Forecasting for SMEs: start small, grow smart

Every business already forecasts. When you decide how much stock to order, how many people to roster for next month, or whether cash will stretch to the next big invoice, you are making a prediction about the future. The only question is whether that prediction is a gut feeling scribbled on the back of an envelope, or something a little more grounded. Forecasting, done well, is simply the discipline of turning the numbers you already have into a structured estimate of what comes next.

The good news for smaller organisations is that you do not need a data science team, a big budget or a fashionable algorithm to start. You need a clear question, some history, and the honesty to check whether your forecast is actually any good. This article walks through where forecasting helps, how to begin with something simple you can trust, and how to grow from there without over-engineering it.

What forecasting is, and where it helps

A forecast is an estimate of a future value, produced from a repeatable method rather than a hunch. That repeatability is the point: it means you can check it, improve it, and hand it to someone else. For most SMEs, the same handful of questions come up again and again.

• Demand: how many units, orders or bookings should we expect next week or next quarter?
• Revenue: what is a realistic top line for the coming months, given how we are trending?
• Cash flow: when will money actually land and leave, and will we stay comfortable in between?
• Stock: how much should we hold so we neither run out nor tie up cash in a full warehouse?
• Staffing: how many people do we need rostered to meet demand without paying for idle hours?

Notice that these are all decisions first and numbers second. A forecast is only useful if it changes what you do. Before building anything, be clear about the decision it will inform and how much accuracy that decision actually needs.

Start small: a baseline you trust

The most common mistake is to reach straight for a sophisticated model. Resist it. Begin with the simplest thing that could work, using the data you already have in your accounting package, till system or spreadsheet.

Two baselines will take you a surprisingly long way. The first is a moving average: take the last few periods and average them to smooth out the noise. The second is last year plus expected growth: take the same period a year ago and adjust it up or down by a sensible percentage. Both are easy to build, easy to explain, and easy to sanity-check against reality.

A simple forecast you trust beats a complex one you do not understand.

This is not a consolation prize. A baseline gives you two things at once: a genuinely usable forecast, and a yardstick. Any fancier model you build later has to prove it beats this baseline before it earns a place. If it cannot, the baseline wins and you have saved yourself a great deal of complexity.

The ingredients that matter

Whether your forecast is simple or advanced, its quality rests on a few basics that are worth getting right.

• Enough clean history. You need a run of past data long enough to show a pattern, ideally two full years or more if your business has yearly seasons. Consistent, well-labelled records matter more than volume.
• Seasonality. Most businesses have rhythms: busy Decembers, quiet summers, end-of-month spikes. A method that ignores these will be confidently wrong at exactly the moments that matter most.
• Known drivers. Promotions, price changes, public holidays, a big campaign or a one-off event all move your numbers. If you know they are coming, the forecast should account for them rather than being surprised.

Much of the real work in forecasting is not modelling at all. It is tidying the history, spotting the odd months that were distorted by a one-off, and writing down the drivers you already carry in your head.

Do not over-engineer it

It is tempting to hand the problem to a black box that spits out a number nobody can question. Avoid this, especially early on. If you cannot explain why the forecast says what it says, you cannot spot when it has gone wrong, defend it to your colleagues, or improve it. Complexity should be earned by evidence, not adopted for its own sake.

Add sophistication only when a simpler method demonstrably falls short and the extra accuracy is worth real money. A forecast that is slightly less accurate but fully understood is usually the better business tool.

This matters more now, not less. In 2026, capable AI forecasting features are built into the tools SMEs already pay for, accounting packages, inventory systems, spreadsheets, and a decent statistical forecast is a few clicks away. That is genuinely useful, but it does not change the discipline. An automated forecast is still a forecast, and it has to clear the same naive baseline before you trust it. The easier these tools make it to generate a confident-looking number, the more valuable it is to have a yardstick that tells you whether the number is actually any good.

Judge it honestly

A forecast without a scorecard is just an opinion. To know whether yours is any good, measure its error in plain terms: how far off were you, on average, in units or euros or percent? Track that over time rather than judging a single lucky or unlucky month.

The single most important habit is to compare every forecast against a naive baseline. The simplest is 'the same as last period'. For a business with clear seasons, the fairer bar is the seasonal version, 'the same period last year', so a busy December is judged against last December, not against a quiet November. Pick whichever naive rule your competitor forecast should be ashamed to lose to, and treat it as the line to clear. If a model cannot consistently beat that, it is not worth the effort of running it, however clever it looks. Beating the naive benchmark is what tells you the forecast is adding real value.

Then keep score. Log what you predicted and what actually happened, and review the gap regularly. Accuracy tracked over time is how a forecast earns trust, and it also tells you honestly when the world has shifted and the method needs rethinking.

Keep a human in the loop

A forecast is an input to a decision, not the decision itself. The model does not know that a key customer just signalled they are leaving, or that a competitor is about to close down the road. People do. The right pattern is that the model provides a grounded starting point and flags what it sees, and a person applies judgement about what it cannot know. Numbers plus human sense beats either one alone.

A maturity path: where to start

You do not have to arrive fully formed. Forecasting maturity is a path you walk one sensible step at a time.

1. 1Gut feeling. Decisions rest on experience alone. It works until it does not, and it cannot be checked.
2. 2A simple baseline. Build a moving average or last-year-plus-growth forecast for one important number, in the tools you already have.
3. 3Measured baseline. Start logging forecast versus actual, and compare against the naive benchmark so you know how good you really are.
4. 4Seasonality and drivers. Fold in your known patterns and events so the forecast reflects the business you actually run.
5. 5Fit-for-purpose models. Only where the evidence justifies it, introduce a more capable model, and keep measuring it against the baseline and keeping a human in the loop.

Pick the one number where a better estimate would most change your decisions, and start at step two this week. A modest, honest forecast that you actually use and check will do more for the business than an ambitious one that never leaves the drawing board.