We use cookies. You have options. Cookies help us keep the site running smoothly and inform some of our advertising, but if you’d like to make adjustments, you can visit our Cookie Notice page for more information.
We’d like to use cookies on your device. Cookies help us keep the site running smoothly and inform some of our advertising, but how we use them is entirely up to you. Accept our recommended settings or customise them to your wishes.

PPC Bid Management Misunderstanding #3: "Portfolio Approach"

When an agency says they use a "portfolio" approach to bid management, what does that mean? The answer is: not much. There is no single, universally accepted meaning to the term "portfolio bid management." Nor will statisticians smile knowingly if you say your system uses "portfolio modeling" because the phrase has no mathematical definition. We use the term portfolio simply to distinguish our approach from what we would describe as atomic approaches. The atomic approach views each ad's data in isolation of all other ads. Atomic systems set bids at the keyword level if they have enough data but fall back on rules when they don't. In other words: bid to data when there's enough and when there's not bid $1 for this category, $0.50 for that category etc. {Note: To be truly "atomic" they'd need to go one level deeper, to the ad level, but I'm not aware of any systems for rent that go below Keyword level} This is a bad idea. You could get away with this approach in 2001 but not now. There's too much value in the torso and tail that is mismanaged under this type of system. A portfolio approach takes data from all the keywords into consideration as it sets bids on individual ads. But there are different ways to do this, hence different meanings of "portfolio." One firm built their portfolio system to find the right position* on the page for each keyword in the portfolio to maximize ROI. The idea is that if you have $1,000 to spend in a day, you may end up bidding down some keywords even though they are efficient, so that even more efficient keywords can be pushed, thus maximizing ROI. But, herein lies the rub, in our view. This system was designed, apparently, to answer the question: "What's the best way to maximize ROI from a fixed budget?" That's really an investment question, not a marketing question. For a direct marketing program (as distinct from a branding campaign) budgeting search doesn't make sense. Particularly because there isn't even a cash-flow consideration -- sales are generated before the advertiser pays their advertising bill -- there is no reason to bid down ads that are efficient to stay within a budget, nor is there ever a direct marketing reason to push the bids of terms beyond their efficiency target in order to burn up excess budget. Budgeting both wastes money and misses opportunities. Our model was built to solve the much more common marketing problem: "How can we generate the most sales within an ROI target?" This is a fundamentally different problem that leads to a fundamentally different solution. Our approach to the portfolio concept relies on the fact that similar keywords, siblings if you will, often behave very much alike, so by studying data across similar keywords we can better predict how traffic from this particular keyword will likely behave. How relationships are defined between keywords is a central issue. Many "portfolio systems" that share our approach only allow clustering based on a couple of attributes and those are often defined by the inflexible hierarchy of the engine account structure. We believe making the portfolio model work at the highest level requires infinitely many well considered, thoughtfully assigned attributes. Simply aggregating data across a collection of attributes -- eg: all the low traffic terms that are: 1) related to the same sub-category; 2) related to the same manufacturer brand; and 3) related to the same style, or material type, or holiday, or match type, etc -- will get you pretty far, provided that you have a smart, multi-layered process for expanding the group incrementally as needed to reach significance. Exactly how we do our modeling is proprietary, naturally, and has evolved over time. We've moved beyond the approach above to look at the impact different attributes have on conversion rate likelihood and AOV, using different MLE techniques for each. We've found that refining our statics has made a difference, and we'll keep pushing the envelope in this regard. To make a VERY long story short, "portfolio bid management" means different things to different people. When shopping you should ask: 1) what problem is your system designed to solve?; 2) how many attributes can be assigned to each ad?; and 3) how are bids set on low traffic ads? Don't expect a terrifically detailed answer to this last question, but they should be able to tell you whether they rely on MLE stats or simple aggregation. * We have tremendous respect for these folks and think they have the second best :-) bidding system in the world -- we do hope they're kidding about the position bidding part.
Join the Discussion