Prakash Patil is the chief consultant of a GST audit practice. He does the work most people would call the hard part: going into a business, reading its books against the law, and working out what it owes and where it is exposed.
That was never what limited him. What limited him was the writing up.
A single audit report took seven to eight days to produce. Not the audit. The report. He now produces the same report in five minutes, and the number that changed as a result is not the one most people expect.
Why can't a good consultant take on more clients?
Quick Answer: Because the constraint is usually delivery, not demand. Prakash could complete about two audits a month, and that ceiling had nothing to do with finding clients. Writing each report consumed seven to eight days, so the practice could only ever serve as many clients as he could write up.
Most advice aimed at consultants assumes the problem is the top of the funnel. Get more leads, get more visible, get better at selling.
Prakash did not have that problem. He had the other one, which is quieter and more expensive: work he had already won, waiting behind work he had already done.
Two audits a month was the ceiling. Every hour spent assembling a report was an hour not spent in a client's office doing the thing clients actually pay for. And because the report is the deliverable, he could not skip it, shorten it, or hand it to someone junior without the quality showing.
That is the shape of the problem in almost every expertise business. The expert is the bottleneck, and the bottleneck is rarely the expertise. It is the documentation of the expertise.
Your first action: Work out your own ceiling. How many clients could you serve this month if delivery took half as long? If that number is meaningfully higher than what you serve today, your constraint is not marketing.
Can AI write a GST audit report?
Quick Answer: Yes, and Prakash's does considerably more than summarise. It produces observations, the legal framework behind each one, the business impact, recommended actions, a calculation of the client's exposure, and his own branded formatting. Seven to eight days of work, in five minutes.
The important detail is how much of the report it actually produces.
A summary would not have helped. A GST audit report is not a document you skim, it is one a client acts on and, occasionally, defends. Prakash's version has to carry the observation, the section of law it rests on, what it means commercially, what he recommends doing about it, and the number the client is exposed to if they do nothing.
All of that now comes out of a single Claude skill he built during a hackathon session, formatted and branded as his practice sends it.
He did not automate the audit. He automated the part of the audit that was stopping him from doing more audits.
The distinction matters if you are trying to work out whether this applies to you. Look at what you produce after the thinking is finished. If the write-up takes longer than the work, that gap is the opportunity.
Your first action: Time your last deliverable honestly, separating the thinking from the writing up. Most professionals have never measured the two apart, and the ratio surprises them.
Does AI replace professional judgement here?
Quick Answer: No. Prakash still does the audit, reads the books, and decides what is wrong and what it means. The skill writes up conclusions he has already reached. Nothing about the legal position is decided by a machine, and a practice built the other way round would not be defensible.
This is the question every professional asks, and it deserves a straight answer rather than reassurance.
The judgement in a GST audit is the whole product. Which transactions to question, which exposures matter, what the law says in a particular set of facts, and what a client should actually do about it. Prakash makes every one of those calls, as he did before.
What changed is the eight days spent turning those calls into a document that reads like a professional firm produced it. Assembling, formatting, cross-referencing, calculating and laying out. Work that requires care, and none of which requires him.
There is also a reason this is the right boundary commercially, not just ethically. If the judgement were automated, the practice would have nothing left to sell. Automating the write-up makes his judgement more available, which is the opposite problem.
Your first action: Draw the line before you build anything. Write down which decisions in your work are the product, and treat everything on the other side of that line as fair game.
What is a Claude skill, and why not just use a prompt?
Quick Answer: A prompt is a request you make each time. A skill is a procedure written down once, that runs the same way every time. For a report that carries legal citations and exposure calculations, consistency is the whole point, and a prompt typed fresh each month cannot give you that.
Prakash's report is not a one-off piece of writing. It is the same document, correctly, forty times a year.
That is what a prompt is bad at. Type it again next month and you will phrase it differently, forget a section, or get a different structure back. For a birthday message, fine. For a document that cites law and calculates what a client owes, that variance is the risk.
Writing it as a skill means the structure, the sections, the legal framing, the calculation method and the formatting are all fixed. He supplies the findings. The procedure does the rest, identically, every time.
The general form: if you do something once, prompt it. If you do it repeatedly and it has to come out the same way, write it down as a skill. Prakash's practice is a good demonstration because the cost of inconsistency in his output is unusually visible.
Your first action: Take the thing you have re-prompted more than five times and write it down properly once, with the structure and standards baked in.
What actually changed in the practice?
Quick Answer: Throughput. Prakash went from around two audits a month to two client visits a week, all of them billable, which is roughly four times the capacity from a single skill. His fees did not change and neither did his marketing.
This is the part worth sitting with, because it is not the number people expect from an AI story.
He did not get faster at winning clients. He did not raise his prices. He built one thing, and the work he could already win stopped queueing behind the work he had already done.
Two audits a month became two client visits a week. Every one of them billable, because the visit is the billable part and the report was never what a client was paying for.
One caution on reading this: four times the capacity is not automatically four times the revenue. Capacity is the ceiling, and filling it is a separate job. What the build removed was the reason he could not have grown even if the clients were queueing at the door.
Your first action: If you free up delivery capacity, decide in advance what fills it. Capacity nobody books is just a shorter working week, which is a fine outcome only if you chose it.
FAQ
Does this work for chartered accountants and auditors outside GST?
The specific report is a GST audit, but nothing about the approach depends on that. Any practice whose deliverable is a structured document built from expert findings has the same shape: tax, statutory audit, due diligence, compliance reviews, legal opinions.
How long did it take Prakash to build?
It came out of a single hackathon session in April 2026. The reason it was fast is that he was not inventing the report, he was writing down one he had already produced hundreds of times.
Is the output safe to send to a client?
He reviews it, and he signs it. The skill produces the document from his findings; the professional responsibility for it stays exactly where it was. Any consultant using AI on client deliverables and not reading them before they go out has made a different and much worse decision.
Do I need to be technical to build a skill like this?
No. The hard part is describing your own procedure precisely, which is a subject-matter task rather than a technical one. If you can brief a competent junior on exactly how the report should be built, you can write a skill.
Can I verify this?
Prakash Patil is on LinkedIn at linkedin.com/in/prakash-patil-gst-consultant. The figures here come from his own reporting during the programme and he has seen the claims attached to his name.
If you want to work out where this would start in your practice, book a call. For a case with three systems rather than one, read what a 64-year-old contractor built.