Purpose
This article provides an overview of the InnovidXP feature Predict, including how it works, metrics, and recommendations.
How does Predict work?
- Predict uses the TV spot and response data you have uploaded into the InnovidXP platform. It models the data, based on the parameters and rules that you have set, and the amount of your planned budget that you want it to consider re-allocating.
- Predict applies advanced algorithms to the modeled data to determine the optimal way for you to adjust your future plan and buy, identifying spots that have underperformed and that you should decrease spend on, along with spots you should buy more of as they perform strongly.
- Predict also calculates the anticipated impact of following these recommendations on your overall KPIs, for example, your total response levels and CPA.
How does Predict help my business?
Predict recommends increases or decreases in your buys to help improve TV efficiency for your business. It employs powerful analytics to help answer the question ‘which TV spots should I buy?’. By analyzing your historic TV performance data and calculating cost, it generates actionable decrease or increase buy recommendations to optimize your future TV buy and drive improved TV efficiency, for example:
Select Export CSV to view in a spreadsheet format, for example:
Using the date range options for optimal calculations
Predict will automatically apply the optimal historical date range for your calculations. You can change the settings on the Core Parameters page.
To apply your own date range, select the Manually Select option. This displays further options, for example:
Use the slider to change the date range, or there are two quick links set up for you to:
- Select the last active three months
- Select the last active four weeks
The recommendations for optimal calculation are as follows:
- The date range should span a minimum of 28 days.
- The date range should include at least 14 days on which spots aired
- There should be a minimum of 21 days between the first and last spot
- There should be at least 500 spots within the date range
Note: Select Apply to save your changes.
Dimensions in Predict
Dimensions are TV spot parameters you want Predict to consider when it generates its recommendations.
Channel is a common default, but you can select up to five; for example, many users add parameters like creative, spot length, and daypart.
Excluding Dimensions
If you want to exclude any Dimensions from Predict's calculations, this can be done using the Rules Configuration functionality.
Select the campaign you want to edit, go to the Rules Configuration page, and scroll to the Optional Advanced Rules area. Here you can create any number of rules that tell Predict what to exclude from its calculation for that Predict configuration.
For example, if you wanted to exclude a particular daypart from the calculation:
1. Select Create Rule to display the following popup:
2. Enter a name for the rule, select Daypart as the Dimension, then select which daypart(s) you want to exclude.
Note: The available values for the selected Dimension are displayed in a dropdown list.
3. Select the Exclude checkbox.
If you want to set a budget instead, leave this checkbox deselected and enter a minimum and/or maximum amount.
4. Select Add Rule and it will display in the list of rules:
5. You can select any rule to edit in this list to edit it, or click Apply to proceed and apply the rules in the list.
Changing the timescale of your media budget plan
Predict generates new recommendations on a 7-day basis, so that the recommendations are always current and optimized based on your latest data. In fact, you can set your Predict configurations to run automatically every week.
However, if you want to adjust this standard timescale you can use the Recommendation Date Range option on the Rules Configuration page.
- In Predict, open the Configuration you want to change and select the Rules Configuration page:
- Move the Recommendation Date Range slider to the number of days for your required media buy cycle.
Limiting the maximum number of TV spots Predict recommends
You can limit the number of TV spots Predict recommends by setting a rule within the Optional Advanced Rules area on the Rules Configuration page and setting a maximum for a particular Dimension.
Limiting the maximum number of TV spots Predict recommends
The maximum number of TV spots can be limited by creating a rule within the Advanced Rules area in Predict.
You can limit the number of TV spots Predict recommends by setting a rule within the Optional Advanced Rules area on the Rules Configuration page and setting a maximum for a particular Dimension.
Calculating the total cost for your campaign
Predict calculates the total cost by analyzing the spots over the chosen time period, which is typically 28 days, and summing the spend associated with these spots.
It is important to note that Predict will adjust the total cost based on any Optional Advanced Rules that you have set. The Rules Configuration page lets you select options and add any optional rules, for example:
Select Apply and Predict uses this cost, time period, and the optional rules to project the amount of money that can be optimized during the recommendation period - this recommendation period can be set as 7 or 28 days and the projected spend will be scaled accordingly, for example:
Grouping channels by Saleshouse
You can select Saleshouse as a Dimension and then see recommendations relating to Saleshouse.
Performance and Buy Metrics
A Performance Metric is the metric you want to use to measure your TV spot performance, and that you therefore want Predict to focus on improving. By default, this is set as 'cost per response' but you can select any metric you wish from the picklist.
A Buy Metric is how you specify your TV buy, so usually this will be by either cost or audience impressions.
Regional recommendations
We can configure Predict to provide regional recommendations in most circumstances. The customer must provide regional information as part of their spot logs. This configuration is currently unavailable for Market Area-level analysis in the U.S.
Different Values
Values shown for totals and performance metrics in Predict may be different from the rest of the portal.
There are two reasons for this:
- Firstly, by default Predict standardizes metrics to a period of 7 days (unless this setting has been adjusted under the Rules Configuration page).
- Secondly, in order to ensure optimized data modeling for Predict recommendations, we exclude all spots with a buy value of zero, as we want to ensure the model avoids recommendations to purchase 'free' (ie. zero cost) spots.
Historical runs
Now that we have introduced the concept of Predict configurations, you will be able to see the last historical run for each configuration, not the last 30 days.
Related content