Five questions that sort every AI product you will ever be shown — and our honest answers to all twenty-one, including the ones where the answer is no.
Every AI product makes the same four promises. It knows your business. It works on its own. It's transformative. It isn't just a chatbot.
The single-purpose tools promise it. The assistants built into your software promise it. The dashboards promise it. The automation platforms promise it. We promise it too.
Which means the promise tells you nothing. You've heard all four this month, probably twice. The difference between something genuinely new and something repackaged is never visible in what a company claims. It only shows up in the answers.
We publish five questions that sort every AI product you will ever be shown, with an honest scorecard of where each kind lands — The Five Questions. This page is us answering our own instrument. Below are the twenty-one follow-ups behind those five, and how AI Integrator answers each one, including the places where our honest answer is "no" or "not yet."
Questions 1 and 2 sort the market into its real groups. Questions 3, 4 and 5 are the ones almost nothing survives. If you have not run them on anyone else yet, start with The Five Questions — it is the instrument; this is our answer sheet.
Nothing on this page describes how any of it is built. That part stays ours. Everything here is something you can go check.
How your company actually works, written down. Your one goal. Every department and what each is responsible for. Who hands work to whom. The decisions you've made, and why you made them. What each piece of software you already pay for can genuinely do. And what's unfinished right now.
None of that is a saved prompt. It's a written record. It gets built during onboarding in a real interview, and it grows every time you decide something. Three months in, it knows more than it did on day one because things got written down — not because anyone retrained anything.
Three months ago you decided to stop quoting jobs under $5,000 because they never made money. Nobody has to remember to tell it that again, and nobody has to remember to tell the new hire. A $3,200 request comes in and it gets flagged on its own.
The honest edge. It knows what got written down. A decision made in a hallway that nobody recorded is invisible to it. Then again, it's invisible to your company too.
Everything about the job. Nothing about the person.
Whoever sits in that chair next inherits the whole picture on day one — what the job is, what it's supposed to move, what's still open, and what was already decided, including why.
Your operations manager leaves. The new one opens her list on day one and sees the same three things her predecessor was chasing, with the history attached — instead of an empty screen and a shared folder nobody can explain.
What it will not tell your company is who came up with what. That's a wall we built on purpose, and it's the thing people argue with us about most. It's spelled out further down, under Where our answer is no, or not yet.
A record changes. And each fact lives in exactly one place, with everything else pointing at it.
There is no pile of saved prompts to keep in sync, because that is precisely how these setups rot. Two copies of the same fact means one of them is quietly wrong, and nobody can tell which.
So when something changes, it changes in the one place that owns it, and everything pointing at it gets fixed in the same pass. If we rename something, the old name stays registered as a nickname — so anything we missed sends you looking for a second, instead of confidently telling you something false.
You change your standard payment terms from 30 days to 15. That gets changed in one place. Every proposal, reminder and follow-up that mentioned 30 days moves with it. You are not hunting through thirty documents hoping you got them all.
On your side. Three different things, kept apart on purpose:
Think of a builder working on your house. He needs to know the layout, where the pipes run and what you want in the end. He does not need to move into the house, and he does not need your bank statements.
Said precisely, because the loose version would be a lie. We don't keep your information. But when you ask for something, the relevant part of it does travel to us and to an AI in order to get the job done. We won't tell you “it never leaves.” That isn't true of anybody, and you should be suspicious of anyone who says it.
A short list for one specific person. Not a dashboard.
Every piece of work has a job, and every job points at your one goal. If nothing is in the way, it just runs and you never hear about it. If something is in the way, it shows up on a named person's morning list — the actual stuck thing, not a red number on a screen somebody has to go look at.
No. That's the entire design.
The normal way round: a person has to remember the AI exists, go open it, type out what they want, then check whether the answer is any good. That's a person doing the work of operating an AI.
Here it runs the other direction. The work is already about ninety percent done from what the system already knows, and the person's job is to look at it and say yes.
If your people have to remember to go use it, then your people are the connection — and that connection breaks in the first genuinely busy week. That is how almost every stalled AI project actually died. Not because the technology failed. Because somebody got busy.
Think of the last piece of software your company bought that only worked if somebody logged in and checked it every morning. How long did that last?
Both, in different places. Routine work follows rules, because using judgment where a rule would do just makes things unpredictable and expensive. Deciding what's worth interrupting a human being for — that's judgment, measured against your one goal.
A rule: send the invoice on the first of the month. Always, no thinking required. A judgment: this customer said they were ready to sign, then went quiet for three weeks — and that matters more this morning than the other four things on your list.
The honest edge. That judgment is only as good as how clearly your goal is written down. Which is exactly why the audit happens first, and why we won't let you skip it.
The prepared work, yes. But nothing would have been sent, filed or changed — and that's on purpose.
A person signing off isn't a formality we'd drop if we could. It's where a human being stays responsible, and it's what makes the next section's answer true.
This is the question almost nobody asks, and it's the one that quietly kills projects. An AI that's wrong doesn't look wrong. It looks exactly like an AI that's right.
There are two checks, and the AI does neither of them. The AI only ever suggests. Our software decides.
The first check. The AI drafts a price-increase letter. Right customer, right tone, right format — a perfect letter. But that customer is on a contract that locks their price until March. The letter is flawless and it is not allowed. It never sends, because “locked until March” is written down as a rule.
The second check. We tell the system to update a renewal date. It reports back that it did. We go look at the actual record — the date never changed, because that field was locked. We caught it because we read it back instead of believing “done.”
The limit, said out loud instead of buried. Check one only works where there's a real rule to check against — a policy, a limit, a number that has to be true. If it's a pure judgment call with nothing to measure it against, we can't catch it. Anyone who tells you they catch everything is selling you something.
Something turns red. Nothing is called finished because somebody says it's finished. It's finished when there's something you can go look at: a test that passes, a page that's actually live, a file that exists, a document that's signed, an email that genuinely sent.
The rule, in four words: saying so isn't proof.
It is the difference between a contractor telling you the inspection passed and a contractor handing you the signed inspection certificate. One of those you can act on.
Yes, and refusing is required rather than encouraged. When it can't find the rule that governs something, it has to stop and say so — never guess and fill in the blank. When none of our expert viewpoints fit a question, the answer is “none of these fit,” not an invented one.
It's also how an AI earns its way in here. Before we trust a client's chosen AI with real work, we hand it a job with a deliberate trap in it and watch what it does. Does it admit the thing it needs isn't there? Or does it make something up? Make something up and it doesn't get used.
Someone asks what your refund policy is for a customer past 90 days. If your company has never actually decided that, it says “that isn't decided yet” — instead of inventing a policy that sounds perfectly reasonable and isn't yours. The second one is how companies find out they promised something they never agreed to.
Yes, and that's the standard, not extra credit. For every safety check we build, we deliberately build the version of the problem designed to slip past it. If it slips past, the check was decoration, and the work stays broken until it isn't.
We build the check that says a deal can't be marked won without a signed document. Then we deliberately try to mark one won with no signature, to see whether the check actually stops us. If it doesn't, it was never a check — it was a sticker on the box.
A check nobody has attacked isn't a check. It's a comment.
All of it. Switching is changing a plug, not starting over.
There are three layers, and only the middle one swaps out:
Each AI brand gets a small plug that fits the same socket. Change the AI and nothing else in the system moves. If you want to run this on a different provider, or on something your own team built, that's a plug — not a project.
The day a better AI comes out, we point it at the same jobs and carry on. Your goal, your departments, your rules, your history — untouched. Compare that with a setup where somebody spent six months hand-tuning instructions for one particular AI. That six months is what gets thrown away.
Picture a chef who hires a temp cook for one shift. One ticket at a time. Never the recipe book, never the keys to the safe, and every plate gets tasted before it leaves the kitchen. Different temp tomorrow, same kitchen, same food.
The written description of how your company works — in your hands, in a form any AI can read. Because it was never on our side to begin with.
There is nothing of yours for us to hold back, because we never took it. That's how the thing is built, not a promise we're making you.
It is the difference between losing your bookkeeper and losing your books. You lose the bookkeeper. The books are yours, they stay on your shelf, and the next person can read them.
What you don't keep, plainly. The method — how we turn that description into finished work. That was always ours. It's the thing the retainer paid for, and it's the only thing that leaves with us.
Yours. Same three-way split as before: your real business data never comes to us at all, the description of how your company works lives in your files, and the method is ours.
The safest information is the information nobody ever took. That's a design decision here, not a privacy page.
You'd lose us. The method, the discipline, and the improvements that would have come from the other companies we work with.
You would not lose your record, your description of your own business, or your ability to run all of it on a different AI. That's a real loss. It's just a bounded one, and you can see exactly where the edges are before you sign anything.
Yes. It's already built that way. Being able to leave isn't a concession we'd make under pressure. It's a rule we designed around from the start.
Now go ask a big AI platform the same question, and listen to how carefully the answer is worded. A company brain you can carry to any AI is the one feature none of them will ever build — because your way out is their lost customer.
That isn't us claiming to be clever. It's an incentive you can check for yourself.
The AI getting better makes this worth more, not less.
What you're buying is your own business, written down so an AI can act on it. No AI company ships that, because nobody outside your company can write it.
As the AI gets smarter, the thing holding you back changes. It stops being “is it smart enough” and becomes “can it read this business at all.” That's the whole bet, stated plainly so you can disagree with it.
An AI can pass the bar exam today. It still can't tell you which of your customers is about to leave — because nobody at your company has ever written down what “about to leave” actually looks like here.
Both. And where the line sits is written down, not promised.
Your side gets better because you use it. Every decision, correction and ruling lands in your record.
The framework gets better across every company we work with — but only as a lesson, never as anyone's information. Before anything can travel, it has to be stripped down to the lesson itself, with every detail that could identify a company removed. If it can't be said without naming someone's specifics, it doesn't travel. Not summarized. Not combined with others. It simply doesn't travel.
And when a better way reaches you, it arrives as a suggestion on your own list. Never quietly switched on inside your business.
A master craftsman trains a lot of apprentices, and gets better at teaching with every one. But apprentice A's actual project never shows up on apprentice B's bench. B just gets the sharper lesson, and builds their own thing, with their own hands.
The contract says what the design says. Your information stays yours. We get better at the method by doing the work. You benefit from everyone's improvements — and nobody benefits from your information.
Whoever's business is actually written down clearly. That's work no vendor can do for you, and it isn't something a competitor can copy off you.
They'd all get the same framework and get different results — the same reason two companies running identical accounting software don't perform the same.
Two contractors buy the identical estimating software on the same day. A year later one is winning noticeably more work. The software wasn't the difference, and everyone in that industry already knows it.
The uncomfortable version, because we'd rather say it now than in month four: if a company won't do the work of getting clear about itself, this doesn't save them.
The person whose morning list goes quiet. That same day.
Which is exactly why problems go to a person instead of a dashboard. A dashboard nobody opens can be broken for three months and look perfectly healthy the entire time.
A more readable company.
The AI models are already good. They get cheaper every month. Next year's beats anything you'd lock yourself into today. The smart AI is the cheap part.
Hire the single smartest person on earth. Give them no access to your files, no org chart, and no idea what you're trying to achieve this year. Now watch them be worth almost nothing to you by Friday.
A genius who can't read your business is useless to you. An average one who can is worth everything.
Collected in one place so they can't be mistaken for an oversight. Each one is a deliberate choice, each one has a reason, and each one has cost us something.
Some owners want that view. They don't get it. It isn't a setting anyone can switch on — it's built into how the information is stored, so a mistake or a permission change can't open it.
It exists because people are honest with a suggestion box the boss can't open, and stop being honest the moment they can.
Someone on your team flags that a process is broken because the training was never finished. That lesson reaches you. Who said it does not. Flip that around and watch how fast the flags stop coming.
We do keep track of who — permanently, and only we can see it — because you can't turn one person's good idea into a general rule without knowing the situation it came from. The promise holds because your company never had access. Not because we threw the record away. Throwing it away would make us worse at our job without making a single person safer.
Repeated here because it's the most important limit on this page. If something is a pure judgment call — no rule, no limit, no number that has to be true — we can't catch it going wrong. That's a real gap, and we'd rather name it than round it up.
“Don't discount below 20%” is a rule — we can catch a breach of it every time. “This email is a bit too pushy for this particular customer” is a judgment. A human still has to make that call, and we'd rather say so than imply we've got it covered.
There's a real interview, and your company has to answer honestly — including about the difference between how you say you do things and how they actually get done. A single-purpose AI tool installs in an afternoon. This doesn't.
And the work doesn't end, because staying pointed at your goal isn't something you complete. It's something you keep.
When the plan needs something built, we help you find the right people for it. We don't manage them, and we won't pretend the line isn't there.
If you cancelled every AI vendor you use on Monday — what would your company still know on Tuesday? If the answer is “nothing,” you didn't buy AI. You rented it.
That's the reason it's in writing. The diagnostic is free, and we'll tell you honestly whether this fits your company or whether one of the other paths is the better call.
Get a free diagnostic