rocket booster Rocket Fuel logo back to home page  
Our platform uses a combination of techniques that come from artificial intelligence, machine learning and statistics to boost campaign performance – throughout the day, and around the clock.


Technology FAQs

1. What’s different about your approach to channels or pre-determined segments?

2. How does your predictive modeling technology work?

3. How do these models apply to my campaign?

4. How does your technology make our online media buy stronger?


What’s different about your approach to channels or pre-determined segments?

Rocket Fuel looks at every impression on our network and selects just the best impressions for your campaign – this set of just "the best impressions for your campaign" is the custom segment our adserver builds and continuously refines for you.

Every time we get a call to serve an ad, our adserver analyzes all of the aspects of that ad call in real time – what page is the ad on, what user is going to see it, what time / day of week is it, and somewhat more fancy things like what data partners may have shared about the particular user, or even weather ("rain today, think I'll do my taxes.")

The Rocket Fuel adserver will serve your ad only if the impression is especially useful for your campaign. And this is true whether your campaign is trying to elicit a brand response (awareness, favorability, intent, etc.), a click, a purchase, or any other measure of success. Ultimately a segment of "the best impressions for your campaign" emerges, but the definition is wrapped up in a lot of math that's hard to scrutinize. It's not as simple as saying "male heads-of-households between 35-45" but you can tell it's working because the performance of the campaign on Rocket Fuel goes up steadily over time.

back to top^

How does your predictive modeling technology work?

Right now we use a combination of techniques that come from Artificial Intelligence, Machine Learning, and Statistics. Throughout the day, our system is updating its models for your campaign as we get impressions, clicks and conversion activity.

For example, a new site may join our network in the morning, and our adserver may run several ads on that site including yours to test its performance. By noon the adserver may shut the site off for you after learning that it performs worse than other sites, but it might keep the site running for another advertiser whose ads are performing well.

This happens without requiring intervention of a human operator, although we can override our system if needed.

back to top^

How do these models apply to my campaign?

Segmentation is defining segments/subsets/pieces/fragments/partitions of an addressable set of opportunities.

In direct marketing, segmentation often refers to defining segments of the targeted consumers. Some actual segments used by direct marketers are "money and brains" and "shotguns and pickup trucks."

At Rocket Fuel, segmentation is about defining segments of impressions, which are defined by not just the consumer on the other end but also the context (the page, time of day, etc).

Traditionally in online advertising, ad networks or publishers offer pre-defined / "off the shelf" segments that are defined via guesswork and are not necessarily optimal for any particular advertiser. For example, an ad network might have an audience segment of "tax software intenders" or "tax services buyers."

This has two shortcomings:

  1. Neither of these segments have been built with the goal of driving the most responses to the particular tax software ads.
  2. If the segments are based only on the user, an advertising strategy based on them might result in poor performance since other factors actually matter a lot
    Rocket Fuel builds a custom segment of the impressions most likely to generate a response for your campaign. Predictive modeling is a class of approaches that tries to build mathematical models from historical data, and use these models to predict future actions. Rocket Fuel's custom segment of "just the best impressions for you" are built using predictive modeling.

How does your technology make our online media buy stronger?

Often we propose running advertisements on contextually relevant sites because clients find it more comfortable. But the best long-term win is to run the campaign with as few constraints as possible, so that our system can figure out for itself where your ad works best. When we say "where" we mean not just which sites, but which times of day, which geographies, which users, etc.

We find that the best sites are often not the ones you would have imagined at the outset, and even a good site won't be uniformly good for any one particular advertiser – it's a matter of selectively picking just the best impressions from that site, and that's exactly what our adserver does.

back to top^

guy gesturing with hands in meeting