I'm not familiar with the POC you mentioned but in my experience, regression methods are well suited in Anaplan as long as you build the auditing and business rules in Anaplan. For example, I used a seasonal ARIMA model (Holt-Winters) to detect trend for a retail client. I had to set up a lot of exception flags in Anaplan so the forecasts didn't fly all over the place. Each week, new data would be loaded and the forecasts would adapt to the new information. The exception flags (and successive trips) help identify the nuances (like when a new store opens, a new product is introduced, store was closed due to weather, consecutive errors in the same direction - that sort of thing.)
If you are using some type of deep learning method, then you'll probably want to do that outside of Anaplan. I use Knime which is an open-source ML tool.
Classification methods, like clustering, seems better suited outside of Anaplan because of the iterations that need to take place. I usually use Python or R. And since Anaplan's API set can be called from either of these, so it's very convenient. You can even put them on a scheduler to improve the automation.
You can go pretty far with this. But hooking up Anaplan to your statistical models is not that difficult - you'll just need to make an investment of time to learn the data integration with Anaplan. Good news is that Anaplan's documentation is excellent, not to mention this Community Site. Most likely, someone has already tried to do what you're doing.
If i understood it correctly then in summary, one can setup extensive models on ML platforms such as Tensor flow ( offered via google cloud) and then have Anaplan model integrate with tensor to feed Anaplan directly with key data points.
Does Anaplan currently on its platform offer any in built ML capabilities ? Can you point to any reference links that demonstrates how ML and AI integrated with Anaplan is advantageous over the traditional FP&A?
Yes, you can integrate tensorflow with Anaplan but because tensorflow doesn't have a built-in connector you'll have to do one of the following.
- create the API calls in tensorflow to import the data into Anaplan. I believe tensorflow is just a python library so if you have a python environment set up, you can just call the APIs directly from python. You just need to install requests, base64, sys, and json libraries as well. My all time favorite best practice post on data integration with python is this one from @chase.hippen
- Use an ETL tool that has a built-in Anaplan connector like Informatica, Mulesoft, Dell Boomi and control the data flow from there.
- Utilize Anaplan Connect which is a free integration tool provided by Anaplan. This will require you to export your data from tensor first.
- manually load the data yourself.
To answer your other question, Anaplan doesn't currently have anything as robust and ML feature rich like tensorflow but I wouldn't be surprised if they're working on it. Just keep in mind that Anaplan is first a dimensional planning framework. It's strength is in connecting different planning applications together by utilizing the same data. Apps just pass their results back and forth. You still need to build the math in Anaplan which you can do but with standard libraries like tensorflow it just seems easier for now to use that and import the results into Anaplan.