This morning I published a post called "AI for bookkeeping: what it can reconcile, and what still needs a human". I wrote it in that order, and when I finished it the second half was longer than the first. The section about what the software does well ran to five bullet points. The sections about where it stops, what it cannot see, and what a person still has to do ran to most of the page.
That was not a decision I made at the top. It is just what came out. And it made me go and look at the rest of what I have published, because I had a suspicion about it.
What I found in my own archive
There were fifty-nine blog URLs in the sitemap when I checked this morning. Reading down the list, I can name five whose main job is to tell you that something will not work, or to slow you down before you spend:
- one on why AI projects fail
- one that is a checklist for interrogating an AI vendor
- one on the documents you are not allowed to sign electronically
- one that takes four widely quoted statistics about lead response times and says plainly that one of them cannot be traced to a source
- one that tells people who have not started yet that they are not behind
That is my own classification of my own writing, so take it as a reading rather than a metric. But the pattern is not subtle. Roughly one post in ten exists to subtract a claim. Today's makes six.
On the personal side it is the same. There is a piece on why I will not put a price in a blog post. There is one about the day I told somebody not to hire me. I am apparently running a marketing operation with a meaningful slice of its output dedicated to talking people out of things.
Why I think it keeps happening
The honest answer is that I am not writing for a market that is deciding between two products. I am writing for one that has been sold software before and is still slightly annoyed about it.
Almost every small business owner I speak to has a system somewhere that was going to change everything and now costs money every month and gets opened on Tuesdays. They are not sceptical about technology in the abstract. They are specifically sceptical about the gap between what a thing was described as and what it turned out to be. That is a much narrower and much more reasonable form of distrust, and it does not respond to enthusiasm at all.
What it responds to, as far as I can tell, is somebody drawing the boundary before they have to find it themselves. If I tell you up front that the reconciliation will match your bank lines and will not tell you whether the invoice behind them is any good, you now have one fewer unpleasant surprise available to you. The claim I did not make is the thing that survives contact.
The part that is not noble
I want to be careful here, because there is a version of this post that is just me congratulating myself for honesty, and that version would be wrong in two specific ways.
The first is that restraint is easier when you have less to lose. I am not defending a number to anyone this quarter. It costs me almost nothing to leave a claim out. Somebody with a team and a target has a genuinely harder version of this decision than I do, and I should not pretend otherwise.
The second is that I cannot prove it works. I have no way to separate the readers who trusted the writing from the readers who would have come anyway, and the honest position is that I am running this on a conviction, not on evidence. The conviction is that overclaiming is a debt you settle later at a worse rate. I believe it. I have not demonstrated it.
There is also a real cost, and it is not abstract. Writing this way makes the material slower to sell from. A post that spends most of its length on limitations does not hand a reader an obvious next action. It hands them a more accurate picture and then asks them to do something with it. Some of them will do nothing, and I will never know which ones.
What I think the actual rule is
Not "be honest", which is too broad to act on and slightly self-satisfied besides. Something narrower:
Do not describe a capability without describing its boundary in the same breath. Not in a footnote, not in the caveats page, not in the call. In the same paragraph, at the same volume.
That is testable, at least on my own work. If a claim appears and the reader has to go somewhere else to find out where it stops, I have written it badly. It is also the only version of this I can apply consistently, because it does not depend on how generous I am feeling on a given morning.
The bet underneath it is that a market which does not trust software yet will not be persuaded by anyone insisting it should. It will be persuaded, slowly and one person at a time, by things turning out to be roughly what they were described as. That is a much less exciting theory of marketing than the one I started with. It is the only one I have seen actually work.
Frequently Asked Questions
Is writing about limitations not just bad marketing? It costs you the reader who wanted permission to buy immediately. It keeps the reader who has been burned before and is reading specifically to find out where the claim breaks. In a market with a lot of the second kind, that trade looks reasonable to me.
How do you decide what to leave out of a claim? The test I use is whether the boundary sits in the same paragraph as the capability. If a reader has to go elsewhere to find out where something stops working, the claim is incomplete, however true the words are.
Does this mean you never say anything positive about software? No. Today's post has five things AI genuinely does well in bookkeeping and says so without hedging. The point is that the limits go next to them rather than somewhere quieter.
Can you show that this approach brings in more work? No. I cannot separate readers who trusted the writing from readers who would have arrived anyway. I am running it on the belief that overclaiming is a debt settled later at a worse rate, and I would rather say that than dress a conviction up as evidence.
Related: AI for bookkeeping: what it can reconcile, and what still needs a human, why I will not put a price in a blog post, and nobody told me I would become my own bookkeeper.