# Wizard Guide
Every screen of the PowerForecast wizard, in order.
The shell
| Sidebar | Does |
|---|---|
| Forecast Wizard | the 4 steps |
| Results | the Results page of the last run (see the User Guide) |
| Forecasts | every forecast ever run, kept by PowerForecast (see below) |
| File operations | Upload a forecast specification (.json), Download the current one (name in the box below) |
| Reset Wizard | clears everything, including the kept draft |
| Documentation | this site |
| Theme · language | violet or amber; English, French, German, Spanish, Dutch |
Switching between Forecast Wizard and Results never loses your work: the wizard keeps a draft and reopens on the step you left.
Every step has an AI assist panel at the top: describe what you want in your own words, click Draft with AI, check the drafted fields, then Apply to fields. Everything stays editable. The questions and the fields come from the PowerForecast template in PowerAdmin.
Step 1 · Goal
Start from an Excel workbook. Choose workbook… reads a workbook with:
| Sheet | Holds |
|---|---|
| Setup (optional) | Setting | Value: What to forecast, Name, Description, How far ahead, Time unit, Likely range, Scenario, Method, Split by |
| Data | the history: the date in the first column, then one column per series (name in the first row). Or three columns Series | Date | Value |
| How to fill | what each sheet holds; never read |
Every line of Setup is optional; what is missing stays for you to set. If the workbook has a problem (a cell that is not a date or a number, a date twice in a series), the panel lists each one with its sheet, row and column, and nothing is changed. Save as workbook writes the current forecast (settings and data) as such a workbook: a good way to start your own.
What do you want to forecast? Demand, Workload, Revenue, Claims, Transactions, or Custom (type your own word). More options holds Split by (per product, region, segment, branch, category, or over time).
Step 2 · Your data
Drop an Excel (.xlsx), CSV or JSON file, or click to browse. The layouts:
| Layout | Example |
|---|---|
| One column per series (recommended) | Date, Shoes, Bags then 2024-01-01, 120, 48 |
| Long | unique_id, ds, y (or Series, Date, Value), one row per series and date |
| JSON | [{ "unique_id": "Shoes", "ds": [...], "y": [...] }] |
Dates: Excel dates, 2024-01-31, 2024-01 (first of the month), 31/01/2024. Numbers: 1200, 1,200, 2,5. An Excel workbook with a Setup sheet also fills the settings it holds.
Once loaded, the step shows the number of series, values, the date range and the detected period, a preview, and any warnings (irregular gaps, different lengths, duplicate dates, short history). Rows out of date order are sorted automatically. Clear Data removes the file.
The time unit of the forecast is taken from the dates: there is no frequency question.
Step 3 · How far ahead
Type a number of periods or click a usual choice (for monthly data: 3, 6, 12, 24 months). The step says "Your data has one value per month, so the forecast does too" and warns when the horizon is more than a third of your shortest history.
Advanced (optional), closed by default:
| Setting | Choices |
|---|---|
| Time unit | hourly … yearly; change it only if the detection was wrong |
| Width of the likely range | 80 %, 90 %, 95 %, 99 % |
| Scenario | baseline, optimistic, pessimistic, stress (a label carried with the run) |
| Forecasting method | Automatic (recommended) or Manual: tick AutoARIMA, ETS, SeasonalNaive, Theta, MSTL |
Step 4 · Review
The summary in plain words (what is forecast, split by, time unit, how far ahead, likely range, scenario, method, series in your data), a name and an optional description. An AI review of the settings runs by itself (you can choose the AI model next to the button); it is advice only. Run Forecast first validates the whole specification against the template. Each problem is listed with its field and owning step. Then it starts the run.
Forecasts
Forecasts (sidebar) lists every forecast PowerForecast keeps: name, status (Done, Failed, Running), number of series, how far ahead, when it started and who ran it. Nothing expires.
| Action | Does |
|---|---|
| Results | opens the Results page of that forecast, with its history and its checks |
| Use as starting point | puts its settings and data into the wizard (step 1) to run a new forecast from it |
| Rename / JSON | a new name; the specification, result and diagnostics in one file |
| History | who ran, renamed, deleted or restored it, and when (names are the ones the browser sent; not verified yet) |
| Delete → Restore / Delete for good | a deleted forecast is hidden (tick Show deleted); only a deleted one can be removed for good, after a confirmation |
| Import files… | the files saved at the end of a run made before forecasts were kept: the results file (required), the specification and diagnostics files (optional; without the specification it cannot be used as a starting point) |
The AI Forecast Critic's trials are kept too, but only shown when Show AI critic trials is ticked.
Running
A progress window shows the elapsed time and the stages (checking your data → choosing a method → forecasting). The engine works through the series one by one: about 6 minutes for the 20 monthly series of the demo (11 years each) on one core; longer and daily series take longer.
When the run ends:
- Your forecast is ready → See the results opens the Results page.
- Four files are saved to your downloads:
forecast_specification_<run>.json,forecast_results_<run>.json,forecast_diagnostics_<run>.json,forecast_profiling_<run>.json. Keep at least the results and diagnostics files to reopen the run later.
If the engine refuses the data, the message names the problem (for example a series with too few values).
Next: the User Guide for the Results page.