Zeta Forecasting
Recent times have witnessed businesses making grave strategic errors because of inaccurate industrywide forecasts in contexts ranging from market research or investments to competitive intelligence. While these inaccuracies in assumptions stemmed from a lack of efficient forecasting techniques, the companies started to realize how past data (related to sales or market growth) can be utilized to determine possible upcoming trends and extrapolate what could happen in the future.
As technologies emerge and industries evolve, gauging the present state of the market has become paramount. Zeta, however, gives more importance to understanding where the market is heading. In light of the same, ZMP Forecaster allows forecasting across its varied tools.
How does Forecasting Work?
At the most basic level, forecasting works by estimating future events* via looking backward at historical events that have already happened.
*We are defining an event here as anything that makes a call to Zeta servers and creates a log. For example, when a client signs into your website, when a prospect opens your email, or when an ad loads on a website page.
How do I Interpret Forecasting Results?
When looking at forecasting results, you'll encounter two main types:
Range forecasts
Specific forecasts
Range Forecasts | Specific Forecasts |
|---|---|
Range forecasts give you a sense of the expected performance based on the previous 14-day period. Based on these historical averages, we show a range of what's likely to happen, with a margin of error. In other words, range forecasts are ±1 standard deviation of a 14-day historical average and provide an upper and lower bound of expected performance for any given day. | On the other hand, specific forecasts are a bit more detailed. They use a time-series statistical method (for those interested, Double Exponential Smoothing) to predict the performance of a metric over a certain time period. This prediction is based on how the metric has behaved over the past 14 days. So, it's like a cumulative estimate of what's expected to occur over time. |
Confidence Scores
When available, Confidence Scores provide a baseline for how accurate we believe our forecasts to be. Confidence scores can look a little different throughout the platform. Here’s how you can expect to see these scores show up:
Written as “within 10% accuracy” | This is Zeta’s claim that forecasts aim to be within 10% of actual results. |
Written as “within 90% accuracy” | This is a goal that Zeta forecasts are accurate 90% of the time. |
Broken out into either a High, Medium, or Low Confidence rating. | These ratings are based on how close our trained model results are to a control group and represent our confidence in meeting the Accuracy and Precision targets. |
Finally, in some areas of the platform historical forecast data can be compared to actual delivery data. This is Zeta’s way of being transparent about how our system is performing and learning over time.
If at any point this data is confusing or alarming, let us know! We aim to ensure our forecasts and forecast methodologies are transparent and always improving.
What Steps Should I Take to Troubleshoot?
As always, if something doesn’t look right, don’t hesitate to reach out to Zeta Support.
Remember to take screenshots or videos and write down the steps you took to get to where you are in order to get to a resolution faster (For example: (1) logged in, (2) navigated to this page, (3) clicked on this button, (4) experienced this strange thing, and (5) if possible, provide URL.)
When it comes to forecasting tools, there are likely some things you can try on your own:
Problem Statement | Not enough data resulting in no results. |
Solution | In this scenario, you can try increasing the dataset that is being forecasted. For example, this may be increasing the look-back or forward date ranges or adding more people to your audience. |
Problem Statement | The forecast stopped loading and timed out. |
Solution | In this scenario, there may be too much data being forecasted and you may need to decrease the amount of data being considered in order to see results. For example, this may be decreasing the look back or look forward date ranges or reducing the size of your audience. |
Problem Statement | The widget or card is just empty 🤷 |
Solution | This is a particularly frustrating problem, so when it happens try the tried and true IT 101 solution and remove the widget then re-add it or do a hard refresh of the page you’re on. If this doesn’t work, reach out to Zeta Support. |
Where can I Find Forecasting Tools within ZMP and Zeta DSP?
Dashboard Widgets in ZMP
There are forecasting widgets available for users in the Dashboard Widget Library.
Information on all widgets is available in the Dashboard section: Widget Dictionary
Experience Builder in ZMP
When building an experience, users can see a forecasted outcome of each applicable node.
Information on forecasting per node is available in the Experiences section: Node Level Insights
Audience Builder in ZMP
After building an audience, users can see forecasted performance by channel in the Audience Builder.
Information on audience forecasting is available in the Audience section: Omnichannel Audience Forecaster
Report Builder in ZMP
Certain Customer Marketing and Media Activation metrics can be forecasted in the Report Builder.
Information on these metrics and how to add them to your report can be found in the Analytics section: Report Builder: Forecasts
Labs in ZMP
The Total Spend Optimizer, New Spend Optimizer, and Scenario Planner tools are available in ZMP for select accounts.
Information on how to get access to these forecasting tools is available in the Streamlight Applications section.
Campaign Builder in Zeta DSP
Users can get up-to-date forecasts based on various parameters while building a Media campaign in Zeta DSP.
Information on the DSP Forecaster is available in the Campaigns section of the Programmatic Knowledge Base: DSP Campaign Forecaster
Forecaster in Zeta DSP
Zeta DSP users have a dedicated forecasting tool that they can access at the advertiser level in the platform.
Information on how to use the DSP Forecaster is available in the Programmatic Knowledge Base: DSP Forecaster