ForecastSolve: demand forecasting in Excel with machine learning models
An Excel add-in that takes your sales history, runs a tournament of 23 models per item, picks the winner by measured accuracy and leaves the forecast in the same sheet you already use. No cloud, no implementation project, no data migration.
- Excel add-in for Windows
- Local engine: data never leaves your PC
- 23 models per item, automatic winner
- Monthly S&OP cycles
- One planner, one license
- From USD 400/month
What is ForecastSolve?
ForecastSolve is demand forecasting software that lives inside Excel. You load your sales history by item and customer, and the engine does what normally requires a demand planning system: it cleans the series, classifies it, tests dozens of models, picks the best one per item and saves the result as a closed S&OP cycle.
It is built for the planner who does the sales forecast in Excel today with moving averages, FORECAST.ETS or an inherited template, and who needs an auditable forecast for catalogs of hundreds to a couple of thousand items, without buying a platform or waiting months for implementation.
It is the first step of the DataSolve ladder. When more people on the team need to take part in the plan, the same engine and the same way of working continue in DemandSolve.
How it works: from data to a closed cycle
Data
Load the sales history and the item master from CSV or from your own sheet. Nothing is transformed without your approval: you choose every adjustment.
Demand
The aggregation analysis evaluates the level combinations (product, family, customer, region) and tells you where forecasting works best. Each series is classified into one of 15 categories crossing maturity and pattern: smooth, erratic, intermittent or lumpy. You flag anomalous months (stock-outs, strikes, pandemic) so they do not contaminate the model, and the substitution module splices the history of a new item with the one it replaced.
Models
A tournament of 23 models per series: exponential smoothing variants, ARIMA, intermittent-demand methods (Croston and SBA) and machine learning models. The winner is chosen by accuracy measured on data the model did not see. Eight error metrics per model and per series are stored and visible in Excel.
Forecast
The calculated forecast (what the model said) is kept apart from the adjusted one (what you decided), with full traceability. The aggregated forecast is broken down to the fine level according to each item's and customer's sales weight.
Report and S&OP cycle
Each month is frozen as a snapshot of the plan. You compare cycle against cycle and against actual sales, with error by item, family and customer, in Excel sheets you can filter, chart and share.
Why FORECAST.ETS is not enough
Excel's FORECAST.ETS function is a fine starting point: it applies a single exponential smoothing model with seasonality (Holt-Winters AAA) to one series at a time. For a real catalog it falls short for four reasons.
- One model for every item. A seasonal product, a new launch and an item with intermittent sales need different models. ForecastSolve tests 23 per item and keeps the one that forecasts that specific item best.
- It knows nothing about intermittent demand. With many zero months, exponential smoothing systematically overestimates. Croston and SBA exist for that, and ForecastSolve applies them automatically to series classified as intermittent.
- It does not measure error per item. You get the number, not how wrong you would have been last year with that method. ForecastSolve stores eight error metrics per model and series, and chooses with them.
- There is no cycle. With formulas, March's forecast disappears when you recalculate in April. ForecastSolve freezes every cycle so you can compare plan against actuals and explain the gap.
If you are looking for a demand forecasting Excel template, ForecastSolve is the next step: the same sheet, with an engine behind it.
Who it is for
One planner with Excel
You build the monthly forecast and today you do it with formulas or a template nobody else understands.
Hundreds to thousands of items
Distribution, consumer goods, pharma, hardware, spare parts: catalogs where item-by-item review no longer scales.
Data that cannot leave
The engine runs on your computer. No cloud, no third party seeing your sales.
Difficult demand
Intermittent series, items that replace each other, anomalous months. Exactly what breaks templates.
A monthly S&OP process
You need to close cycles, compare against actuals and account for the error, not just produce a number.
Before buying a platform
You want a serious forecast today and to decide with data whether you will ever need a multi-user system.
Compared with the alternatives
| Excel template | ForecastSolve | Cloud platform | |
|---|---|---|---|
| Models per item | One, whichever you can build | 23, chosen automatically | Several, often opaque |
| Intermittent demand | No | Croston and SBA, automatic | Depends on vendor |
| Error per item and model | No | 8 metrics stored | Yes |
| Comparable S&OP cycles | Manual, gets lost | Every month frozen | Yes |
| Where the data lives | Your PC | Your PC | Vendor servers |
| Implementation | None | Install the add-in | Weeks or months |
| Users | One | One | Team |
Qualitative comparison between product categories, not against a specific vendor.
What ForecastSolve does not do
We would rather you know before writing to us.
- It does not handle millions of items. The practical limit is around 2,000 series per cycle, due to computing time on a PC.
- It does not integrate with your ERP. Data comes in by CSV or from the sheet.
- It is not multi-user or collaborative. For a team there is DemandSolve.
- It does not compute safety stock or reorder points. It delivers the forecast; inventory policy is another step.
- It does not run on Mac or Excel for the web.
- It does not forecast in real time. The tournament runs once per cycle, usually overnight.
Frequently asked questions
How much does it cost and when is it available?
ForecastSolve costs from USD 400 per month per planner license. The final amount depends on catalog size and the support you need. It is in its launch phase: write to us and we will confirm terms and availability for your case.
Do I need to know statistics or programming?
No. You work in Excel as always. The engine classifies, tests models and picks the winner; you review, adjust what you know about the business and close the cycle. The error metrics are there for anyone who wants to audit them.
What data do I need to start?
A monthly sales history by item (ideally 24 months or more, though it works with less) and an item master with the hierarchies you use: family, category, customer, region. A CSV exported from the ERP is enough.
What about new items or discontinued ones?
Classification crosses pattern with maturity: a launch is not forecast like a mature item. And the substitution module lets you say which product replaced which, to splice their histories instead of starting from zero.
How long does it take to compute?
It depends on catalog size and your computer. The full tournament is demanding per item, so ForecastSolve runs the heavy models only on the series that warrant them, and it runs once per cycle, usually overnight.
How does it relate to DemandSolve?
Same approach at two scales. ForecastSolve is for one person in Excel. DemandSolve is for a team that needs to work on the same plan, with roles, approval flows and optional ERP integration. You move up when the process asks for it.
Does your forecast live in Excel? Let's talk.
Tell us how many items you manage and how you close the cycle today. We will tell you frankly whether ForecastSolve fits.
Email us →