Through 2023, the firm focused on training staff on how to use chatbots and write effective prompts.
In 2024, it started building agents, including the TaxBot mentioned above.
Munnelly said building that bot started with locating tax advice written by partners, which he said was "stored all over the place" – often on tax partners' laptops. KPMG found as much of that advice as it could and placed it in a RAG model along with Australia's tax code to produce an Agent that creates tax advice.
"It is very efficient," Munnelly told the Forrester conference. "It does what our team used to do in about two weeks, in a day. It will strip through our documents and the legislation and produce a 25-page document for a client as a first draft.
"That speed is important," he added. "If we have a client who is about to do a merger, and they want to understand the tax implications, getting that knowledge in a day is much more important than getting it in two weeks' time."
"That is really changing our business and how we work."
Munnelly said KPMG built the agent by writing a 100-page prompt it fed into Workbench. The Register asked for details of the prompt and Munnelly said a substantial team worked on it for months, and the resulting agent asks for four or five inputs before it starts working on tax advice, then asks a human for direction before generating a document.
Only tax agents can use the tool, because its output is not suitable for people without deep tax expertise. //
The chief digital officer said KPMG has deployed agents that do frustrating and time-consuming work people would rather avoid, and that staff surveys suggest employee satisfaction has risen as AI frees them to spend more time working on challenging tasks, leading them to rate the firm as more innovative.
"They just don't want to do the boring stuff," Munnelly said. "They want to get out there and help clients with chewy problems." //
An_Old_DogSilver badge
Sprawling, Unmaintainable, Spreadsheet Macros: The New Generation
-
Does this new, faster method produce complete and accurate results? No.
-
Is this 100-page LLM prompt effectively-maintainable software? Probably not.
-
Does this smack of corporate-image-spinmeistering over rationality and logic? Yes.
