AI or business rules: how to choose

Most processes do not need a choice between AI and traditional software; they need each part of the work in the right place. AI is very good at reading, understanding and drafting. Business rules are very good at calculating and deciding the same way every time. This is how we decide which part goes where.

Where AI earns its place

When the input is messy: documents, emails, requests in plain language, voice. When the output is a draft that a person reviews: a summary, an answer, a classification to confirm. When being roughly right is already useful, and someone catches the rest.

  • Reading and classifying documents and emails
  • Understanding a request in plain words and turning it into a plan
  • Summarizing, explaining and drafting answers
  • Finding what matters in large amounts of text

Where business rules earn theirs

When the same input must always give the same output: money, permissions, eligibility, anything an auditor will ask about. Rules can be tested, explained line by line and changed with a review.

  • Calculations and amounts
  • Permissions and access
  • Eligibility, approvals and deadlines
  • Anything that has to pass an audit

Most processes need both

The useful pattern is to let each one do its part: AI reads and proposes, a person decides where it matters, and a rules engine applies what was approved. Sendero, our product, works this way: the assistant turns a request into rules you can read, the owner approves them, and an engine applies them exactly.

In one consulting case, a client planned to have a language model calculate the figures in a business process and explain them. We moved the calculations into tested business rules and kept AI for interpreting and explaining the results. The figures became auditable, and the AI kept the part where it adds the most.

Five questions that decide it

  • Must the same input always give the same output?
  • Will someone have to explain or audit the result?
  • Is the input free text, documents or voice?
  • What does a wrong answer cost, and who would notice it?
  • Can a person review the result before it is applied?

The consulting caseHow Sendero works ↗

Questions

When should I use an LLM instead of business rules?
When the input is unstructured, such as text, documents or voice, and a person can review the result. For calculations, permissions and anything audited, rules are safer: the same input always gives the same output.
Can AI and a rules engine work together?
Yes, and that is usually the best design: the AI interprets and proposes, a person approves where it matters, and the rules engine applies the decision the same way every time.
Is it simpler to use AI for everything?
It looks simpler at the start. Over time, figures that can change from one run to the next are hard to test and to audit, and errors are expensive to find. Rules take longer to write once and then behave the same way every time.

Not sure which one your process needs?

Tell us about it. The first conversation is free.

Write to us