Both are automation
A common mistake is treating AI as the automation and everything before it as manual work.
ERPs have automated things for decades. Reordering rules. Approval routing. Recurring invoices. Scheduled reports. Follow-up sequences.
That is rule-based automation, and for a large share of what businesses need, it is still the better answer.
The question is not which is more advanced. It is which fits this particular task.
The difference in one line
Rule-based automation does what you told it.
AI-based automation does what usually works.
Everything else follows from that.
A rule is written by a person. It is explicit, predictable and auditable. It handles exactly the cases it was written for and fails on everything else.
A model learns from examples. It handles cases nobody anticipated, and it is right most of the time rather than always.
FIGURE 1: THE CORE DIFFERENCE
Rules
- Written by a person
- Always behaves the same way
- Explainable line by line
- Breaks on anything unanticipated
Patterns
- Learned from your data
- Usually right, occasionally not
- Hard to explain in detail
- Handles cases nobody wrote a rule for
Where rules win
Five situations. Notice how many ordinary business tasks fall here.
When the logic is definite. “Orders over 50,000 need director approval.” There is no judgement. A rule is correct every time and a model would be worse.
When you must explain it. Anything an auditor, regulator or customer may question. “Because the rule says so” is an answer. “The model determined it” is not.
When errors are expensive. Payment thresholds, credit limits, tax treatment. You want predictable, not usually-right.
When you have little data. A model needs examples. A rule needs one person who knows the policy.
When the rule changes by decision. Your approval limit changes because management decided. A model learning from history would still be applying the old one.
Where AI wins
Four situations.
When the variation is unbounded. Supplier invoices come in thousands of layouts. Writing rules for each is impossible; recognising an invoice total is exactly what a model does well.
When the pattern is real but nobody can articulate it. Which leads convert. Which invoices get paid late. There is a pattern; nobody can write it as a rule.
When it should adapt. Demand shifts. A fixed reorder minimum does not. A forecast does.
When the input is unstructured. Documents, free text, images. Rules need structure.
FIGURE 2: WHICH TOOL FOR WHICH JOB
Use rules when
- The logic is a definite policy
- You must explain the decision
- An error is expensive
- The rule changes by management decision
Use AI when
- The input varies without limit
- A pattern exists that nobody can write down
- Conditions shift over time
- The input is documents or free text
Most systems need both
The practical answer is rarely one or the other. It is usually a chain.
A worked example — processing a supplier bill:
AI reads the scanned document and extracts vendor, dates, amounts and lines. Rules could not handle the variety of layouts.
Rules check it. Does the total match the purchase order within tolerance? Is the vendor approved? Is the amount within the approval limit? These are policies and they must be exact.
AI flags it if the amount is unusual for this vendor. No rule captures “unusual” well.
Rules route it. Over the threshold, it goes to a director. That is policy.
A person reviews and posts.
Each part uses the right tool. Extraction where variety is unbounded, rules where the policy is definite, a person where money moves.
FIGURE 3: THE TWO WORKING TOGETHER
AI reads
- Any layout, any format
Rules check
- Policy limits, approvals, tolerances
AI flags
- Anything unusual for this vendor
A person posts
- The commit step stays human
The mistake in each direction
Using AI where a rule would do. More expensive, less predictable, harder to explain. If your approval limit is 50,000, write a rule. Do not train a model to guess where the limit is.
Using rules where the variety defeats them. The classic version is invoice processing built as a template per supplier. It works, then a supplier changes their layout, and you are writing rules forever.
A quick test: if you can write the logic down in a sentence or two, use a rule. If you would need pages of exceptions and still miss cases, use AI.
What people forget about rules
Two things.
Rules also rot. A reorder minimum set three years ago is still running with last year’s demand. A rule is predictable, not correct — it keeps doing what it was told even when what it was told is outdated.
Review your rules yearly. The same discipline you would apply to anything else.
Rules are easier to audit and easier to forget. Both are true. Nobody questions a rule that has been running for years, which is exactly why it should be checked.
What people forget about AI
It needs data. Not just any data — enough of it, clean, with outcomes recorded. Most disappointments trace back to this rather than to the model.
It cannot explain itself in detail. For an operational nudge, fine. For anything you must defend, use a rule.
It changes. Retrained on new data, it may behave differently. A rule does not drift.
Deciding for a given task
Five questions.
Can I write the logic in a sentence? Yes — rule.
Must I explain the decision? Yes — rule.
Is the input structured or unstructured? Unstructured — AI.
Do I have hundreds of examples with recorded outcomes? No — rule.
Does an error cost real money? Yes — whichever you use, keep a person on the commit.
The short version
Rules and AI are both automation, and they suit different problems.
Rules for definite policy, explainability and expensive errors. AI for unbounded variation, patterns nobody can write down, and unstructured input.
Most good systems use both — AI to read and flag, rules to check and route, a person to commit.
And review your rules yearly. Predictable is not the same as correct.
Not sure whether your automation problem needs a rule or a model?
Get in touch. We will look at the specific task — and quite often the answer is a rule you can write this afternoon.
