Can the software do the quote and a human just check it?

The way he put it was better than any way I have put it. At the moment, he said, our estimators do the whole estimation process, and then they stick it under the nose of someone else to go, does that look right? What he wanted to know was whether AI had got good enough to do that first part, the long part, and leave the sniff test with the human. The answer is yes. But the reason it works in his business and falls over in others has very little to do with how clever the software is.

The check has to already exist

His process already has a second pair of eyes in it. Somebody senior already looks at a finished quote and decides whether it smells right. He described a job the other week where an estimator had missed a shield, just had not quite looked down at the bottom of the drawing, and the second reader caught it. People are human, he said, you will always miss something.

So basically, when we put software into that process, nobody is being asked to trust a machine with anything. The judgement stays exactly where it already sat. All that changes is who spends the hour doing the long bit in the middle. That is a small change to how the business runs, which is why it tends to stick.

Now take the same request in a business where nobody checks the quote before it goes out. No second reader, no sniff test, one person prices it and sends it. Put software in there and you have nothing to catch it when it is wrong, and no agreed idea of what right even looks like. That is not an AI problem. The process was already running without a brake.

What does that look right actually means

If a human checker is going to stay in the loop, they need something to check against, and it usually turns out nobody has written that down. So before anything gets built, we ask for about ten jobs that have already been priced and already gone out of the door, including the awkward ones.

He offered that himself, which told me he has thought about this properly. He said he would not pick terrible drawings, but he would pick ones he knows have pitfalls in them, because a human would stumble over those too. That is exactly the right instinct. The bar is not perfection. The bar is your current process, on a bad day, with a tired estimator.

Then the test is simple. Run the ten through the system. For each one, would the checker have accepted the answer and sent it, or sent it back? That gives you a first time right figure for the software, next to the first time right figure you already get from people. Two numbers, same scale, no argument about them.

What this means if you are weighing it up

Look at the work you want to hand over and ask who checks the output today. If the answer is a named person who already does it as part of their week, that job is a good candidate and you can start on it now. If the answer is nobody, you are not blocked on AI. You are missing a step that your business needed anyway, and adding software will just make the mistakes happen faster.

The quiet part is that most of the businesses I speak to already have the check. They have lived through enough expensive mistakes to have invented it. They just never thought of it as the thing that makes automation safe.

If a second person already looks at the work before it leaves the building, you are ready. If nobody does, start there instead.