Predictive analytics for SMEs: you do not need to be Amazon

"Predictive analytics" sounds like a multinational's data department, with data scientists coding complex models only Amazon or Netflix could afford. And it is true that the most sophisticated prediction systems do require that level of resources, but a fairly substantial part of what makes predictive analytics useful (anticipating what is going to happen before it happens, instead of just describing what already happened) is available today, with considerably less effort than it looks, to any SME with minimally organised data.

The difference between descriptive and predictive analytics, in simple terms

The analytics most businesses use today is descriptive: it counts what has already happened (how many visits there were last month, how many sales closed). Predictive analytics takes it a step further: it uses that same historical data to estimate what is likely to happen in the future, whether next week, next month or next quarter. It is not about guessing the future with absolute certainty, but about reducing uncertainty enough to make better decisions ahead of time.

Audience predictions in Google Analytics: the most accessible entry point

Google Analytics 4 includes, at no extra cost and with no complex setup, machine-learning-based predictive metrics: purchase probability (which active users are most likely to complete a purchase in the coming days), churn probability (which active users are least likely to interact with the site again), and predicted revenue. These metrics are generated automatically if there is enough historical data volume, and can be used directly to create remarketing audiences (for example, targeting a special offer at people with high purchase probability who have not yet converted).

Demand forecasting with your own historical data, no special tool required

A business with at least two or three years of monthly sales data can identify fairly clear seasonal patterns using nothing more sophisticated than a spreadsheet: which months are consistently stronger or weaker, whether there is a sustained growth or decline trend over time, and how much each year deviates from that general pattern. This form of forecasting, though rudimentary compared to an advanced statistical model, is surprisingly useful for practical decisions like when to build up stock, when to hire seasonal staff, or when to launch an acquisition campaign.

Predictive lead scoring: prioritising without guessing blindly

Many CRMs, even ones aimed at SMEs, include some form of lead scoring that combines behavioural signals (pages visited, response time to an email, email opens) to estimate which contacts are most likely to become customers. You do not need a complex AI model for this to add value: even a simple formula, based on the sales team's own accumulated experience of which signals have predicted sales in the past, already notably improves prioritisation compared to treating all leads equally.

Predicting churn among recurring customers

For businesses with customers who buy or subscribe recurrently (subscriptions, maintenance services, consumables), analysing the historical behaviour pattern before a customer cancels (a noticeable drop in usage frequency, say, or no response to the last few communications) lets you build fairly simple early-warning signals: if a current customer starts showing that same pattern, you can intervene proactively (a call, a loyalty offer) before the cancellation happens, instead of reacting afterward.

Generative AI tools as an accelerator for this kind of analysis

With no need to code or have advanced statistics knowledge, generative AI tools can analyse an exported dataset (a sales history in a spreadsheet, for example) and suggest patterns, seasonality or correlations that would go unnoticed at first glance. It does not replace rigorous analysis by a specialist, but it democratises the first level of pattern detection for businesses that previously had no practical access to this kind of capability at all.

The real limits: what accessible predictive analytics cannot deliver

It is important to be realistic about what these accessible tools can and cannot do. They do not replace a rigorous statistical model custom-built for the business, they work better the more volume and history of data is available (with very little data, predictions are unreliable), and no prediction, however good the tool, removes uncertainty entirely: they are an aid to deciding with better information, not a guarantee of being right.

An example of demand forecasting done with a spreadsheet

A pool supplies shop had four years of monthly sales data. Plotting that data on a simple chart made a pattern crystal clear that the owner "knew" intuitively but had never quantified: March sales fairly reliably predicted the total volume of the summer season, a correlation that repeated year after year. From that pattern, the shop started using March sales as an early signal for deciding how far in advance to build up stock and hire seasonal staff, instead of waiting until June to react on already-late data. No paid software or complex statistical model was needed, just carefully looking at four years of data that already existed and that nobody had cross-referenced that way before.

A common mistake: confusing correlation with causation

When working with patterns detected in your own data, it is easy to fall into the temptation of assuming two figures moving together have a cause-and-effect relationship, when in reality both could be responding to a third, unconsidered factor. A business that detects its sales rise when its social media spend also rises might wrongly conclude social media is causing those sales, when in reality both things could simply be rising because they coincide with the business's high season. Before making important decisions based on a detected pattern, it is worth checking whether an equally plausible alternative explanation exists.

Start with one specific question, not with "doing predictive analytics"

The most common mistake when trying to introduce predictive analytics into a small business is framing it as an abstract project ("let's do predictive analytics") instead of starting from a very specific business question that already exists unanswered ("when should I start hiring extra staff for Christmas?", "which customers are most at risk of not renewing this quarter?"). Starting from the specific question forces you to use only the data and tools genuinely needed to answer it, instead of trying to build a general predictive system that ends up too ambitious for the time and resources available.

Step by step: building your first seasonality analysis with a spreadsheet

First, export at least the last two or three years of monthly sales or revenue into a spreadsheet, with one column per month and one row per year. Second, create a line chart with one year per series, overlapping the months on the horizontal axis: this lets you see at a glance whether the pattern repeats year to year or is erratic. Third, calculate the correlation between each month and the total for the following quarter or season (most spreadsheets have a built-in correlation function), to identify objectively, not just visually, which month best anticipates the later result. Fourth, once the most predictive month or months are identified, define a concrete threshold ("if March sales exceed X, build up stock in April") that turns the detected pattern into a clear operational decision. Fifth, review the pattern every year with new data, because business or market conditions can change and a pattern that worked for several years can stop being reliable without warning. This exercise, requiring no programming or advanced statistics knowledge, is the most accessible entry point into predictive analytics for any SME.

Frequently asked questions