Forecast business practice changes on your emissions with what-if analysis

What-if analysis is a custom AI model that forecasts the impact of several business practice changes on your organization's carbon emissions footprint. It helps you create more informed carbon reduction strategies and accelerate your overall sustainability goals. For example, you can forecast the impact of switching to renewable energy sources, such as wind or solar power, or switching suppliers through supplier-specific factors.

This article shows you how to create a what-if analysis forecast. It also provides considerations, details, and information to help you get the most out of your forecast.

Important considerations

Keep these considerations in mind while creating your forecast scenario.

• You can layer between one and three strategies within a single scenario by using the Number of scenarios field.

• The Variable(s) field is multiselect. Choose the data or quantities that you'd like to forecast changing. For example, you might choose to modify the Cost, Distance, and Goods quantity being transported as part of your 4. Upstream transportation and distribution scenario.

• Every scenario requires the following fields:

  • Name
  • Organizational unit
  • Data type
  • Calculation model
  • AR version
  • Number of strategies
  • Variable(s)

• Facility isn't a required field, but you can use it to further narrow down the data for your forecast.

Create a what-if analysis forecast

To get started with what-if analysis, follow these steps:

1. On the navigation pane, select What-if analysis.

2. On the What-if analysis page, select New scenario.

3. On the New What-if scenario pane, enter the following details:

   • Name
   • Organizational unit
   • Data type
   • Calculation model
   • AR version

4. Enter the number of strategies that you want to layer through the Number of strategies field.

5. Select the Variable(s) that you want to forecast changing as part of your strategy or strategies. Enter the historical values for these variables, which represent the current strategy your organization is using. For example, if you're forecasting the impact of changing your Mobile combustion vehicle fleet, enter the current vehicle and fuel type under Current strategy.

1. After you complete the form, select Save & Close.

2. Select your created scenario, and then select Run scenario. After the scenario job completes, you receive an in-app notification alerting you of the results, with a hyperlink leading you to them. This page displays your scenario details and a graph visualizing your historic data and associated strategy forecasts.

Supported scenarios

Each scenario offers different levels of customizability. Tailor the forecast to your organization's data and needs by using variables. This table lists all the available variables for each data type.

Forecast aspects

• Existing strategy: The existing strategy forecast shows your projected emissions if you make no changes to the current way you generate emissions for that category. For example, if you're forecasting the impact of switching from coal to biofuel for a particular facility, the existing strategy forecast represents the projected emissions of continuing to use coal.

• New strategies: The new strategy forecast shows your projected emissions if you switch to the new business strategy represented by your forecast scenario. For example, if you're forecasting the impact of switching from coal to biofuel for a particular facility, the new strategy forecast represents the projected emissions of switching to biofuel. You can explore one to three new strategies depending on how you set up your scenario.

• Prediction intervals: Prediction intervals represent the estimate of an interval in which a future observation falls with a certain probability (95% confidence), given the historical data. Prediction intervals essentially represent the uncertainty associated with a forecast.

Model failures and informational messages

This section explains errors or problems you might encounter with forecasts.

We made some adjustments to generate this forecast

• Switching to a fallback forecasting method: The system uses a fallback forecasting method if the number of historical data points or the data quality isn't good enough to fit (S)ARIMA or ETS models. Two specific cases require switching to a fallback methodology:

  • Too many missing data points in an otherwise relatively uniformly spaced historical data time series
  • Irregularly spaced historical data

• Data uniformity check and frequency adjustment: Before forecasting, the system aggregates your data on a monthly level to generate a monthly baseline and what-if forecast. However, if the aggregation doesn't present a relatively uniform monthly cadence, the system attempts further aggregation to two, three, four, or six months. If it isn't possible for the series to achieve relative uniformity under these adjustments, then a simpler fallback model is used for forecasting.

Unable to generate forecast

• Historic data is too sparse: To ensure a successful forecast, the system requires your historical data to have a frequency of at least one data point every six months. If your data is more sparse than that interval, the forecast fails.

• No or too few historical data points: The what-if analysis forecasting models require at least six data points (after frequency adjustment, described earlier in Data uniformity check and frequency adjustment) to successfully generate a forecast.

Supported time series forecasting models

Sustainability Manager supports (Seasonal) Auto Regressive Integrated Moving Average ((S)ARIMA) and Error Trend Seasonality (ETS) univariate time series forecasting models for generating forecasts on activity data. The model selection framework selects the best forecasting model based on the historical activity data. The generated activity-level forecasts go through the calculation model to transform them into emissions-level forecasts.

ARIMA and ETS are the most widely used time series forecasting methods. ETS models rely on the descriptions of trend and seasonality in the data, while ARIMA models describe the autocorrelations in the data. To learn more about these models, refer to Chapter 7 (Exponential Smoothing) and Chapter 8 (ARIMA models) of the Forecasting: Principles and Practice textbook.

On certain occasions, such as when the historical data is too little or highly irregular, the system selects a simple fallback model instead of ARIMA or ETS.