Modern bookkeeping software is genuinely good. It connects to your accounts, pulls transactions automatically, learns your categories, and gets a lot right without being asked.
So the question is fair: if the software does the bookkeeping, what is the bookkeeper for?
The honest answer is that software is very good at recording and has almost no ability to verify. Those are different jobs, and the second one is where the money is.
Software records. It does not check.
A bank feed tells you what it received. It cannot tell you what it missed.
When a connection breaks for nine days and resumes without backfilling, the software shows you a clean, plausible, incomplete ledger. Nothing is flagged, because from the software’s point of view nothing went wrong. The only thing that catches it is reconciling against the actual statement, which is a deliberate act somebody has to perform.
Same with duplicates after a reconnect. Each copy looks entirely legitimate on its own. The software has no opinion about whether you really bought the same thing twice. A feed is a copy of the bank’s record, not the record itself, and checking a copy against itself proves nothing.
Software categorizes. It does not know your business.
Auto-categorization is pattern matching against what you did before, and it is genuinely useful. But it cannot distinguish things that look identical and are not:
- A hardware store charge that is job materials versus one that is a repair to your own premises. Same merchant, different treatment.
- A transfer to your other account, which is neither income nor expense, versus a payment to a supplier.
- A payment that is partly business and partly personal.
- An owner draw, which is not an expense at all and never appears on your Profit and Loss.
The software will confidently place all of those somewhere. Being wrong consistently is not better than being wrong randomly; it just makes the error harder to see.
Software answers. It does not ask.
This is the difference that actually shows up in money.
A person doing your books notices that a subscription you forgot about has been renewing for eleven months. That a supplier’s pricing crept up 30 percent since spring. That your gross margin has slipped three months running. That the customer you talk about most is not the one paying you best.
None of those are questions the software was asked, so none of them get answered. They surface because somebody was looking at your numbers as your business rather than as data to be sorted.
That trickle of small questions is what accurate books actually sound like. When it goes quiet, it usually does not mean the transactions became obvious. It means they are being guessed at.
Software will not chase you
The unglamorous one, and probably the most valuable.
Software does not care whether you ever look at it. It will keep importing transactions into an increasingly wrong ledger for a year without complaint, and the reports will keep generating, and they will keep looking normal.
Most catch-up work I take on is not from people who chose badly. It is from people who bought good software, used it well for two months, got busy, and had nothing in their life creating the deadline. A monthly close that somebody else is responsible for is a forcing function, and forcing functions are what actually keep books current.
What the division of labour really is
Software is genuinely better at: importing transactions, applying consistent rules at volume, arithmetic, generating reports on demand, storing documents, never getting bored.
A person is required for: proving the data is complete, resolving the transactions only you can explain, keeping personal and business properly separated, distinguishing draws from expenses and transfers from income, noticing patterns nobody asked about, and creating the deadline that makes the month actually close.
Nobody sensible does this without software. The tool does the volume; the judgment and the verification are what you are actually paying a person for.
The test worth applying
Forget the theory and ask three concrete questions about your own books:
- When was every account last reconciled against its statement? Not the feed. The statement.
- What is sitting in your uncategorized bucket right now?
- When did anyone last ask you a question about a transaction?
If the answers are “not sure”, “quite a lot”, and “I cannot remember”, then the software is recording faithfully and nobody is verifying anything. That is the gap, and it does not close by buying a better tool.
Where I land on it
I put clients on good software, in their own name, so the books belong to them and nothing is locked to me. That matters: your records should be yours, and you should be able to take them anywhere.
Then I do the part the software cannot. Reconcile against statements. Ask the questions. Notice the things nobody asked about. Close the month whether or not anybody feels like it.
If you are not sure which of those is currently happening in your books, send me three months of statements and I will build you a real Profit and Loss from them, free. Comparing it against what your current reports say tends to answer the question quickly, and sometimes the answer is that you are doing fine and should keep going as you are.