Sales and demand forecast · Panadería La Espiga
Example business: Panadería La Espiga — a bakery and café with daily sales and strong weekly seasonality.
The ten sections organize this illustrative example. Each product report has its own scope and format.
Analysis overview
Variable, frequency, cut-off date and horizon.
- Forecast variable
- Weekly net sales (USD)
- Frequency
- Weekly, aggregated from daily sales
- Cut-off date
- June 29, 2025
- Horizon
- 4 weeks (through July 27, 2025)
Executive summary
Main forecast and relevant warnings.
The central forecast for the next 4 weeks is USD 42,100 in sales, with a 90% prediction interval between USD 37,600 and USD 46,800. The method beat the benchmark in validation (7.8% error versus 12.4%), and the series shows a mild upward trend with stable weekly seasonality.
- Weekends concentrate 46% of sales: the daily forecast in the annex reflects this.
- Warning: the July 20 holiday has no recent equivalent and adds uncertainty to that week.
- The interval, not the central value, is the recommended basis for purchasing decisions.
Historical quality and sufficiency
Gaps, changes and treatment applied.
The history covers 78 continuous weeks. Two maintenance closures (3 days in total) were treated as missing and imputed with the same-weekday average of the 4 neighboring weeks. There are no general price changes within the history that break the series.
- 546 sales days; 3 imputed (0.5%).
- Weekly frequency chosen for the 4-week horizon: the daily one is delivered as a breakdown.
- Sufficient coverage: 78 weeks of history for a 4-week horizon.
Historical behavior
Evolution, patterns and variability.
The last 12 weeks average USD 9,320 with mild, sustained growth. Weekly variability is moderate (a 6.9% deviation over the mean), and the intra-week pattern is stable: Saturdays sell 2.1 times a Tuesday.
Weekly sales (last 12 weeks)
Weekly net sales series before the cut-off.
| # | History |
|---|---|
| W-12 | 8.1K |
| W-11 | 8.4K |
| W-10 | 8.2K |
| W-9 | 8.9K |
| W-8 | 9.1K |
| W-7 | 8.7K |
| W-6 | 9.4K |
| W-5 | 9.2K |
| W-4 | 9.8K |
| W-3 | 9.6K |
| W-2 | 10.1K |
| W-1 | 10.3K |
Explore with the pointer or keyboard. Tap a value to keep its detail visible.
Temporal validation
Model comparison and performance against a reference method.
Validation held out the final 8 weeks of history: the forecast was computed without seeing them and compared against what actually happened. The Valerian engine roughly halved the error versus repeating the same week of the previous year adjusted for trend.
Forecast
Charts, values per period and prediction intervals.
Each week’s central value comes with its 90% prediction interval. The horizon total is USD 42,100 (between 37,600 and 46,800).
History and forecast with interval
Last 12 observed weeks and 4 projected weeks with a 90% band.
| # | History | Forecast |
|---|---|---|
| W-12 | 8.1K | |
| W-11 | 8.4K | |
| W-10 | 8.2K | |
| W-9 | 8.9K | |
| W-8 | 9.1K | |
| W-7 | 8.7K | |
| W-6 | 9.4K | |
| W-5 | 9.2K | |
| W-4 | 9.8K | |
| W-3 | 9.6K | |
| W-2 | 10.1K | |
| W-1 | 10.3K | |
| W+1 | 10.4K | |
| W+2 | 10.6K | |
| W+3 | 10.2K | |
| W+4 | 10.9K |
Explore with the pointer or keyboard. Tap a value to keep its detail visible.
- History
- Forecast
- Likely range
Breakdown and limitations
Results per product or channel when feasible, and conditions that could alter the forecast.
The category breakdown is feasible for the three main lines; minor categories lack sufficient history and are grouped. The forecast assumes stable conditions: a heat wave, a road closure or a price change would partially invalidate it.
- The July 14–20 week includes the holiday with no recent equivalent: its interval is the widest.
- The forecast does not incorporate planned promotions; if launched, it must be regenerated.
AI analysis
Interpretation of the forecast and operational implications, without presenting estimates as certainties.
Key findings
Demand grows beyond seasonality
The last 12 weeks grow 2.1% monthly once the weekly pattern is removed. The forecast projects that trend continuing, not a level shift.
Evidence: Historical series (element 4) and the annex decomposition.
The method is reliable for weekly purchasing
A 7.8% mean error with 92% interval coverage in validation is sufficient for planning short-shelf-life input purchases.
Evidence: Validation table (element 5).
Horizon risk concentrates in one week
The holiday week has the widest interval (USD 8,900–11,500). Treating it as an operational exception lowers the cost of being wrong in the other three.
Evidence: Weekly forecast (element 6) and limitations (element 7).
Business interpretation
For purchasing and production, plan perishables on the conservative scenario and long-life inputs on the expected one: stock-out and waste costs are not symmetric. These figures are estimates with declared uncertainty, not sales commitments.
Hypotheses and limitations
- Recent growth may be tied to the neighboring office opening in April; if that demand is stable the trend holds, but there is no data to isolate its effect.
- The holiday’s behavior was estimated from similar-profile 2024 holidays; that week’s error may exceed the validation average.
Prioritized recommendations
Buy perishables on the conservative scenario
Use the lower band (USD 37,600 total) for dairy and fruit, adjusting mid-week with actual accumulated sales.
Special plan for the holiday week
Define holiday production separately with the team, using the analogous 2024 holiday as the reference and the interval as the bound.
Record promotions and events alongside sales
Logging promotions, holidays and events in the sales file lets future forecasts incorporate them and cut error in atypical weeks.
Proposed follow-up
Follow-up plan
Decisions to evaluate, indicators and update timing.
Methodology and annexes
Model, parameters, metrics and projection tables.
Valerian forecasting engine, calibrated with temporal validation over 8 held-out weeks and compared with repeating last year’s week. 90% prediction intervals computed from validation residuals. MAPE = mean absolute percentage error.
- Annex A: daily forecast by category for the 4 weeks.
- Annex B: validation residuals by week.
- Annex C: missing-day treatment and imputations.