Advertising optimization can be very complex and confusing. Not to mention that in the Internet advertising space there is an abundance of available inventory. The problem is, only a subset of that inventory will work for your advertisers and campaigns. The challenge then becomes how do you spend a minimal amount of money to identify the subset of inventory that will perform the best but at the same time be the most cost effective? In short, how can you successfully optimize Internet Advertising without breaking the bank or taking up all of your free time?

Exact Drive is working diligently to solve this challenge and the optimization strategies and systems we already have in place are off to an excellent start. From a 30,000 foot view, an end user sets up the appropriate goals and fixed targeting criteria in Exact Drive’s self-serve advertising platform and our optimization systems immediately start learning about performance against several different variables including: inventory placement, creatives displayed, user audience, frequency cap, CPMs, etc..

Key Features

- Efficient learning algorithms minimize cost and maximize ROI

- Automatic advertising optimization to publisher and advertiser attributes

- Modifier system allows integration of continuous variables without drastic increases in learning costs

- Custom breakouts on any targeting variable available

Advertising Optimization

The purpose of quick advertising optimization is to find slices of inventory that provide a positive return-on-investment (ROI) for your campaigns. With Exact Drive you can optimize to a Cost per Click (CPC) or Cost per Acquisition (CPA).

The concept of advertising optimization isn't very complicated: it's taking past performance and applying it to a future goal, such as cost per click (CPC) or cost per acquisition (CPA). If you're likely to get a click, spend a lot; if you're not, spend a little or don't buy at all.

Even though it appears easy to provide example advertising optimization that way, in practice it can be hard to do. Should we calculate the click through rate for a whole website or a specific section? Does the creative itself make a difference to performance? Does frequency? Recency? Should we break out click through rate by day of the week? Geographic region? Can we bucket groups of similar sites with low volume and still optimize (apply a click through rate) to that? These are just a handful of questions that can be related to advertising optimization.

There's one more piece to consider, and that's the "Learning Stage." Before you can decide how to calculate or optimize click through rates or conversion rates or any other performance metric, you have to get some initial data of some kind. Basically, buying ad impressions to learn on usually has a much lower return on investment (ROI) than buying "optimized" because you are flying relatively blind, so you want to strategize a bit when buying and bidding in the learn stage.

All of this makes advertising optimization more complex than it seems. Good advertising optimization has a lot of "levers" that can be tweaked, and we would like to offer our clients the ability to tweak them using their own intuition and experience in combination with our analytics and reporting system.

Here are some of the types of things you can affect your campaign within the Exact Drive advertising optimization system:

Learn bids: We use a specially calculated metric called Estimated Average Price to get started bidding.

Throttling: We cap bids according to how confident we are in our performance calculations. This means, basically how much data we have to base click through rates and conversion rates on. A good rule of thum is to dial your throttling up or down according to how fast you need or want data.

Copy or pre-populate Learn: If you have learn data from other sources, we can get you started ahead of the game.

Getting out of Learn: How much data do you need to be confident of performance accuracy? You can trust us or adjust the number of events you need to get out of learn.

Frequency-Recency Modifier: All bids, learn or optimized, are adjusted via a frequency-recency modifier.

Projected learn modifier: This can affect how fast you decide to "give up" on certain inventory or get out of learn and into optimized bidding.

Targeted Inventory: We optimize to groups of objects, such as a creative-campaign-pixel-inventory tag bucket. If you target different inventory types, such as above and below the fold, in different campaigns, you will effectively break out optimization.

Learning

For cost-per-click (CPC) campaigns, the advertising optimization engine will use historical performance data: a click-through-rate (CTR) to determine whether to bid and how much to bid, and for CPA, it will use a historical conversion rate. But when you first add a campaign to Exact Drive, the advertising optimization system knows nothing about how that campaign will perform. We must first buy some inventory to acquire data so that we can "learn" how specific advertising campaigns and creatives perform on different slices of inventory (website publishers).

How do we decide what to bid in the learn stage? We use platform-wide historical prices for individual slices inventory.

After we buy enough impressions, one of two things happens:

Optimized Bidding

Once the advertising campaign is optimized it is time to start bidding smartly. With historical performance data, the process for coming up with a bid is relatively simple. We take the historical conversion rate, modify this based on a Frequency and Recency model and location on page also known as a cadence modifier, and apply your desired margin or goal metric.

Give Up

Some website inventory may not perform at all, and it isn't worthwhile to keep bidding if you are getting few or no clicks (low click through rate) and conversions (low conversion rate). If we have NO successes, or very few, this campaign will probably never be competitive on this website inventory. To make a decision to "give up" we estimate a "maximum CPM" based on the number of ad impressions and events we have gotten so far, and compare that to the estimated average price of the inventory. The decision of when to give up is called "learn persistence" and can be adjusted per campaign. When we give up, we bid optimized CPMs, but they will generally trend to zero.

Please Note

Advertising optimization is based on a unique combination of several factors including creative, campaign, and inventory source (and conversion pixel for CPA goals), meaning you won't want to dilute your budget over a huge number of combinations. For example, if one campaign has 20 creatives attached, EACH creative-campaign combo needs a significant number of events to get out of the learn stage. Or if a campaign targeted all available real-time media buying inventory, EACH website inventory source would require its own events to get out of learn. We recommend that to choose website inventory to get started on, you consider the types of website inventory that have worked for this type of offer in the past. You may wish to use our inventory categories, your own research, or choose the website inventory sources with the highest volume to get started.

Advertising Camaign Budget

In the learn phase, bids are based on the estimated average price (EAP) and not cost-per-click (CPC) or cost-per-acquisition (CPA) goals, so the impressions we buy in learn are not focused on performance but rather the accumulation of information. Therefore, you may want to limit the total amount you spend learning daily or for the lifetime of a campaign. Again, keep in mind, when an advertising campaign first starts, we have no data and thus every impression is in the learn stage. Limiting your daily learn budget during the initial stages of an advertising campaign may impact the speed at which you learn.

Advertising Optimization: Bidding

The first stage of advertising optimization is learning, where we gather data about the click-through rate (CTR) or conversion rate for a certain combination of items, such as creative-campaign-conversion pixel-inventory combination. Once we have gained enough data, we will use that information to start bidding the right price using some variation on the below algorithm, where the cost-per-acquisition (CPA) goal, or cost-per-click (CPC) goal and margin modifier are set by you.

convs/imps * 1000 * CPA Goal * Cadence Modifier

Gain Confidence

There is always a balance between getting quickly out of the learn stage and being confident that our click-through rate (CTR) or conversion rate is accurate. The more events (clicks or conversions) we have, the more accurate we are, but we don't want to stay in the learn stage longer than we have to. For this reason we have a throttling stage immediately after we leave the learning stage where we cap our bids according to how confident we are that our calculation is accurate. The more data we have, the more confident we are, and the less we cap.

Advertising Optimization Factors

- Cadence Modifier

The Cadence Modifier is a multiplier determined by our statistics team based on frequency, or how many times a user has seen a particular advertisement creative. For a low frequency impression, the cadence modifier will be high. For a high frequency impression, the cadence modifier will be low. The cadence modifier algorithm is planar and approximates historical user response to different frequencies and recencies. Currently, the cadence modifier is constant across our self-serve advertising platform, but soon it will be specific according to client and website inventory source. In the case of no-cookie users, the cadence modifier will default to 1.

- Throttling

When your advertising campaign is past the learn stage but still isn't fully confident, we use a form of throttling to make sure you don't overspend. We do this by capping bids on an optimized node depending on how confident we are in the performance data. The cap starts at EAP and scales towards ECP as we become more confident in the performance.

The formula for this is: Max CPM = EAP + (ECP - EAP) * Prediction Confidence where

Prediction Confidence = events^2 / confidence threshold ^2

A Simplified Use Case Of Advertising Optimization

Let's say you want people to sign up for your cheese of the month club, and a year's subscription earns you $50 in revenue. You set up an advertisement campaign that targets Oklahoma-based users on who are not currently subscribed to your club and serve display ads to them on photobucket.com.

 • How likely is it that showing the user a display advertisement will convince him to sign up? 

• How much would you pay to show him/her a display advertising based on this likelihood?

During the learning stage, Exact Drive will acquire enough 25-30 year old Oklahoman men on MySpace.com to get a statistically significant number of conversions. Let's say Exact Drive finds out that the conversion rate is 1 per 5,000 (0.02%) when they are shown advertisement creative A. In other words, if you pay a $1 CPM for such advertisement impressions, a conversion costs you $5.

In the next stage you need to decide what to bid. Since your estimate revenue for a conversion is $50, let's set your cost-per-acquisition (CPA) goal to $50. Hypothetically any acquisition that costs less than $50 nets you profit, although you would like to acquire for much less. That is where the margin comes in.

If you decide you want to make a profit margin of 80%, then the advertising optimization system will optimize to a cost-per-acquisition (CPA) of $10. Ignoring the Cadence Modifier, which will adjust the bid according to frequency and recency, you get:

1/5,000 * 10 * 1000 = $2 per Conversion

If interested in trying out Exact Drive's self-serve advertising platform please register below for free and kick the tires.