The small business problem

Most writing about AI in business assumes resources you do not have.

It assumes years of clean historical data. A person who can own the project. Budget for consultants. Volume high enough that a small percentage improvement is real money.

A twenty-person company has none of that in the same measure. So the useful question is narrower: what actually pays for itself at your size?

The answer is a short list, and it is not the exciting part of the technology.

What works at small scale

Three things. All of them are about volume of dull work, not sophistication.

1. Reading supplier documents

Why it works small. Even at fifty bills a month, that is real hours of typing. And you can verify it instantly — the document is right there.

What you need. Clean vendor records. That is it.

Realistic saving. A few hours a month at fifty bills; considerably more at two hundred.

2. Drafting text

Product descriptions, routine emails, knowledge base articles.

Why it works small. It removes the blank page, which is the expensive part when one person does everything.

What you need. Nothing. It works from your product data.

The catch. Read every word. At small scale you probably will, because it is one person and they care.

3. Summarising before a conversation

A customer’s history condensed before a call.

Why it works small. In a small company, whoever picks up the phone is often not the person who dealt with them last. This closes that gap.

What does not work at small scale

Being direct here saves you money.

Sales prediction. Needs thousands of closed deals. If you close forty a year, a score is a guess with a confident interface.

Lead scoring. Same problem. If you get thirty enquiries a month, sort them yourself in five minutes.

Demand forecasting. Needs years of history and stable patterns. A business winning four large contracts a year has no pattern to learn.

Recommendations from a model. Under a few thousand orders, a hand-set list of “these go together” is more accurate. Not a lesser option — genuinely more accurate.

Anomaly detection. Needs enough transactions to know what normal looks like. At low volume it flags everything, and you stop reading the flags.

The thing that matters more than any of it

Your data.

This is more true at small scale, not less, because you have less of it. Every duplicate customer is a larger proportion of your history.

Four things, and they cost nothing but attention:

Merge duplicate customers. “ABC Trading” and “ABC Trading Ltd” are one company.

Fix product costs. Every margin figure depends on these. In small businesses they are frequently years out of date.

Make units of measure consistent. Decide the rule and apply it.

Agree what revenue means. Invoiced, ordered or delivered. Write it down.

A week spent on this is worth more than any AI feature you could switch on. It is also useful immediately, whether or not you adopt anything.

A realistic plan

Six steps, in order, over a few months.

1. Fix your data. A week. Duplicates, costs, units, definitions.

2. Measure a baseline. How long does bill entry take? How many bills a month? Write the numbers down.

3. Turn on document reading. Set up an email alias for supplier bills.

4. Run it for a month with review. Note where it gets things wrong — usually tax codes and account allocation.

5. Compare against your baseline. Did the hours actually drop? If not, find out why before adding anything else.

6. Add drafting and summaries if step 5 was positive.

Then stop and use it for six months before considering anything else.

What to be careful with

Buying a tool because a competitor mentioned it. The question is what it does for your specific costs.

Adopting several things at once. You will not know what helped.

Removing review to save more time. At small scale one duplicate payment can exceed a year of saved hours.

Paying for capability you cannot feed. Forecasting on eight months of data produces a confident number with no basis.

What it costs

Worth being straight about, because budgets are real.

Odoo Enterprise is a per-user monthly cost, and AI features are increasingly part of it rather than a separate product. For a small team the licence is a modest line.

Implementation help is usually the larger cost, and it is the one worth spending on — mostly on getting the data and configuration right rather than on features.

Your own time is the hidden cost. Someone has to make decisions and check output. In a small company that is likely you.

The realistic saving is hours per week, not a headcount. At small scale that time usually goes into work that was not getting done rather than into savings on a payroll line.

The advantage you actually have

One thing small businesses do better than large ones, and it is worth knowing.

You can fix your data. A twenty-person company with three thousand customers can clean its records in a week. A large enterprise with two million cannot.

Since data quality is the main determinant of whether any of this works, that is a genuine advantage — and it is available to you now, for the cost of some attention.

The short version

At small scale, AI pays for reading documents, drafting text and summarising history. Those three are worth having and they work at any volume.

It does not pay for prediction, scoring or forecasting, because those need data volumes you do not have — and they will produce confident numbers anyway.

Fix your data first. It costs a week, it helps immediately, and it decides whether anything else you try is worth reading.