I have been in practice long enough to remember the days when a client would walk in with a shoebox of bills, and your heart would sink a little. Not because the work was hard, accounting is never really hard, but because the process was exhausting. The data entry. The cross-checking. The "please send me that invoice again" emails. The 2B mismatch conversations that somehow always happened the night before a filing deadline.
What changed for me was not one big moment. It was a series of small ones.
When I Started Building Instead of Complaining
A few years ago, I started noticing a pattern. Every tool I was paying for solved 80% of my problem and left the last 20% to me — manually, in a spreadsheet, at 11pm. And that 20% was always the same things: reconciliations that needed judgement, reports that needed reshaping, entries that needed context a generic software didn't have.
So I started building small tools. Not enterprise software. Just quiet, focused utilities that did one thing well. A reconciliation checker. A GSTR-2B matching tool. An extraction helper that could read a scanned invoice and pull out the numbers.
None of these were glamorous. Most of them looked terrible. But they worked for my workflow, and that was enough.
What AI Actually Changed
Let me be honest about something: AI did not change accounting. Double entry is still double entry. GST rules are still GST rules. A ledger that does not balance does not balance whether you used a quill or a language model.
What AI changed was the distance between a document and a decision.
Before, if a client sent me a bank statement, I had three choices: enter it manually, export it to Excel and write formulas, or use whatever import tool the software provided and spend an hour fixing what it got wrong.
Now, I can drop that statement into a tool, and within seconds it tells me: here are the probable ledger heads, here are the entries I think you want, here are the ones I am not sure about. I review. I approve. I move on.
The AI does not replace my judgement. It removes the friction before my judgement is needed.
The Tools That Actually Help
Over time, I have built and used a set of tools that I now genuinely cannot imagine working without:
Document extraction — Give it a purchase bill, a salary slip, a bank advice. It reads it, structures it, and hands you a draft entry. The accuracy is not perfect. But 80% accuracy on a first pass still saves enormous time compared to typing from scratch.
Reconciliation matching — The 2B mismatch problem is one every CA knows intimately. Vendor filed, you did not claim. You claimed, vendor did not file. You claimed a different amount. A good matching tool does not solve this, but it narrows it down fast. It shows you the mismatches, ranks them by likely cause, and lets you act. That alone saves hours each month.
Smart narration and classification — Teaching software to recognise your own patterns is underrated. Once it knows that a payment to a particular vendor is usually a freight charge and not a purchase, it stops asking. That kind of context-aware classification is the real productivity gain — not the flashy AI chat interface, but the quiet, learned intelligence underneath.
Reports on demand — The ability to ask "show me all cash payments above ten thousand this month" and get an instant answer, not a filter exercise in a spreadsheet, is genuinely useful. Not just for me, but for clients who want to understand their own numbers.
The Shift I Did Not Expect
When I started building tools for myself, I assumed I was optimising my own workflow. What I did not expect was that the tools would change what I could offer.
When reconciliation takes four hours, you do it once a month at the end of the period. When it takes twenty minutes, you do it weekly. When it takes five, you start doing it before every filing as a matter of habit. The economics of the task change, so the behaviour around the task changes.
That is the real unlock. Not that AI is faster. It is that the speed changes what is economically worth doing.
I now offer clients a level of monitoring that I simply could not have justified billing for before, because the marginal cost of doing it has dropped so much that it fits inside any reasonable retainer. That is a practice model change, not just a process change.
What I Tell Other Practitioners
A few things I have learned that might save you the wrong turns:
Start narrow. Do not try to automate everything. Pick the one task that costs you the most time and the least judgement — usually data entry or matching — and solve that first. Prove the value before you expand.
Own your review layer. Any AI tool that posts entries without review is a liability. The value is in the suggestion, not the automation. You want AI to shorten the distance to the decision, not to make the decision.
Your clients' data is not a test case. Build tools that treat financial data as sensitive by default, scoped by client, auditable by action, and never mixed. The bar for tools that touch accounting data should be higher than for tools that suggest restaurant recommendations.
The free tools are often enough. Some of the most useful things I use cost nothing. Open source libraries, community-built extractors, spreadsheet plugins. The gap between free and paid is not always quality; sometimes it is just marketing. Evaluate on fit, not price.
Where This Is Going
I think the next few years will see a quiet but significant change in how accounting practices are structured. The tools that used to require an enterprise budget are becoming accessible to solo practitioners and small firms. The work that used to require dedicated data entry staff can increasingly be handled by a well-designed extraction layer.
This does not mean accountants are being replaced. If anything, the practitioners who are building or adopting these tools are finding they can handle more clients, do better work, and spend more of their time on the things that actually require a trained mind : interpretation, compliance, planning, judgement.
The shoebox of bills has not disappeared. But what I do with it in the first thirty minutes has changed completely.
And that, quietly, changes everything.
If you are working on similar tools, or have thoughts on where this is heading, I would genuinely like to hear it.
Frequently Asked Questions
What is the main point of this article?
I have been in practice long enough to remember the days when a client would walk in with a shoebox of bills, and your heart would sink a little. Not because the work was hard, accounting is never really hard, but because the process was exhausting. The data entry. The cross-checking. The "please send me that invoice again" emails. The 2B mismatch conversations that somehow always happened the night before a filing deadline.
How can business owners apply this in practice?
Before, if a client sent me a bank statement, I had three choices: enter it manually, export it to Excel and write formulas, or use whatever import tool the software provided and spend an hour fixing what it got wrong.