One of the most common questions I get when it comes to AI in accounting is “where do I even start?”
Given the near-endless possibilities, it’s no wonder that it’s overwhelming.
Whether you’re just starting out, or if you want to expand your firm’s existing capability, here are a bunch of real use cases for AI. These are the ones that I’ve found work well – not just a list of things that sound good in theory but fall down in practice.
Important note: However you decide to use AI – you either need to remove all sensitive client data, or be sure that the system you’re using does not use your data to train their model. If you’re using ChatGPT, you can opt out of chat history & training.
1. Excel Macros & Google Sheets Script
Just about every time I need to write a macro or script, I’ve forgotten where I need to go to even start writing it.
That’s the insane thing about asking ChatGPT to write you a script – it’ll give you a step-by-step guide on where to go to save the code.
Seriously you can just describe what you want it to do and get a ready-made macro along with instructions on what to click and where to go to save the code.
Here’s a prompt I used:
I want to add a button on my Google Sheet which duplicates the current sheet, renames the new one to the date, in the format YYYYMMDD, then switches to the new sheet. How do I do that?
The result:
When I tested it, I needed a tweak.
Can you modify the script so that the new sheet appears immediately to the right of the current sheet. Right now it adds at the end
Then it was as easy as copying and pasting the new code.
2. Excel & Google Sheets formulas
Complex formulas can take a bit of time to get right, especially when you have 5+ nested layers of formulas. Often GPT will just give you what you need.
Tip: Tell it what column or cell the data you want to include is in. If you don’t, it’ll use a default like “A2” and you’ll have to replace all instances of that in the formula.
Here’s a really basic example. This formula would have been easy to write manually, but it was a lot easier to just paste the current tax brackets into ChatGPT and have it write the formula.
here are tax rates in australia, please create an excel formula that takes a value in A2 and outputs the payable tax
Taxable income
Tax on this income
0 – $18,200
Nil
$18,201 – $45,000
19c for each $1 over $18,200
$45,001 – $120,000
$5,092 plus 32.5c for each $1 over $45,000
$120,001 – $180,000
$29,467 plus 37c for each $1 over $120,000
$180,001 and over
$51,667 plus 45c for each $1 over $180,000
The result:
3. Transaction queries
For uncoded or unreconciled transactions, AI can help you identify what those are without having to ask your client.
Of course, you may want to ask anyway to be sure, but sometimes it’ll find something that is obvious, and you can eliminate that from your queries.
For your prompt, give it the chart of accounts, business name and business location along with some uncategorised transactions. Tell it to pick a category from the accounts and add a note about why it thinks it’s the right one. The answers are often pretty accurate.
For some reason, I’ve found more accurate answers when I allow it to use “Unknown” as a category. When I asked it not to say Unknown, the accuracy of the others suffered. This is worth experimenting with.
Here is a list of uncoded transactions for a software company that is based in Brisbane, Australia.
Rydges World SquareOPI Sydney NSW
Soak Mermaid Beach Mermaid BeacAUS
WiFi Onboard Brisbane AUS
INTNL TRANSACTION FEE
INTNL TRANSACTION FEE
CHATGPT SUBSCRIPTION SAN FRANCISCCA ##0224 20.00 US DOLLAR
NOTION LABS- INC. SAN FRANCISCCA ##0224 20.00 US DOLLAR
PI* SBO FINANCIAL BRISBANE QLD
IINET BATCH PERTH GPO WA
XERO AU INV-MASKED HAWTHORN VIC
Google ADSMASKED Sydney AUS
For each of these, categorize it into one of the categories below, and provide a note about what you think this transaction is. Make a table with 3 columns - the transaction, the category and the note. If you are not confident about the category, write "Unknown"
CATEGORIES
Marketing
Video Marketing
Amortisation expense
Parking
Bank Fees
Paypal Fees
Stripe Fees
Card Processing fees
((insert full list of categories here))
Result:
4. Handling basic client questions & support
Many email clients and support systems have introduced AI drafting. With some, you can point it at your existing content or support documentation to learn from.
In accounting, you could provide a library of your frequently asked questions, along with answers.
Your options depend on how you want to allow clients to contact you. For example, if clients email you directly, you’ll have very limited options. This screenshot below from Superhuman drafts a few different replies to each email. They’re OK, but not amazing.
On the other end of the spectrum, if you give clients the ability to contact you through a chatbot-style app, you can have it automatically provide answers. A great example is Intercom’s Fin, which provides references to where in the documentation it found the answer, referring the customer to it. Then they can specify if their question was answered, or if they need more help.
I’ve been on the client side of this interaction and I was so happy with it. It answered my question exactly and gave me a link to the exact place in the documentation where that answer came from. I thought it was so good, that I posted about it on LinkedIn.
The middle ground is if you’re using some kind of centralised helpdesk or unified inbox, where all your emails come into a different system, like Front, Helpscout or Missive. The AI capabilities of these vary, but often at a minimum, you can draft replies. If they aren’t able to be trained on your own documentation, I find the drafts are usually pretty average.
5. Weekly summaries of client questions
As your business grows, you may find yourself having less and less direct client contact. But it’s this contact with clients that can generate some of the best business improvement ideas.
AI can bridge this gap by summarising what you receive from clients, including feedback, questions, or issues they have.
This too is highly dependent on what systems you are using, and it’ll need a bit of automation wizardry. You’ll need to be proficient with Zapier or Make.
In our business this process is made up of 3 automation:
- Store the ID of all conversations that are closed in our support system
- After a week has passed, summarise them and add the summary to a spreadsheet
- Each week, get all the rows in the spreadsheet and create a summary of those
Part 3 looks like this, in Make:
6. Reviewing documents or creating summaries
This is one of those ones that’s regularly thrown around by AI influencers, and there’s even entire products dedicated to “asking your documents questions”.
It’s been pretty handy for me on a few occasions.
I’ve used an AI summary for a first pass on contracts, terms and conditions and random things.
While you should absolutely not use AI as a substitute for a proper legal review, it’s helped me identify red flags by turning a monster document full of legalese into one that an actual human can read.
Here’s an example prompt:
You're a legal expert who distils legal contracts into easy-to-understand dot points. <Describe your business and what you do>. The below agreement was sent to us by a customer and they want us to sign it. I would like to know what the implications are for us. Could you please summarise this to create a list of implications that we need to be aware of?
7. Extracting information or testimonials
This has to be one of my favourite use cases for AI – extracting information from larger documents or transcripts.
We use this extensively during our case study process. We do a 15-minute call with a customer about how they use our product, what they like, and what they don’t like.
Then we feed that into a few prompts. Note, this is specific to our SaaS product, but the process is the same for client calls that you do at your firm.
Generating a review:
Below is a transcript from an interview with a customer of our SaaS product. Pretend you are the person who was interviewed, and answer the following questions based on what they said. List each question, followed by your answer.
- What do you like best about Content Snare?
- What do you dislike about Content Snare?
- What problems is Content Snare solving and how is that benefiting you?
TRANSCRIPT:
Generating quotes and testimonials:
Below is a transcript from a conversation with a customer of our software product Content Snare. The customer is a (insert industry here) who uses Content Snare to collect information from clients. Please extract the most impactful quotes spoken by the customer. I would like quotes that highlight the benefits of the product, explain how their life or business has improved, or show a positive experience they had with Content Snare. You can combine quotes from different parts of the transcript into one. You can remove filler words. You can paraphrase a small amount. Please give me the quotes in list form.
TRANSCRIPT:
I’m AMAZED at how well this replicates the exact quotes I would have pulled out of the transcript – all in a few seconds.
This is just an example – you can use the same process to extract anything from transcripts, like:
- Action items
- Client questions
- Objections (great for helping to create a sales process)
- Feature requests
- Basic client info to store in your CRM
And pretty much anything.
8. Content ideas
This one has been done to death, but it does belong in this list. If you’re stuck on what kind of content to create for your clients and prospects, ChatGPT can help. But you need to get creative – asking “what can I write about” ain’t going to cut it.
A great example of this is to ask it to give you 10 sub-niches for your main niche, then 5 topics within those niches.
I’ve had success with variations on this prompt:
You are a content strategist. You want to write content that will help our audience of accountants and bookkeepers run a more profitable, more efficient business. What are some content ideas for this audience? You should consider commonly asked questions, keywords that they might type into Google, or anything that an accountant or bookkeeper should know. Return the results as a list.
Once you have the initial list, you can ask it to dig into particular topics more. For example, it suggested time management for accountants. I asked it to generate 40 topics closely related to “time management for accountants”. Asking for a big list is helpful, as many suggestions will suck, but there is usually a handful of really good ideas in there.
Another trick is to prime it with some of your FAQs, client call summaries, or whatever other data you have that could help it identify client struggles.
9. Creating reports
Paste in some data or upload a spreadsheet, and ask it to visualise or graph something from it.
This has fairly limited use cases for me, but it does come in handy. I find it’s best for one-off reports. For anything ongoing, you’re better off using a proper reporting or dashboard tool.
10. Meeting transcripts & summaries
This has to be the most talked about, and probably over-stated use case of all AI, but it’s worth doing.
If your memory is anything like mine, you’ll forget what happened in a call on Friday simply because it’s now Monday and last week feels like 6 years ago.
I use Fireflies.ai to give me dot points and transcripts which have saved me a few times.
The added benefit of having all your previous transcripts on file is that you can use them in other ways in the future. For example, we can use past demo call transcripts to identify common objections or questions to make it easier to train new salespeople. Likewise, you could have AI scour transcripts for content ideas.
The earlier to start recording, the more information you’ll have. Again, just be careful about sensitive client data and do not record the calls where that’s likely to come up. Destroy or edit the transcript if it does.
Focus on what’s useful
With the pace of improvement, there will be new ways you can use AI every other day. The hard part is identifying which prompts and ideas are actually useful – because let’s be honest, a majority of what’s shared isn’t that useful in a professional context.
But occasionally, you’ll find something that truly revolutionalises a process that used to take you hours. That’s what many of the above have done for me. Let me know if any of these stick out to you, and if you’d like me to dig deeper in a future post.