We had the opportunity to test drive the Predicta Forecasting application over the last couple of weeks. This is an Excel add-in software available as off-the-shelf (or more appropriately off-the-net) package for forecast modeling and simulation of safety stock calculations.
Predicta gracioulsy offered us an Enterprise version for this evaluation and testing.
Right off the gate, the big advantage is it is an application that easily plugs into your Excel 2010 (the company says it is compatible with Excel 2007 and up).
The look and feel is great. The charts are very pretty and it provides a variety of statistics if you are an intensive modeler trying to develop models all day. What we found interesting was the software also provided Theil’s U-statistic in addition to the commonly reported error metrics such as MPE, MAPE, RMSE and R-squared.
The engine behind the tool provides popular statistical modeling options such as exponential smoothing, double exponential smoothing, Holt-Winters methods, and Regression methods. The reporting for analysis is pretty detailed and the graphs are Excel graphs so adjustments is easy – we just did not find an easy way to unhide the tab to manipulate the underlying data that is used in the graph.
The Predicta data-application-model is driven by Excel as your main spreadsheet while Predicta is an Excel Add-on that sits with buttons available for what you want to do as an additional menu item. You can see all the models and the functional options all available as buttons:
- Exponential Smoothing (Modeling)
- Holt-Winters (Modeling)
- One-Click (creates a forecast on one-click)
- Best-fit (picks the best model among a list of options)
- Safety stock calculations (function)
Contrast this with Forecastpro, another popular Excel package available in the market. Their data-application-model is where Excel just acts as a data provider while you work completely in the software. You change models, change parameters and test different things out and you only go back to excel if you want to reload the data and load back the results.
So there are pros and cons with these approaches. While you work completely in Excel in Predicta, this can also be tedious to do modeling since it creates sheets and sheets of excel output each time you change a model or a minor parameter adjustment.
Predicta says the Enterprise version can do modeling of unlimited number of SKUs. However, we could not load more than one SKU while using the best fit method and not more than three SKUs while using the Holt-Winters method.
The one click is the only option that allows you to model multiple SKUs at the same time. However, the challenge here is you have to select a particular method ahead of time. The software does not pick the best fit of all available models. So for batch forecasting, you will be forced to use the one-click method but this depends on the planner’s prior knowledge of the data and demand profile.
In our forecast comparisons, the software did not yield the most efficient Holt Winters forecasts. While researching this, the Product manager admitted that the parameter search is using a grid that is only .10 apart. The company promised that they will be improving this shortly so more granular parameters can be estimated so this will improve the efficiency of the estimates.
The strangest thing was – the forecast horizon was limited to just one period when picking the simpler models – exponential smoothing, Moving Average etc. Although we know that the forecast will be a constant, this is an inconvenience for the planner not to get the forecasts copied over for the horizon they want.
Given this is the first release of the software, more improvements are imperative to make this competitive with other tools in the space. The biggest challenge I see is the data management – where the software produces tons of data for every little model test intead of over-writing existing outputs.
We also noticed that there is no methodology to address SKUs with intermittent demand.
1. I think it will be better to think of the average customer as a planner that is looking for results rather than a rocket scientist who will like to admire the beauty of the models and the charts all day.
2. Provide an user option to either over-write or keep existing results when running another iteration of the model.
3. Provide a best fit or an expert engine option that suggests only one forecast and results instead of throwing off multiple models and identifying the best fit. Forecasters are looking for a good forecast along with statistics that substantiate the best fit.
4. The one-click model should be intelligent enough to provide the best fit in each case instead of requesting the planner to pick one model option for all the SKUs that are modeled in one-click. This is makes the one click unusable.
5. Allow multiple skus to be picked when using the best fit or the single model. I understand this may cause performance issues but that is a challenge that needs to be addressed.
A nice thing about Predicta is it gives a clean calculation of safety stock, re-order point and Min-Max in one shot. Although these are not optimized for dynamic lot sizes and variable lead times, this is a good start and gives a good feel for the impact of variability on your inventory parameters. The safety stock is calculated using the standard deviation of demand (instead of in-sample forecast error) and the lead time demand is just the forecasted demand from the model that is selected.
We would like to thank Predicta for making the software available to us to do this testing. I hope the company can improve the software substantially before the next release.
You can find more info on the software and purchase a copy at http://predictasolutions.com/.