Most small business owners plan by feel. They look at last month's revenue, think about what's in the pipeline, and make a judgment call on whether to hire, invest, or hold back. That instinct has real value — but it has limits. When growth accelerates, when costs become unpredictable, or when a major client relationship changes, gut feel stops being enough. What replaces it is forecast analysis: structured, data-driven modelling of what is likely to happen next.
The good news is that most Irish SMEs already have the data needed to build meaningful forecasts. The challenge is that it is rarely structured in a way that makes forecasting easy — and most business owners don't know where to start. This article explains how Data Sensum approaches forecast analysis engagements, and what the output looks like in practice.
What forecast analysis actually is — and isn't
Forecast analysis is not guesswork dressed up in spreadsheets. A well-built forecast is a structured model that takes historical patterns, adjusts for known future variables (a new contract, a seasonal dip, a planned cost increase), and produces a range of likely outcomes rather than a single optimistic number.
It is also not the exclusive domain of large companies with data science teams. The techniques that underpin useful business forecasting — trend analysis, seasonality adjustment, scenario modelling — are entirely achievable with Excel or Power BI for a business of any size, provided the underlying data is clean and consistently structured.
What separates a useful forecast from a useless one is not the sophistication of the tool. It is the quality of the input data, the honesty of the assumptions, and the discipline to update the model as new information comes in.
A useful rule of thumb: If your forecast is always optimistic — if actual results regularly fall below what the model predicted — the problem is almost never the model. It is usually an assumption that is being applied too generously. Useful forecasts are built to be wrong in predictable ways, not to tell you what you want to hear.
The four forecasting problems Data Sensum solves most often
1. Revenue forecasting
Revenue forecasting is the most common starting point. The goal is a rolling model that projects expected revenue over the next three, six, or twelve months — broken down by product line, client segment, or sales channel as the business requires.
We build revenue forecasts by pulling historical transaction data from accounting systems (Xero, QuickBooks, Sage) and combining it with sales pipeline data from the CRM. The model identifies seasonal patterns, adjusts for known pipeline activity, and produces a monthly projection with a confidence range based on historical variance. For businesses with a high proportion of recurring revenue, this can be extremely accurate. For project-based businesses, it is more directional — but still far more useful than an unaided estimate.
2. Demand and inventory forecasting
For businesses that hold stock — retail, wholesale, manufacturing, distribution — demand forecasting determines how much of each product or SKU to hold at any given time. Understocking costs sales and damages customer relationships. Overstocking ties up cash and creates write-off risk.
We build demand models that analyse historical sales velocity by product, incorporate seasonality, and flag which lines are trending up or down. Where a business uses a WMS or ERP system, we connect directly to the stock data. Where inventory is tracked in spreadsheets, we clean and structure the data before building the model. The output is a reorder recommendation engine: each week, the model flags which lines are approaching reorder point and recommends the quantity to order based on projected demand and supplier lead times.
3. Cash flow forecasting
Cash flow forecasting is the forecast that matters most to survival. A business can be profitable on paper while running out of cash — and that is the scenario that kills otherwise viable companies. We build rolling 13-week cash flow models that combine accounts receivable data (what is owed to the business and when it is expected), accounts payable (what the business owes and when it falls due), and a forward schedule of planned expenditure.
The model updates automatically when connected to the accounting system, so the view is always current. It highlights the weeks in which cash will be tightest and gives the management team enough lead time to act — whether that means chasing debtors earlier, delaying a non-critical purchase, or drawing on a credit facility before it becomes urgent.
4. Budget versus actuals tracking
Many businesses set an annual budget once a year and then never look at it again until the year-end review. That is not budgeting — it is a planning exercise with no feedback loop. Effective budget management requires a monthly variance report that compares actual results against the budget line by line, explains the variance, and adjusts the full-year forecast accordingly.
We build budget versus actuals models that pull directly from the accounting system and present the variance analysis in a clear, colour-coded dashboard. When actuals deviate significantly from budget — in either direction — the model automatically updates the full-year projection so the management team can see what the deviation means for the year as a whole, not just the month in question.
How we build a forecast model
Every forecast engagement begins with a data audit. Before we write a single formula, we assess the quality and completeness of the historical data available. Forecasts are only as reliable as the data they are built on — and it is better to know about a data quality problem before building the model than after.
Once the data foundation is established, we define the forecast structure: the time horizon, the granularity (by week, month, or quarter), the segmentation (by product, region, client, or channel), and the key assumptions that will drive the model. We document all assumptions explicitly — not just the numbers, but the reasoning behind them — so that the model can be revisited and updated as circumstances change.
The model itself is built to be maintained by the client's own team. We do not deliver black-box tools that only we can update. Every model includes a clearly labelled assumptions section, a data refresh process that can be run by a non-technical user, and documentation that explains how the model works and how to interpret the outputs.
A real example: A Dublin-based wholesale distributor came to us with no formal forecasting process. Their buying decisions were based on intuition and a rough rule of thumb for reorder quantities. We built a demand forecasting model connected to their Sage system that analysed two years of sales history, identified seasonality by product category, and automated weekly reorder recommendations. In the first six months, their overstock write-offs fell by 60% and their out-of-stock incidents dropped by 45%.
Choosing the right tool for your forecast
Excel with Power Query is the right choice for most SMEs. It is familiar, flexible, and does not require any additional software licences. A well-structured Excel forecast model, refreshed automatically via Power Query connections to your source systems, is reliable and maintainable by any competent Excel user. For businesses with monthly or weekly forecasting cycles and data volumes in the tens of thousands of rows, Excel handles everything required.
Power BI becomes the better option when the forecast needs to be shared across a team, when the underlying data volumes are larger, or when you want to combine the forecast with broader operational dashboards in the same interface. Power BI's DAX calculation engine handles complex time-intelligence functions — rolling averages, year-on-year comparisons, moving totals — efficiently and at scale.
Python-based models are worth considering when the forecasting requirement goes beyond what spreadsheet tools can handle — for example, when you need to model multiple interacting variables, run probabilistic simulations, or integrate machine learning-based demand signals. We use Python for these engagements, delivering the output in a format (usually a dashboard or scheduled report) that non-technical users can access and interpret without understanding the underlying code.
What to expect from a forecasting engagement with Data Sensum
We start with the free productivity audit, which includes a review of your current data landscape and a clear assessment of what forecasting is feasible and valuable given what you have. We will not recommend building a complex multi-variable model if a simpler rolling average would serve you just as well. The goal is the most useful model for your specific situation — not the most impressive one.
A typical forecast model engagement takes two to four weeks from scoping to delivery, depending on the complexity of the data sources and the number of forecast types required. We deliver the model with full documentation, a training session for the people who will maintain it, and a 30-day support window after go-live during which we resolve any issues that arise in production use.
Ready to Replace Gut Feel with a Reliable Forecast?
Our free audit identifies which forecasting models would deliver the most value for your business — no jargon, no commitment.
Free Productivity Audit →