Cutting through the noise
Every software product now claims to have AI in it. Most of the claims are vague, and some are simply a marketing label applied to features that existed before.
So it is worth being clear about what the term actually means, because you are going to be making buying decisions with it.
AI is software that produces useful output from patterns in data, rather than from rules a person wrote in advance.
That is the whole idea. Traditional software does what it was told. AI software produces an answer based on what it has seen before.
The difference in practice
Take reading an invoice.
Traditional approach. A developer writes rules. “The total is the number after the words ‘Amount Due’.” It works on invoices matching that layout and fails on the rest. Every new supplier format needs new rules.
AI approach. The system has seen millions of invoices. It recognises what an invoice total looks like regardless of layout, wording or language. It handles formats nobody programmed for.
The trade-off is important. The rule-based version is always right or always wrong, and you know which. The AI version is usually right, occasionally wrong, and you have to check.
FIGURE 1: TWO WAYS SOFTWARE PRODUCES AN ANSWER
Rules written by a person
- Predictable and explainable
- Right or wrong in the same way every time
- Breaks on anything unanticipated
Patterns learned from data
- Handles cases nobody programmed for
- Usually right, sometimes wrong
- Needs a person to check the output
The kinds you will meet
Three types come up in business software. They do different things.
Extraction
Pulling structured information out of unstructured documents. Reading a scanned invoice and identifying the supplier, date, amounts and line items.
Mature and reliable. This is the AI most likely to save your business real time today, and the easiest to verify — you can see whether the number is right.
Prediction
Estimating something from historical patterns. Which deals are likely to close. How much stock you will need. Which invoices will be paid late.
Useful when you have enough history. Its accuracy depends entirely on the quality and quantity of your past data. A prediction from two hundred records is a guess with a confident interface.
Generation
Producing text. Drafting an email, summarising a long record, writing a product description.
Fast and needs review. It produces fluent output, and fluent is not the same as correct. Everything generated should be read before it goes to a customer.
FIGURE 2: THREE KINDS OF AI YOU WILL ACTUALLY ENCOUNTER
Extraction
- Reads documents and pulls out structured data. The most reliable, and the easiest to check.
Prediction
- Estimates outcomes from your history. Only as good as the data behind it.
Generation
- Drafts text. Fast, fluent, and always needs a human read before it goes out.
What it is not
Three corrections worth making, because they drive bad decisions.
It does not understand your business. It recognises patterns. An AI that predicts your busiest month has not grasped that you sell umbrellas — it has noticed a shape in the numbers.
It is not always right. It produces a most-likely answer. On routine cases it is very good. On unusual ones it can be confidently wrong, which is more dangerous than being obviously wrong.
It does not remove the need for good data. It makes bad data worse, faster. If your product costs are wrong, an AI margin forecast is a wrong number with more authority behind it.
The question that matters
Not “does this have AI?” but: what does it do, and how would I know if it were wrong?
Good answer: “It reads supplier invoices and fills in the vendor, date and amounts. You review the draft before posting. It is right most of the time and you catch the rest.”
Poor answer: “It uses AI to optimise your operations.”
The first tells you what happens and where a human checks. The second tells you nothing.
FIGURE 3: FOUR QUESTIONS TO ASK ABOUT ANY AI FEATURE
What exactly does it do?
- A specific task, in a sentence. Not “optimises” or “enhances”.
Where does a person check it?
- Every useful AI feature has a review step. If it has none, ask why.
What data does it need?
- Prediction needs history. Without enough, it is a confident guess.
What happens when it is wrong?
- A draft you correct is fine. An automatic posting is not.
Where your data goes
A practical question that gets skipped.
Some AI runs on the vendor’s servers, which means your data is sent there. Some runs on your own infrastructure.
For most business data this is acceptable, and it is the same decision you already made when choosing cloud software. But you should know which it is, particularly for anything sensitive — payroll, customer personal data, contracts.
Ask the vendor directly: is our data sent anywhere, is it retained, and is it used to train their models?
Where it earns its place
The pattern is consistent across businesses.
AI is most valuable on work that is repetitive, high-volume, and easy to verify.
Reading invoices is a good example — it happens constantly, it is dull, and you can see immediately whether the number is right.
It is least valuable on work that is rare, high-stakes, or hard to check. Deciding whether to extend credit to a major customer is not a task to hand over, because the cost of a confident wrong answer is high and the mistake is not obvious.
What this means for Odoo
Odoo now includes AI across the system rather than as a separate add-on. Reading documents, drafting text, summarising records, automating repetitive steps.
The structural advantage is what it can see. Because Odoo runs on one database, the AI has the customer’s orders, invoices, payment behaviour and support history in view at once. An AI bolted onto a single disconnected tool only ever sees one slice.
The following articles in this series go through the specific uses one at a time.
The short version
AI is pattern-based software. It is genuinely good at reading documents, reasonably good at predicting from data you have plenty of, and fast at drafting text that someone still needs to read.
Treat it as a capable assistant whose work you check — not as an oracle, and not as a marketing word.
Judge every AI feature by one question: what does it do, and how would you know if it were wrong?
Wondering where AI would actually help in your business?
Get in touch. We will look at where your team spends time on repetitive work — that is where AI pays, and it is rarely where the marketing points.
