Demo Video: Anaplan PlanIQ

Anaplan PlanIQ allows customers to use statistical artificial intelligence (AI) and machine-learning (ML)-based forecasting techniques to access accurate forecasts.

Train forecast models with AI/ML to equip users with quality metrics and harness forecast models to generate predictions.

Easily embed AI/ML forecasting into existing processes, schedule predictions and monitor quality.

To find out how Anaplan can help, contact us for a personalised demo.

Transcript

This video will show you how you can set up and utilise Plan IQ within a demand forecasting model to enhance and improve your forecast accuracy. Plan IQ uses algorithm training, along with machine learning, to make intelligent forecasts based on your historical data. So, what is Plan IQ?

Plan IQ is embedded in the Anaplan platform for quick and easy setup. So, you can select your Anaplan module, the data, and, within that, for model training. You can then validate that data and then surface and correct any data issues. Within Plan IQ, you have a choice of multiple statistical, machine learning, and neural network algorithms from Amazon Forecast. And Plan IQ enables you to schedule, automate, or run ad hoc forecasts as needed, aligning to your business planning cycles.

So, what are the benefits? Well, Plan IQ delivers three key benefits that help teams achieve intelligent decision making. Teams can improve the accuracy of their predictions. They can make forecasting more accessible by empowering more users to unlock new insights, and they can automatically track new insights as your business grows. Without Plan IQ, teams across finance, supply chain, sales, and human resources can make highly accurate decisions with machine learning-driven forecasting. This joint solution combines Anaplan’s powerful connected planning platform with Amazon’s intelligent machine learning capabilities and their deep learning algorithms to generate accurate future forecasts fast.

To set the scene, we’re starting with a pretty typical demand management process in Anaplan. In this model, the historical demand has already been loaded into Anaplan, so we have the latest actuals. A baseline forecast has then been automatically calculated based on prior-year data, and the user is alerted to forecast exceptions that can be reviewed and commented on.

Before we go straight into Plan IQ, the next step for a demand planner as part of this process would be to go through a data cleansing exercise or outlier correction exercise. In this screen, we’re able to bring in our latest actuals and review, on a product-by-product basis, any outliers we have. This is a really important process to address these outliers prior to us going into Plan IQ and completing the forecasting process, as we don’t want the outliers to impact our Plan IQ forecasting process. There may be some outliers you want to keep, and some you may want to remove. It all depends on the demand planner.

Once we’ve addressed the outliers, we’re ready to push our data through our demand forecast. This is typically where Plan IQ will sit in our process. It is used alongside our more traditional statistical forecasting algorithms to essentially enhance our forecasting ability with machine learning and AI. In the past, this level of forecasting would require data scientists to manage and train the models. However, Plan IQ has been designed for the Anaplan model builders to manage the process.

On the Plan IQ overview page, we can see all our traditional statistical forecasting methods and algorithms that have been automatically run alongside our seven Plan IQ algorithms. Against each, we can see our forecast accuracy, and depending on the accuracy driver that we pick, the optimal forecast has been presented automatically by Anaplan on a skew-by-skew basis. If we want to override the optimal forecast, we can choose to do so.

The Plan IQ details page shows all of our products across the forecast horizon for each algorithm, side by side. Here we are forecasting out for the next three months at the weekly level. On the right-hand side, I can see the definition behind each of the Plan IQ algorithms, which are also available in more detail on Anapedia. We can also see our MAPE, or Mean Average Percentage Error, which is one of many methods for determining forecast accuracy, alongside our forecast bounds. Plan IQ allows you to select quantiles that provide an upper and lower boundary for your forecasts. The Plan IQ forecast produces a distribution of possible values for your forecast between those upper and lower bounds, rather than just a single point forecast. This helps to address the uncertainty in your forecasting.

So, in order to run our Plan IQ forecast, there are five key stages to go through. First is preparing our data. Second is collecting our data. Then, we’ll create our forecast model. Then, we’ll run our forecast, and then finally, we’ll configure a forecast action to use it in our Plan UX.

The first stage is to prepare your time series data that is going to drive the forecast. In this example, our data is actuals that we’ve loaded into Anaplan at the weekly level by SKU. Here, we are using four years of history, but the amount of history you feed into Plan IQ really depends on your business. A question you should really ask is, is there any relevance or significance with your actual sales data from X years ago? If not, do not include it in Plan IQ.

The second set of time series data to include is what we call related data. Related data is numeric time series data related to the items for which a forecast should be generated. Related data includes your historical data points leading up to that forecast horizon, but it can also include future data points to cover the length of the forecast horizon. While this data is optional, it does help Plan IQ improve the forecast accuracy, so it is worth entering if you have it. In my example, I have included related data for weather, inflation, unemployment, events, and promotional periods, but you can include anything relevant, such as maybe your product prices.

Once you have your data, you then need to go into Plan IQ and set up your data collection. So, I’m going to click on “New Data Collection” here. I’m going to give it a name, and then we can define the workspace and model that contains that source data I’ve just shown, along with that time scale, which in this case is weekly. You then need to map your data, identify what the unique item identifier is, what your time dimension is, and what your prediction target is. If you have related data, you can also repeat this process to collect that, again defining the time scale and our data mapping. Finally, if you’re providing additional attributes, such as the product size or type, you can also include this here. Once ready, click “Create Data Collection.”

Now that you’ve collected your data, you can create your forecast model. A forecast model is an algorithm trained on data from that data collection to support our forecast. To do that, we need to click on “New Forecast Model,” give it a name, and define the data collection that we chose on the previous screen, and our algorithm. The definitions behind which you can find here but also in more detail on Anapedia. Once you’ve picked your algorithm, you can then define your forecast horizon, how far out you wish to forecast and the time interval. Lastly, Anaplan will allow you to include a country-specific holiday calendar to assist with the forecast, should you wish to include it. So, I’m going to pick my UK holidays. Once that’s complete, click “Create Forecast Model.”

The final part of Plan IQ is to create your forecast actions and push your forecast back into your Anaplan model. So, we’re going to create a new forecast action and then define which model the data from the previous step we want to push back into Anaplan and into which workspace do we want to push that back into and which model. Lastly, you can pick the forecast quantiles that provide those upper and lower boundaries I talked about for the forecast. If you don’t want to get too technical, leave them on the defaults of 0.1 and 0.9, and that will provide a range known as an 80% confidence interval. Once you’ve created your action, you can choose to run this action either on demand or you can schedule them. So, I can choose to run this on demand or I can choose to schedule it. Once run, the data will flow back into that model you defined. You can then pick one of those algorithms as your optimal forecast on a skew-by-skew basis. And that is it. You’ve configured Plan IQ with your time series data, you’ve created a forecast model based on an algorithm, and you’ve pushed that forecast back into Anaplan for review and then identified your demand forecast method.

You can then continue on with the typical demand management process in Anaplan by kicking off the next step, which in this case is our consensus demand planning process. So, thanks for watching. Although it is an incredibly powerful enhancement, each of your businesses will have different forecasting needs. It is not a one-size-fits-all, so please do speak to us about where Plan IQ might add value in your Anaplan landscape and forecasting maturity. [Music]