Revenue forecasts that purchasing and budgeting can rely on.
With an honest range.
Dometrics builds forecasting models from shop, POS and marketing data: revenue and demand for the coming weeks and months, with seasonality, promotions and ad budget as drivers. Visible in the dashboard, tested against reality.
| Category | Forecast Q1 | Range | Trend |
|---|---|---|---|
| Portafilter machines | €468k | ± 7% | ↑ 9% |
| Accessories | €301k | ± 10% | ↑ 4% |
| Coffee subscription | €339k | ± 5% | ↑ 14% |
Why planning often misses.
Most plans start with last year and a mark-up. That works as long as nothing happens. But something always happens.
Last year plus a percentage
Holidays shift, promotions fall differently, the weather does not cooperate. A plan that only looks at last year explains no deviation and learns nothing from it.
Stock too high or too low
Too much stock ties up capital and ends up in clearance sales, too little costs revenue in the strongest week. Both start with an inaccurate demand estimate.
Budget follows gut feeling
How much ad budget makes sense next month is estimated. The relationship between budget and revenue is in the data, but goes unused.
One number, no risk
Plans state a number but no uncertainty. Whether a target is likely or ambitious remains unclear until it is too late.
From historical data to a reliable range.
A forecast is only as good as its validation. That is why every model is first tested against the past before it makes any statement about the future.
- 01 · Data
Consolidate history
Revenue, orders and items from the shop or POS, ad spend per channel, prices, promotions, public holidays.
- 02 · Cleansing
Flag special cases
Promotions, supply outages and outliers are flagged so that the model classifies them correctly.
- 03 · Model
Calculate the forecast
Time series models with seasonality and drivers, validated in a backtest against known months.
- 04 · Scenarios
What if
More budget, different prices, an earlier promotion: scenarios show the effect in the dashboard, with a range.
- 05 · Reconciliation
Forecast versus reality
Every month, the gap between forecast and result is measured. The model keeps learning.
Prior-year planning versus forecasting model.
| Criterion | Last year plus a percentage | Forecasting model by Dometrics |
|---|---|---|
| Basis | One comparison period | Several years of history with seasonal patterns |
| Drivers | Estimated or ignored | Budget, prices, promotions and holidays in the model |
| Output | A single number | Expected value with a range |
| Quality | Unknown until month-end | Forecast error measured in the backtest and reported |
| Updates | Once a year | Continuously with every new day of data |
What a good revenue forecast has to deliver.
A revenue forecast is not a crystal ball. It is a structured answer to the question of what is likely to happen if past patterns continue, and how known influences affect that. Its value lies less in the individual number than in being honest about uncertainty.
A range instead of a precise landing
Every forecast is uncertain. Good forecasts state how uncertain. Instead of “€412,000 revenue in March”, the statement reads something like “in March, very likely between €380,000 and €440,000, expected value €412,000”. Purchasing cares about the lower bound, capacity planning about the upper bound, controlling about the expected value. A number without a range withholds precisely the information that safeguards decisions.
Which data feeds the model
- History: revenue, orders and items on a daily basis from the shop, POS or inventory management system.
- Calendar: public holidays, school holidays, bridge days and industry-specific events.
- Marketing: ad spend per channel, newsletter sends, promotions and discounts.
- Prices and range: price changes, new launches, delisted items.
- Optional external data: weather, market data or search interest, if they demonstrably add explanatory power.
Method: the backtest decides
There are proven methods for time series forecasting, from exponential smoothing and ARIMA to regression and machine learning models [1]. Which of them works best for a given company cannot be said in advance, only measured. In the backtest, each method forecasts past months as if they were the future. The errors are compared, and the method with the best accuracy wins. Often it is not the most complex one.
Good to know: A widely used error measure is the mean absolute percentage error, or MAPE. It indicates by what percentage the forecast was off on average. If a provider does not report such a value, the quality of its forecast is unknown.
Where the forecast makes a difference
| Area | Decision | Benefit of the forecast |
|---|---|---|
| Purchasing and stock | Order quantities, safety stock | Fewer stock-outs and less clearance |
| Marketing | Monthly budget per channel | Budget where demand is expected |
| Staffing | Shift planning, seasonal staff | Staffing matched to expected workload |
| Finance | Liquidity planning | Early warning of foreseeable deviations |
Key point: A forecast does not replace a decision. It shows which assumptions a decision makes and how risky it is.
Scenarios involving ad budget require reliable measurement, as described under GDPR-compliant conversion tracking. In sales, lead scoring builds on the same data. In Lower Austria, data projects of this kind are eligible for funding through digi4Wirtschaft, the regional digitalisation programme.
What the forecasting project includes.
- Data connection to shop, POS, inventory management and ad accounts
- Cleansing of historical data and flagging of special cases
- Forecasting model at total, category or item level
- Backtest with reported forecast error
- Dashboard with forecast, range and scenarios
- Automatic updates with new data
- Monthly reconciliation and model maintenance in ongoing operation
How the project runs
Data check
Review of which history is available and in what quality. Result: forecast level and expected accuracy.
Model
Build, backtest and compare several methods. The model with the lowest error wins, not the most complex one.
Dashboard
Forecast, range and scenarios are integrated into the existing reporting.
Operation
Ongoing updates, monthly reconciliation, refinement when new drivers emerge.
Frequently asked questions
How accurate is an AI revenue forecast?
That depends on the volume of data, the stability of the business and the forecast horizon. Instead of making promises, Dometrics measures accuracy in a backtest: the model forecasts past months as if they were the future, and the error is reported. This shows before deployment how far the forecast can be trusted.
How much data does a revenue forecast need?
For seasonal patterns, at least two, ideally three years of daily history make sense. Forecasts are possible with a shorter history, but with a wider range. Completeness matters more than volume: gaps and unflagged promotions distort any model.
Which methods are used?
Proven time series methods such as exponential smoothing and ARIMA, regression models with drivers, and machine learning methods such as gradient boosting. The backtest decides which method is used, not the latest trend.
Can the forecast also take ad budget into account?
Yes. Ad spend per channel is included in the model as a driver. This makes it possible to calculate scenarios, for example how a higher budget next month is likely to affect revenue and the range. The prerequisite is clean measurement of spend and revenue.
Which companies benefit from a forecast?
Companies whose decisions depend on future demand: eCommerce and retail for purchasing and stock, hospitality for staffing and pricing, manufacturing for capacity. The more expensive a planning error, the faster the model pays for itself.
Does the model replace planning within the company?
No. The model provides a reliable baseline and shows which assumptions a plan makes. Targets, strategy and decisions remain with the company. The gain is that deviations become visible early.
