal value is and how many recovered deals cover the build. One of my clients, a finance broker, had 319 dormant contacts his team had written off. The reactivation work brought back $49,000. That number only meant something because we knew his average deal value before we started, not after.
If the payback maths only works on the vendor’s assumptions and not on yours, you do not have a deal. You have a hope.
The Ownership Questions: Who Holds the System If You Walk?
This is the section founders skip and the section that costs them the most.
Ask: if we terminate in twelve months, what do we keep? Write the answer down. Then compare it to what the contract says, because those two things are often different.
There are three things worth owning, and a decent contract is specific about all three.
The accounts. The automation platform, the phone service, the model access, the CRM. These should be in your company’s name with your billing details, with the vendor holding access as a user. If everything runs inside the vendor’s accounts, you cannot leave without rebuilding. Some vendors have good reasons for this on shared infrastructure, but they should say so plainly rather than let you assume otherwise.
The configuration. The prompts, the context documents, the workflows, the call scripts, the decision rules. This is the actual intellectual property of an AI build. It is the part that took the thinking. Ask whether you receive copies of it, in a readable format, and whether you are licensed to keep using it after termination. “You get an export” and “you get to keep using it” are two different clauses.
The data. Your customer records, call transcripts, conversation history, and anything the system generated about your customers. Ask for the export format and the notice period. A CSV export on 30 days’ notice is reasonable. No export mechanism at all is a lock-in you did not agree to.
None of this means a vendor holding some things is automatically bad. Plenty of good arrangements run on the vendor’s platform, and that can be the right call for a small business that does not want to administer four accounts. The point is that you should know which arrangement you signed, and your accountant should know before they treat the spend.
The question I would put to any vendor, including my own business: if you disappeared tomorrow, what would still work? A straight answer to that tells you everything about how the deal is built.
The Risk Questions Your Accountant Won’t Think to Ask
Your accountant will cover money and structure well. There are three questions they will probably miss because they sit outside the usual scope, and you should raise them yourself.
Where does customer data go, and who can read it? AI systems process real customer information: names, phone numbers, health details in some industries, financial details in others. Under the Privacy Act 2020, you remain responsible for that information even when a supplier handles it on your behalf. Ask which countries the data is stored in, whether any of it is used to train models, and what happens to it after you leave. Get the answers in the contract, not in an email.
Who is accountable when the AI gets it wrong? It will get something wrong. It will quote a price it should not have, or book an appointment that cannot be honoured, or tell a customer something that is not quite right. That is not a reason to avoid AI; it is a reason to have agreed in advance who fixes it and who bears the cost. Ask what the review and approval steps are before anything goes to a customer, and how errors get corrected. A vendor who claims their system never makes mistakes has either not run one in production or is not being honest with you.
What happens in the 70 per cent of the work that is not building? MIT’s research into enterprise AI found roughly 95 per cent of initiatives deliver no measurable return, and the failures cluster around process and adoption rather than technology. The build is the easy part. The hard part is changing how your team works so the system actually gets used. Ask what the vendor does after go-live. If the answer is a handover document and a support email address, expect to be in the 95 per cent. Good implementation partners stay involved through the part where habits change, which is also the part nobody enjoys.
This is worth reading alongside what to do when an AI automation vendor has already let you down, because the warning signs are usually visible in the paperwork before they show up in the delivery.
What a Good Answer Sounds Like
You are not looking for perfect answers. You are looking for specific ones.
A good vendor answers the ownership question in about fifteen seconds, because they have answered it fifty times. They name the accounts, they tell you whose name each sits in, and they tell you what the export looks like. They do not need to check with anyone.
A good vendor gives you a monthly running cost range for third-party tools without flinching, and the range is narrow enough to be useful. They tell you what is included in support and what is quoted separately, and they put the hourly rate in the document.
A good vendor asks you questions back. What is your average deal value? How many enquiries do you get a week? Who on your team will own this once it is live? If nobody has asked you those, they are selling a product, not solving a problem, and the payback maths is guesswork.
The tell I trust most is how they handle the exit conversation. A vendor who walks you calmly through what happens if you leave is a vendor who expects you to stay. One who gets uncomfortable has built the deal around you not being able to.
Take the answers back to your accountant in writing. Not the pitch deck, not the call notes. The contract, the payment schedule, and the written answers to the ownership and data questions. Fifteen minutes of their time at that point is worth more than an hour of it a year later. If you want a structured way to check whether the build is working once it is live, the 90-day AI automation checkpoint covers what to measure and when.
The Contract Is a Proxy for the Relationship
Everything above comes down to one thing. The contract tells you how the vendor thinks about the relationship before any work has started, which is the only honest signal you get.
A clear payment schedule means they know how long the work takes. Clear ownership terms mean they expect to earn the renewal rather than trap it. Clear data clauses mean they have thought about your risk, not just their delivery. Vague documents come from vague thinking, and vague thinking produces systems that technically exist and practically do nothing.
The real goal is not a good contract. It is a business with an AI brain that holds what you know and an AI workforce that does the repeated work, both of which stay yours as the business grows. That only happens if you own the thing you paid for. Everything your accountant checks is in service of that.
Ask the questions before you sign. They get much more expensive afterwards.
Get a Second Opinion Before You Commit
If you have a proposal in front of you and you are not sure what you are reading, book a 30-minute discovery call. Bring the document. I will tell you what I would ask, what I would push back on, and whether the structure is normal, even if the answer is that the deal you have looks fine and you should sign it.
No pitch required. I would rather you signed a good contract with someone else than a bad one with anyone.
Frequently Asked Questions
What should I ask my accountant before signing an AI automation contract?
Ask three things. First, how should this be treated in the accounts, as a capital item or an operating expense, and what does the answer depend on? Second, does the payment schedule expose us if delivery slips? Third, what is the total twelve-month commitment including ongoing fees and third-party costs? Those three answers will tell you whether the deal is structured normally.
Who owns the AI system after the contract ends?
It depends entirely on what the contract says, and many say nothing. Look for clauses covering the platform accounts, the configuration such as prompts and workflows, and your customer data. Ideally, the accounts sit in your company’s name, and you receive copies of the configuration with a licence to keep using it. If the contract is silent, assume you keep nothing and negotiate before signing.
Is AI automation a capital expense or an operating expense?
That is a call for your accountant based on what you actually receive. If the build produces assets your business owns and controls, it may be treated as capital. If you are paying for access to something the supplier hosts and operates, it usually behaves as an operating expense regardless of how the invoice is worded. Give your accountant the contract, not the proposal, before they decide.
How much should an AI automation project cost a small business?
There is no standard benchmark, which is exactly why the structure matters more than the number. Compare it against something you already price confidently, such as a part-time hire or the cost of the problem you are solving. If the vendor cannot show you payback using your own deal values and enquiry volumes rather than their assumptions, the price question cannot be answered properly yet.
What are the warning signs in an AI automation proposal?
Payment milestones tied to dates instead of deliverables. No named ongoing cost for third-party tools. No mention of data location or export. Deliverables written as activities rather than outcomes. Reluctance to discuss what happens if you terminate. Any one of these is a conversation. Three or more together usually means the vendor has not done this often enough to have answers.
Do I need a lawyer as well as an accountant?
For most small builds, an accountant reading the commercial terms is enough. Bring in a lawyer when the contract involves sensitive customer data such as health or financial records, when the total commitment is large relative to your annual profit, or when the intellectual property clauses are unclear. A short review at that point costs far less than an unwinding later.
What happens if the AI makes a mistake with a customer?
Agree in advance who fixes it and who carries the cost. Good implementations keep a human approval step on anything sensitive, log every interaction so mistakes can be traced, and have a defined correction process. Ask the vendor to walk you through a real example of something going wrong and how it was handled. Their answer tells you whether they have run these systems in production.
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.