Most founders should learn how to spot the red flags AI automation company candidates often hide during the initial sales pitch to avoid being burned. They paid someone to build an AI tool, the demo looked sharp, the invoice cleared, and six months later nobody in the business uses it. The real warning signs are almost never in the technical build; they are in the conversation, and they are visible if you know what you are listening for.
This post is the list I would give a mate before he signed anything. What to watch for in the pitch, in the proposal, in the build, and in the months after go-live. Plus the questions that separate an operator who will still be answering your messages next year from one who is optimising for the deposit.
Key Takeaways
- The clearest red flags an AI automation company shows appear in the sales process, not the build: no discovery questions, constant tool talk, and a quote inside the first call.
- If they price the work before mapping a single one of your actual processes, they are selling a product they already built, not solving your problem.
- Ask who owns the accounts, the data, and the automations after handover. A vague answer means you are renting your own operations back off them.
- Around 95% of business AI projects return nothing measurable, and nearly all of them started with the technology instead of the process.
- Build work is roughly 30% of the engagement. Training, adoption, and iteration are the other 70%, and that is where most projects quietly die.
- A good operator will tell you what not to automate. Anyone who says yes to every idea has not thought about your business for more than five minutes.
- Ask for one client result with a real number and a real name attached. “Saved hours every week” is a slogan, not proof.
- No documentation, no handover session, and no named support contact means the system breaks the first time your process changes.
The Sales Process Tells You Almost Everything
By the end of the first call, you should know whether this is going to work. Not because of what they showed you, but because of what they asked.
Red flag: they talked about tools before they asked about your business. If the first fifteen minutes cover which platform they build on, which models they use, and how many integrations they support, you are in a product demo. The name of the automation platform is the least interesting thing about the engagement. What matters is whether they understand where your time actually goes and what breaks when you step away. I have sat on calls where the prospect never once got asked what their week looks like.
Red flag: no discovery questions about process. Automation is a process problem with a technology answer, in that order. Anyone serious will want to know how a lead actually moves through your business, who touches it, where it stalls, what happens when someone is on leave. If nobody asks you to walk through a single real workflow start to finish, they cannot possibly know what to build.
Red flag: a price in the first conversation. A quote before scoping means one of two things. Either they are selling a fixed package that exists already, and you are being fitted to it, or they are guessing and will discover the real scope after you have signed. Neither ends well. Rough ranges are fine and honest. A firm number attached to a specific build, before anyone has looked at how you operate, is a warning.
Red flag: urgency that comes from their calendar, not your problem. “We only have two build slots left this month” is a closing technique, not a fact about your business. Real urgency sounds different. It sounds like “you are losing roughly forty enquiries a month to slow follow-up, so every month you wait costs you about this much.” That is urgency you can verify.
Red flag: they say yes to everything. Ask them what they would not automate in your business. If the answer is that everything is automatable, they have not thought about it. Judgement calls, high-stakes client conversations, and anything with a genuine legal or safety consequence belong to humans, at least for now. An operator who draws that line in front of you is telling you they have done this before and watched something go wrong.
Red Flags in the Proposal and the Pricing
The proposal is where good intentions meet reality. Read it like a contract, because that is what it becomes.
Red flag: deliverables described as outcomes with no mechanics. “AI-powered lead management” is not a deliverable. “New enquiries from the website form receive an SMS within sixty seconds, a follow-up call attempt within five minutes, and a four-touch email sequence over ten days, with all activity logged against the contact record” is a deliverable. If you cannot picture exactly what will exist at the end, neither can they.
Red flag: no clarity on who owns what. This is the one that costs people the most and gets asked the least. Whose account is the CRM on? Whose phone numbers are those? Who holds the API keys? If the automations live inside the provider’s own workspace and you cannot get an export, you do not own a system. You own a dependency, and the renewal conversation will reflect that. The right answer is that the accounts are in your business name, the data is exportable, and you could take it to another provider tomorrow if you wanted to.
Red flag: pricing that hides the ongoing cost. Automation is not a one-off. Platforms charge monthly, messages and calls cost money per unit, and someone needs to maintain the thing when your process changes. A proposal that shows a big setup number and goes quiet on running costs is setting you up for a surprise. Ask for a twelve-month total, including third-party platform fees and usage. If they cannot produce that, they have not built enough of these. I have written more about how to read automation quotes in our guide to AI automation pricing.
Red flag: guaranteed returns with no conditions attached. Anyone promising a specific revenue lift with no dependency on your lead volume, your close rate, or your team actually using the system is either naive or hoping you do not read closely. Honest guarantees have conditions, because outcomes depend on both sides doing their part. A conditional guarantee is a sign of someone who has been held to one before.
Red flag: no exit. What happens if it does not work? What happens if you want to leave in month four? If the proposal has no answer, ask for one in writing before you sign. The response tells you how confident they actually are.
The Proof They Show You (and the Proof They Avoid)
Every provider has a portfolio page. Almost none of them have proof.
Red flag: results with no name, no number, or no mechanism. “Increased efficiency for a national retailer” tells you nothing. Real proof has three parts: what the situation was, what was built, and what changed numerically. When I talk about database reactivation, the example I use is James, a finance broker with 319 dormant contacts his team had written off. A multi-touch SMS and email reactivation recovered $49,000 from a list he already owned. You can interrogate that. You know the size of the list, the channel, and the outcome. A claim you cannot interrogate is marketing.
Red flag: case studies borrowed from someone else’s business. This happens more than you would think in the AI space. A provider quotes numbers from a programme they bought, a workshop they attended, or a founder they follow online, and presents it as their own delivery record. Ask directly: did you personally build this, for a client who paid you? The pause tells you what you need to know.
Red flag: no reference call on offer. If a provider has happy clients, they will have one who will take a ten-minute call. If they resist, or they offer a written testimonial instead, that is information. When I point people at voice AI results, I point at specifics: a dental practice owner, Dr Claire, who was losing close to half her inbound calls to a two-person reception team and lifted booked appointments 44% after the overflow calls started getting answered. If you asked me to put you on the phone with someone in that position, I would.
Red flag: the demo is the only proof. Demos are built to work. They run on clean data, a happy path, and a script the presenter has rehearsed. Ask what happens when the caller has a strong accent, when the CRM field is empty, when the customer replies with something the system was not expecting. A provider who has run these in production will have real answers and real failure stories. One who has not will tell you the AI handles it.
What Happens After the Build Is Where Most Projects Die
MIT’s research into business AI adoption found roughly 95% of initiatives deliver no measurable return. The pattern in the failures is not bad technology. It is that the technology got installed into a business that was never prepared to use it.
Red flag: no adoption plan. Development is maybe 30% of the work. The other 70% is training, changing how people actually operate, and fixing the ten things nobody anticipated in week two. If the proposal has a build timeline and a go-live date but no training sessions, no documentation, and no review points, the project ends the day the invoice is paid. That is when it should be starting.
Red flag: no documentation or handover. Ask what you receive at the end besides a working system. The answer should include a written record of what was built, how it is configured, what the fallback is when something fails, and who to contact. Without that, your business now has a black box that only one external party understands. You have not removed a single point of failure. You have moved it outside the company.
Red flag: nobody is measuring anything. Automation without a baseline is a story you tell yourself. Before anything is built, someone should write down where you are now: current response time, current answer rate, current conversion, current hours spent on the task. Otherwise, in three months you will have a vague sense that things feel better and no way to justify the spend. Research published by Harvard Business Review found firms contacting a lead within an hour were nearly seven times more likely to have a meaningful conversation than those waiting even sixty-one minutes. That is measurable. Pick the numbers that matter to you and record them before you start.
Red flag: support is undefined. “We’re always here if you need us” is not a support arrangement. What is the response time? Is there a monthly review? Who fixes it at 4 pm on a Friday when the booking automation stops firing? If the answer lives in a friendly sentence rather than a written commitment, you will find out how it works at the worst possible moment.
Red flag: they build in isolation from your team. The people who will use the system every day should be in at least one conversation before it is built. Not because they need to approve the technology, but because they know where the process actually bends. Systems built purely from the founder’s description of how things work tend to break on contact with how things really work.
The Deeper Problem Nobody Mentions in the Pitch
Here is the one that costs the most, and it is not really a red flag in the provider. It is a red flag in the brief.
Most automation projects fail because they automate a task inside a business that has no shared intelligence. Your CRM does not know what your accounting system knows. Your project tool does not know what was agreed on the client call. The context that makes any of it useful lives in one place, which is your head. So you bolt an automation onto the side, it does its one job, and it never compounds, because there is nothing for it to plug into.
That is why the second automation is rarely as good as the first, and the fifth is a maintenance liability. Each one is a separate thing to check, with its own login and its own way of breaking. You end up managing your automations the way you used to manage the tasks.
The alternative is to build the brain before you build the workforce. Capture what the business knows, in a form a system can read. Connect the data so the numbers live in one place instead of six dashboards. Then automate, so that every task you hand over runs with the full picture behind it rather than a narrow trigger.
The red flag version of this: a provider who never asks about your context, your data, or how anything connects. They will happily build task number one. They just will not tell you that task number one is worth a fraction of what it could be, or that in eighteen months you will be paying someone to untangle eleven disconnected automations. This is the difference between hiring a builder and hiring a partner, and I have written about that distinction in more detail in what an AI automation agency actually does.
The Questions That Sort Good From Bad in Ten Minutes
You do not need to be technical to run this. Ask these, and listen to how they answer rather than what they answer.
- “Walk me through a client whose project did not go well. What happened?” Everyone has one. A provider who claims otherwise is either new or not being straight with you. The useful answer includes what they changed afterwards.
- “What would you not automate in my business?” Tests whether they have actually thought about your situation or are pattern-matching to a package.
- “After handover, whose name is on the accounts and can I export everything?” The single highest-value question on this list.
- “What does month four look like?” Anyone can describe the build. Fewer can describe the support rhythm, the review cadence, and what happens when you want a change.
- “What do you need from me and my team, and how many hours?” If the answer is “nothing, we handle it all,” they are either overselling or planning to build something your team was never consulted on.
- “What numbers will we look at in ninety days to decide if this worked?” A provider who has a ready answer has done this before. One who improvises is about to sell you something unmeasurable.
- “Which part of this could I do myself, and would you tell me if I could?” The answer to this tells you whether they are optimising for your outcome or their invoice.
A quick word on scale as well. If you are a small operation, some of what gets sold as enterprise AI transformation is genuinely overkill for you, and a good provider will say so. There is a sensible starting point for most businesses under fifty people, and it is usually smaller and more specific than what turns up in the proposal. I have covered that starting point in AI agents for small business.
Conclusion
The warning signs are consistent, and they cluster in three places. In the sale, they talk about tools instead of asking about process, and they price before they scope. In the proposal, ownership is vague, ongoing costs are hidden, and there is no exit. After the build, there is no training, no documentation, no measurement, and no defined support.
Underneath all of it sits the real issue. Automating one task inside a business that has no shared brain gives you one working task and a new thing to maintain. Automating inside a business where the context is captured, the data is connected, and the system understands how you actually operate gives you something that compounds. The first is a purchase. The second is a business that can start running without you in it.
If you have been burned before, that was probably the difference, and it was never really about the technology. You just need to know how to spot the red flags AI automation company vendors display before the contract is even signed.
If you are weighing up a provider right now, or you have a proposal in front of you and something about it does not sit right, book a 30-minute Discovery Call. No pitch. Bring the proposal, and I will tell you honestly what I would push back on, what I would ask for in writing, and whether the scope makes sense for the size of your business. Worst case, you get a second opinion and go ahead with them anyway.
Frequently Asked Questions
How do I know if an AI automation company is legitimate?
Look for evidence rather than credentials. A legitimate provider will name specific clients, quote specific numbers, and offer a reference call. They will ask detailed questions about your processes before proposing anything, and they will be able to describe a project that went badly and what they changed as a result. Polished branding and platform partnerships prove nothing about delivery quality.
What questions should I ask an AI automation company before signing?
Ask who owns the accounts and data after handover, what happens in month four, what they would not automate in your business, what they need from your team in hours, and which numbers you will review at ninety days to judge success. Also ask about a project that failed. The quality of the answers matters more than the content, because hesitation on ownership or support is the real signal.
Should I own the automations and data an AI automation company builds?
Yes. Accounts should be in your business name, your data should be exportable at any time, and you should be able to move the work to another provider without starting over. If the automations live inside the provider’s workspace and you have no access, you are renting your own operations. Get the ownership position confirmed in writing before you pay a deposit.
How much should AI automation cost for a small business?
It varies enormously by scope, but the useful comparison is against what the problem currently costs you. A single automation addressing one clear bottleneck usually sits well below the cost of a part-time hire. Full system builds cost more. What matters more than the setup figure is the twelve-month total including platform fees and usage, which many proposals leave out entirely.
Is it better to hire an AI automation company or build it in-house?
Research consistently shows businesses that buy from specialists reach production roughly twice as often as those building internally. In-house works when you have genuine capacity and someone who will own it long-term. For most founder-led businesses, the bottleneck is attention, not capability, and a DIY build becomes another half-finished project. Hire out the build, but insist on documentation so the knowledge stays with you.
What happens if the AI automation company disappears or I want to leave?
This should be answered in the proposal, not discovered later. You want accounts in your name, exportable data, written documentation of the configuration, and no proprietary layer that only they can access. If a provider cannot describe a clean exit, that tells you how confident they are in retaining you on merit rather than on lock-in.
How long should an AI automation project take before I see results?
A focused single-process automation should be live within a few weeks and showing measurable change within thirty days. Larger builds take longer, but they should still deliver something usable early rather than everything at the end. Be wary of timelines with a single go-live date months away and no interim milestone. Long silent build periods are where projects drift.
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.