Your Clients Do Not Need a Faster Accountant. They Need to Hear From You.

Ask a room of practice owners what they want AI to do for them and you get a fairly consistent answer. Speed up the compliance work. Get through the recons faster. Shorten the time between receiving a client’s records and issuing something.

It is a reasonable answer. It is also, in my experience, the answer that changes the least about a practice.

Here is why. Time you save on compliance work has a habit of refilling with more compliance work at exactly the same rate. You take on another two clients, or you finally get to the backlog, or the work simply expands into the space you cleared. The efficiency is real, but it very rarely arrives in the fee. Clients are not paying you for the hours. They are paying you for the file being right and being on time, and both of those things were already happening before you got faster.

The place where these tools genuinely change something is at the client boundary. Not in how quickly you produce the work, but in how often the client hears from you and what it feels like when they do.

The gap most practices do not look at

Try this exercise on one client, and be honest about it. Write down every single time your practice made contact with them over the last twelve months. Not every time they contacted you. Every time you went to them.

For most clients on most books, the list looks something like this. An engagement letter renewal. A request for documents. A follow up request for the documents they did not send. The annual financial statements going out for signature. Two provisional tax reminders. Possibly an invoice query.

Every item on that list is transactional. Every item is something you had to do in order to complete work you were already contracted to do. There is nothing on there that a client would describe to a friend as advice.

Now write down the moments in that same year where contact would have been genuinely valuable and did not happen. The client who bought a second property and did not think to mention it. The one whose turnover crossed the VAT registration threshold in month seven and only found out in month eleven. The one who was quietly considering bringing someone into the business. The one who got a letter from SARS, panicked, googled it, and only phoned you three days later.

That second list is where your fee lives. It is also where retention lives, and it is almost entirely where referrals come from. Nobody has ever referred their accountant because the annual financial statements arrived promptly. They refer the accountant who phoned them about something before they had to ask.

Why your AI output sounds generic

Most practitioners who have tried these tools for client communication have had the same experience. The output is fine. It is also bland, slightly American, and reads like it was written by somebody who has never met your client.

The instinct is to assume you are prompting badly, and there is an entire industry built on selling you better wording. I do not think wording is the lever. The problem is almost always missing context.

The tool does not know that your client base is mostly owner managed businesses in the R5 million to R40 million turnover range. It does not know that you do not do audits and refer them out. It does not know your February is a write off, that you write to clients in plain language and never in tax jargon, or that you have a rule about never giving an opinion on a SARS matter over email. It does not know any of that because nobody told it.

The single most useful thing a practice can build here is not a clever prompt. It is a reusable block of context about the firm itself, covering who you are, who your clients are, how you speak to them, what you do not do, what your compliance year looks like, and what your professional non negotiables are. Write it once, properly, and paste it in front of everything you ask for afterwards. The difference between output produced with that block and without it is not marginal. It is the difference between something you would delete and something you would send.

Where these tools must not go

I want to be direct about this, because the enthusiasm in this space tends to skip over it.

General purpose AI tools produce confident, fluent, well structured answers that are sometimes simply wrong. Not obviously wrong. Wrong in the way that survives a quick read. Almost every practitioner I speak to has already been handed a plausible and incorrect answer on a technical point, often with a section reference attached to it that does not say what the tool claims it says.

So the line is straightforward. These tools have no place in forming a technical position. They do not determine your treatment, they do not interpret legislation for you, and they do not decide what goes on a return. What they are useful for is communication: explaining a position you have already formed, drafting the letter that carries it, turning a set of numbers into something a client can actually read, and helping you reach clients you would otherwise not have reached this quarter.

The rest of it matters too. Client information going into a public tool is a POPIA question and a confidentiality question before it is a technology question. And whatever goes out under your firm’s name is yours, professionally, regardless of what drafted it. That responsibility does not move.

What to do about it

If you take one thing from this, take the exercise. Pick a client, map the year, and look at what you actually sent them. Most practitioners find the gap uncomfortable, and that discomfort is the useful part.

In the CPD session I am presenting for The Tax Faculty, we do exactly that, live, on one of your own clients. Then we build the practice context block together, and use it to write one real client communication that fills the biggest gap on your own map. You leave with an asset you will reuse for years and something you could send that afternoon.

The firms that get value out of this are not the ones with the best tools. They are the ones who worked out which client conversations were not happening, and then made them happen on purpose.

There are not comments for this article at the moment, check back later.
You must be logged in to add a comment, log in now.

Explore Smarty