Bid Management Science Explained

Bid management is vitally important for SEM — that’s not much of a revelation. These days, most big spenders use a bid management technology, but don’t really think about how these bids are calculated.

I thought it would be worthwhile to outline some of the science behind bid management so that everyone can see “how the sausage is made.”

Rules-Based Bidding

The simplest form of bid management is rules-based bidding. A rules-based bid essentially looks at each keyword independently to make a bid, rather than looking at how keywords might work together to achieve a goal (that is portfolio-based, which I’ll talk about in detail later).

I say “simplest,” but really rules-based bidding is still quite nuanced. Here’s how it works: let’s say a client has a goal of a $50 cost per acquisition (CPA) and is getting a 10% conversion rate. A rules-based bid system would calculate a bid at $5 per click ($50 X .1).

But wait – there’s more! That $5 bid might be adjusted by the bid tool based on a variety of additional data. A few of the more common variables include:

• Actual query (vs. keyword)
• Geography
• Time of day and day of week
• Device
• Average revenue (for e-commerce)
• Clustered performance of similar keywords with sparse data
• Differential between bid and actual CPC

In other words, an average conversion rate of 10% might only be 5% on mobile and 15% on desktop, and those mobile clicks might convert at 2x in Florida after 5:00 p.m. So all of sudden, the simple rules-based system is now making very complex decisions based on granular data.

Positional Bidding

The most sophisticated systems go beyond just “what is the optimal bid to achieve a CPA objective” and instead look at “what is the optimal bid to maximize performance.”

For instance, in the example above, it may be true that a $5 bid will get a customer to a $50 CPA — but what if a $4 bid would actually get the customer roughly the same number of conversions at a $40 CPA? This is a much tougher calculation to determine because it requires an algorithm that can model outcomes at different bids.

The way I have always thought about this is “optimal performance by position.” In other words, for each position in an ad auction, there will generally be three important variables for a bid system to consider:

• CPC price
• Volume of clicks
• Conversion rate
• Bonus, but less impactful: average revenue

Although we think of Google as a “cost-per-click” (CPC) system, Google actually makes its money through “cost-per-thousand” (CPM) pricing. Your position in Google’s results is determined by the combination of your CPC times your click-through rate (CTR). CPC X CTR = CPM.

What is often shocking to folks that have never analyzed a given keyword by position is, in many cases, your CPC actually increases as your position decreases. If a keyword happens to generate a ton of clicks in top position and few clicks in lower positions, the top bidder might actually pay a lower CPC (but a higher CPM) than lower bidders with a higher CPC (but lower CPM).

So, when optimizing for position, a bid system should actually calculate which position generates the optimal “profit per-thousand-impressions,” which I’ll abbreviate as PPM. PPM is calculated by subtracting revenue per-thousand-impressions (RPM) from CPM. CPM – RPM = PPM. But remember – volume also varies by position.

If you’ve ever looked at your “top versus side” reports in AdWords, you might see that CTR and volume is often 10x higher for ads at the top of the page than on the side (and in the case of mobile ads, volume pretty much disappears if you aren’t in the top two results).

So obviously, optimizing exclusively to PPM could lead a bid system to bid for a low position with high profit but little to no volume.

The challenge, therefore, for a bid system is to find the bid that puts the customer in the optimal position to maximize profit dollars (or revenue, or conversions, etc.) across the total volume of available clicks.

In the hypothetical example below, you can see that optimizing to RPM, PPM, total profit and total number of conversions would actually result in significantly different results:

In this example, to extract the most profit (assuming that is the goal), the bid system should find a way to adjust bidding to maintain position four in the results.

Portfolio-Based Bidding

As you can see, rules-based bidding isn’t that simple after all. When you factor in variables like geography, query, and time-of-day and then try to combine that with positional optimization (and the sparse data that usually prevents you from getting a full picture of the bid landscape of a given keyword), getting bids right for a rules-based system is a tall order. Portfolio-based bidding takes this to a whole new level!

The concept behind portfolio bidding is that a customer’s objectives should be measured at an account level and not at a keyword level.

For example, let’s say a customer has a budget of $1,000 and a max CPA threshold of $100. There are two keywords that can each drive 100 leads, but one of these keywords costs $100 CPA per lead and the other costs $50 CPA per lead. In a rules-based system, the system might bid each of these as aggressively as possible, because they both hit the client’s CPA objectives.

In such a scenario, it is possible that the keyword with a $100 CPA might get conversions more quickly and thus $900 of the $1,000 would be spent on the $100 keyword and the remaining $100 on the $50 keyword. This would result in a total of 11 conversions.

With portfolio-bidding, the system would instead attempt to maximize $50 leads first and then only buy $100 leads after. This would result in 10 $50 leads ($500) and five $100 leads ($500) for a total of 15 conversions.

Now, add to this the positional and geo/time/device analysis we have already discussed, and you have a system that is calculated for both the optimal position of the individual keyword and the optimal budget allocation across the entire account. That’s portfolio bidding at its finest.

Show Me The Algorithm!

Every bid management company out there talks about its proprietary technology and amazing results. Underneath all the fancy sales talk are algorithms using models like the ones I just described above.

When making a choice between different platforms, I find it useful to actually talk to the data scientists behind the scenes and ask them how the algorithms actually work. Not all algorithms are portfolio-based, and not all algorithms account for position when determining bids.

Understanding the science behind the platform will help you “optimize” your own decision-making process!

  (Stock image via Shutterstock.com. Used under license.)