Some jobs in your business need one correct answer, every time. Others need interpretation and judgment. Software can help with both, but each requires a different tool, and many AI projects go wrong when the two kinds of work get mixed up.
Take payroll. At the end of the month, every employee must receive exactly the right paycheque. There is no room for interpretation, and the same hours and rates must produce the same amount every time. That is deterministic work, meaning the same input always produces the same output. Rules-based software, including what is often called robotic process automation (RPA), is built for it. As a recent Harvard Business Review article notes, traditional software behaves this way, so you can test and document it.¹
Now take a different job: reviewing your financial statements and explaining what they say about the health of the business. There is no single correct answer to look up. The work is spotting patterns, weighing what matters and putting it into plain language. This is where AI tools built on large language models excel. They are probabilistic, working from patterns and likelihood,¹ so the same request can come back worded differently. That is acceptable here, because the value lies in the insight.
Each tool is better at a different kind of numbers work, and most business processes need both types of software. I covered why AI should be reserved for the steps that need reasoning in AI Belongs in the Workflow. The practical point here is how the two fit together on a single job.
Here is an example to illustrate the point. Rules-based software pulls the figures from your accounting system and calculates the ratios the same way every month. AI reads the results and explains what changed and why it may matter. A person reviews that explanation before it informs a decision.
A customer quote works the same way. AI drafts the cover note and summarizes what the customer asked for, while rules-based software calculates the price from your price list, so the number is right every time. A person checks both before the quote goes out.
Before you assign any task to AI, ask whether it needs the same answer every time or a thoughtful one. Match each job to the right tool, and you get both the precision and the insight.
Sources
¹ M. Alejandra Parra-Orlandoni and Paulo Carvão, “You Outsourced the AI—but You Still Own the Risk,” Harvard Business Review, July 9, 2026.