That's the short version. The longer one is about what changes when the AI reads what's on file, and what a person still does.
What goes wrong
Ask an assistant which of your customers are overdue. Then ask what it read to answer.
The first question to ask about any assistant: what is it reading? An assistant that guesses instead of reading what's on file is not one to trust with your records.
What we do about it
In our tools, the AI reads what is on file. It reports what is on file. It never invents a number to make things look better.
The assistant we build says so when it doesn't know.
When the AI answers from your data, the answer is checked against your records, word for word, before you see it. Our trust page says the same, and says what the AI may and may not do.
The same rule is in how we build everything: the AI reports what is on file, and it never invents a number to make things look better.
That is a narrow claim. It covers our tools, not every assistant you might use. We say what ours is built to do, and where a person steps in.
An example: RevWin
RevWin is capture planning software for architecture, engineering and construction firms. It turns Don Sherman's method into a tool. RevWin is live, and we own it and run it. It is the capture method, not the system of record.
RevWin gives a firm go/no-go discipline, intelligence on the decision-makers, relationship scoring, and a capture plan generated from the firm's own pursuit data. It doubles what one business-development pro or seller-doer can carry. It reports what the firm has logged. It never invents a relationship or a score to make a pursuit look better than it is.
Picture a firm that has logged no contact at a client it wants to win. RevWin does not fill that gap with a relationship nobody logged. RevWin keeps its own brand, and Inertia is the maker.
Who approves
A person approves anything that bills, sends, or signs. The AI gets the work ready, and a person decides whether it goes out. Our trust page lists what the AI may and may not do, and what a person approves.
The ChatGPT or Claude you already use can draft into the tool. A person still approves.
That is the shape of the assistant we build: it reads what's on file, says so when it doesn't know, and waits for a person to approve before anything goes out.
What to ask any vendor
You can put the same questions to any tool that offers you an AI assistant. Does it read your records, or does it answer from nowhere in particular? Does it say so when the answer isn't there? Who approves before something goes out to a customer?
Those are the same things we say about ours. If a vendor can't answer them, you know where you stand.
When you don't need us
If your assistant lets you attach a file, you can ask it about the document directly. You don't need us for that, and we'll tell you so.
We find the real problem before anything is built, and if what you asked for won't help, we say so.
Where we come in is when you want an assistant that reads your records instead of guessing. Software that just runs is what that looks like: a job that runs on a schedule with no one running it, such as monthly billing reports, finding leads, or pulling in data from other apps.