Prompt your colleagues: what Tax professionals can learn from interacting with an LLM

Discover how prompt engineering can enhance delegation skills for tax professionals.

In brief:

• Prompt engineering improves communication and delegation.
• Clear instructions lead to better outcomes in tax tasks.
• Learning from LLM interactions can streamline workflows.

Innovation is crucial in the ever-changing landscape of taxation and technology. Where accuracy and precision are essential, prompt engineering can significantly enhance the quality of AI-generated results. The effectiveness of language models (LLM) heavily relies on this technique—the art of formulating specific instructions to obtain optimal responses. That's why prompt engineering is akin to coaching your own personal assistant or intern; the more effort you invest in it, the better the outcomes. Tax professionals can learn valuable lessons from interacting with a LLM to improve delegation.

Building confidence: the next evolution of AI

During my first year at EY, eager and fresh out of university, a colleague casually asked me to “do some thinking around” the tax consequences of gifting an estate. Determined to impress, I went all in, pulling out legislation and case law left and right. Some time I proudly handed over a 25-page memo packed with footnotes, appendices, and probably a few unnecessary Latin phrases. The colleague glanced at it, laughed, and said, “This should have been three pages max.” While there was some useful thinking in there, it was a classic case of miscommunication. Looking back, some prompt engineering would have saved us both a lot of time. If he had said, “Summarize in three pages the key tax consequences of gifting an estate, focusing on exemptions and recent case law,” I would have known exactly what to deliver. Turns out, clear instructions matter—whether you’re delegating to a colleague or prompting an AI.

Lessons from LLMs for delegation:

1. Start with context: In LLM prompting, providing context ensures the model understands the problem space. Similarly, when delegating, explain the background. For instance, when asking someone to analyze VAT implications for cross-border services, share relevant client details and past rulings and similar example cases.
2. Be specific about outcomes: LLMs thrive on specific instructions. Humans, too, appreciate knowing what success looks like. Instead of “Prepare a tax compliance report,” specify, “Draft a tax compliance report for Q4 2024, highlighting deviations from prior periods and including recommendations for optimizing deductions.”
3. Iterate and refine: Prompting an LLM often requires iterative refinement. Similarly, delegation benefits from check-ins and feedback. Ask your colleague to share a draft midway, ensuring alignment before they invest too much time.
4. Anticipate questions: A well-engineered prompt anticipates ambiguities. Humans are no different. If a task might raise questions, preemptively address them. For example, when requesting an analysis of treaty benefits, specify whether you’re focusing on withholding taxes or permanent establishment definitions.
5. Inject examples: Examples help LLMs and humans alike. If asking a colleague to draft an audit response, share a previous response that aligns with your expectations.
    

Clear communication is key

Prompt engineering is more than just a tech skill; it’s a framework for clear communication. For tax professionals, mastering these techniques can bridge the gap between intent and outcome in human interactions. After all, whether dealing with LLMs or colleagues, clarity, context, and specificity are your best friends, especially if you need specific results.

And remember: Delegating tasks without context is like filing taxes without the right data —you’re just asking for trouble.