Breaking down the Budget: Will custom GPTs change how accounting firms communicate?by
Developments in generative artificial intelligence could see accounting firms build customised systems based on Treasury documentation to summarise future fiscal events and answer sector-specific questions from clients and staff. Will 2024 be the year of the BudgetGPT?
For many accounting firms, the aftermath of fiscal events such as Budgets or Autumn Statements is a flurry of activity.
Once the Chancellor has sat down after delivering their speech, tax professionals the length and breadth of the country scurry off to scour the small print in the documentation and discover what the changes actually mean for their clients – with the result of these labours often appearing the next morning in the form of emails, reports and perhaps even a tax table mouse mat or two.
But could these fiscal follow-ups soon be joined, or even replaced, by a Budget bot that can churn out unique content relevant to different sectors, companies or individuals, and allow internal staff or clients to ask questions about the changes?
According to two accounting industry insiders, developments in the fast-moving world of generative artificial intelligence (AI) could make this a reality as early as the next Budget.
Earlier this month, and before its much-publicised senior management meltdown, ChatGPT parent company OpenAI opened up the ability to create custom versions of ChatGPT known as GPTs.
These GPTs (and others currently being developed by Microsoft, Google and other AI rivals) leverage the power of the large language model but can be tailored to focus on one specific task within a closed system.
As a trial on his own OpenAI account Alastair Wilson, a tax partner at top 10 firm Azets, used this new capability to create a BudgetGPT.
Wilson set up a GPT within his ChatGPT account, uploaded the documentation released by the Treasury on Autumn Statement day (the Green Book, costings and so on) to create a knowledge base, gave it a tone of voice (friendly or professional, long or short answers), and specifically told the system not to hallucinate (make up answers if it wasn’t sure).
He then interrogated the GPT to verify the accuracy of its answers and check how it was responding to the types of questions clients or junior members of staff might ask.
“In every single case it was spot on,” said Wilson. “For example, I asked it to summarise the changes for R&D-intensive businesses. I was impressed with the response because it didn’t just think about the answer from a tax perspective – it also went through things like grant funding and the British Business Bank’s Future Fund extension, and talked about things like full expensing in the context of R&D-intensive companies.
“Tax practitioners on Budget Day have reams of documents to sift through and pick out what’s relevant to their clients,” he added. “In simple terms, this can do a lot of that heavy lifting for you.
“If the Chancellor sits down at 1.30pm having delivered the Budget, you could have a summary ready to publish on your firm’s website by 2pm, or a chatbot embedded on your site ready to answer clients’ questions. If you want a PowerPoint for a presentation on the event you can ask it to create one, or it can put the results in an Excel document with the bar chart.
“I’d be amazed if they weren’t being used by a good number of firms at the next Budget, perhaps with their own additional commentary to add some differentiation,” concluded Wilson.
A world of AI possibilities
Celso Pinto, founder and CEO of practice management software Pixie, had a similar experience on Autumn Statement day. A team member had the idea of using a GPT to summarise the changes and ask it questions for internal purposes.
While the copy produced by the experimental system came across as somewhat synthesised, it produced accurate results. Pinto believes GPTs will have an increasingly large part to play in the accounting landscape – and not just for standalone fiscal events.
“For a closed, secure ecosystem there are plenty of other applications,” he said. “I could see it deployed in accounting firms for things like training materials, for example. This will give more power to junior staff – it will allow them to access the source material and provide context and information that seniors may not have the time to provide.
“Also, take a list of your clients based on the fees they pay and your last contact with them, add that to your custom GPT and it can produce 10 clients you should get in touch with. It’s not that you can’t already do this with a spreadsheet – it’s just quicker, in natural language and potentially more accessible to others in your firm.”
For Pinto, the advancement of such tools also represents a shift in the balance of power when it comes to who can use technology.
“These tools are really powerful, partly because they don’t take a lot to build,” he said. “Beforehand it required a lot of technical knowledge to construct and populate these systems. Now you can just set it up, add the documents and interrogate it – anyone can create a GPT, drop in documents and then query it. They don’t have to come from vendors, instead, it’s straight from the source.”
To access OpenAI’s GPT functionality, at the time of writing, you need to be a paid ChatGPT subscriber.