Almost nobody in document work owns the documents they handle.
A bureau processes a client’s boxes. A shared service captures another division’s invoices. A records team looks after a practice’s files. The person doing the work and the person it belongs to are two different people, and the relationship lives in that gap.
Two habits close it, not a feature on a datasheet. The work gets checked before it leaves. The owner can see where it stands without asking.
Ask a prospect what they want and you will hear accuracy, speed, integration, price. Nobody has ever asked me for quality control. What they are buying is not being embarrassed in front of a client, and that is a different thing to sell.
Take the checking first, because it cost us the longer argument.
Software reads a document, pulls out the fields, and checks them against rules, faster than any team of people could. What it cannot do is take responsibility. A rule can confirm a date is a date. It cannot tell you whether this set of documents matches what someone expects. So the last step before work leaves us is a person, looking at the whole set together, who either releases it or sends it back.
Inside the business, the argument was whether that step admits failure. If the extraction is good, why inspect it at all? I would defend the answer we settled on anywhere. Automation earns trust by being checkable, not by being confident. A system that reports its own certainty is asking to be believed. A system that stops and puts the work in front of a person is offering to be checked. Only one of those survives an audit.
That is not a closed question, and the strongest argument against us is the best thing I read all month.
W. Edwards Deming’s third point for management says to stop depending on inspection for quality, and build quality into the work from the start instead. He is right. Inspection is late by definition. By the time somebody looks, the quality is already there or it is not, and you have paid twice: once to make the error, once to find it.
What I would add is what the gate earns while it stands there. Every mark an inspector puts on a field tells you where quality leaked upstream, and in which document type. A gate that teaches you nothing is a tax. A gate that shows you which step keeps producing the same problem is the cheapest research you will ever run, and the plan is to need it less each year.
The second habit is the same idea, pointed the other way.
A client’s most common contact with a bureau is not a complaint. It is a question about where their work is. Each of those calls costs both sides half an hour and chips away at the relationship. The answer was sitting in the system the whole time. Nobody sent it.
So one page now goes out every week, without being asked, and it answers five things in order. Are we on track? Where is your work? What is stuck? What went wrong? What do we need from you?
The fifth one turned out to be the whole point. A report that only reports status is a scoreboard, and people stop reading scoreboards by about week four. A report that names what is waiting on the client is a conversation. It moves work that had been quietly parked because nobody wanted to make the call. It puts the client’s own delay on record while it is happening, and that is the only moment that is fair to anyone. Raising it at month end, once a deadline has gone, is not evidence. It is an argument.
There is a third thing worth saying. Running a national, multi-region operation on our platform for years teaches you what a pilot never can. Which failures only show up at volume. Which processes quietly depend on one person’s memory. How much of what a client values sits in the running of the thing, not the building of it. Most of what I believe here came out of work like that. A relationship is judged on how it runs, day after unremarkable day, not on how it was won.
Which is what I would say to anyone holding documents that are not theirs. Being trusted is a practice, not a claim. It is made of small repeated acts that show from the outside: the check that happens every time, not only when somebody is watching, and the page that arrives before it is asked for. You cannot put that in a proposal. You can only let people watch you do it.
Show your work.
Fifteen minutes on our sample set, or bring your own worst documents and watch them become data.