• http://www.clickequations.com Craig Danuloff

    Niraj – Hope you don’t mind some push back on these ideas.

    Your first point seems to be that manual management of text-ad-testing is better than letting Google shift traffic toward ads with a higher CTR. Certainly the idea of considering conversion rate as well as click-through-rate is important (although how often they really diverge is another question). But CTR drives Quality Score so it certainly can’t be dismissed as a core determinant of ‘best’ performing ad. And deciding which to pause based on comparison to an average of the others is certainly a slightly more advanced way to make that decision. But fundamentally the AdWords algo will reduce impressions of low performing ads (CTR measured) and increase impressions of high performing, so when a person isn’t carefully monitoring a test leaving campaigns set on ‘rotate evenly’ can be quite damaging. Everyone has seen campaigns were one low performing CTR ad lingers forever wasting returns while set on ‘even’ distribution.

    The second point against considering revenue attribution is curious. Two things are true, 1) if attribution doesn’t shift revenue for any keyword, then it won’t harm any bid calculations, 2) in a small minority of cases using attribution radically shifts the revenue and therefore the appropriate bid calculation. Suggesting that attribution isn’t important because it doesn’t impact all keywords all the time is like saying someone shouldn’t wear a seat belt because the usually don’t have a car accident. And applying attribution isn’t (or shouldn’t be) a lot of work – either your software supports it or it doesn’t. If it doesn’t I would agree with you that hand-calculation isn’t worth the trouble…

    Finally, the third point seems itself based on three other myths – that hitting your ROI target is the optimal bid (isn’t the real goal to pay the least for the most) and that moving up in position creates higher ROI (sometimes yes, sometimes no), and that raising bid price raises position (will for some queries matched to the keyword while introducing others at lower position, possibly lowering overall KW ave Pos). Really the whole idea of ‘headroom’ as an indicator of potential bid increase seems strange – again because the goal isn’t to give the engine as much money as possible and the bid itself didn’t assure that it was the optimal spend in terms of return. Maybe I’m missing something you’re saying with this point because I really don’t get it – admitting that. this is a hugely complicated subject to fit into a few paragraphs. Also confused as to how the word ‘portfolio’ applies to any of this.

    I think the idea of myth busting in PPC is a great one. Hope this discussion can further that goal.

    - Craig / ClickEquations

  • http://www.efrontier.com Efficient Frontier

    Since it is our specialty, we are compelled to oppose myth 3 in which Niraj suggests portfolio management is easy to implement.

    Portfolio Theory was developed by a Nobel Prize winning economist and uses mathematical models to weight risk and return across numerous variables. In the same way, true portfolio management for search is founded upon keyword modeling and is about making bidding decisions on all keywords at the same time. Looking at keywords individually will almost always return a less than optimal solution.

    There are three main factors that determine the quality of a model:

    • Availability of historical data – It is about looking at overall impact on performance while accounting for bidding effects one keyword might have on the other. There is an exponential relationship between clicks and position – using simple rules may cause you to bid too high and burn budget for a poor return.

    • Capability to properly model tail terms – 30-40% of spend comes from the long tail. You could not use a simple ROI rule to bid these up or down. A much better solution is to use finite mixture models.

    • Adaptability of the model – Campaigns are influenced by factors such as competitors, news and search engine algorithm changes. Our models ensure keyword trends are identified and responded to immediately. It is impossible to do this at scale using the approach suggested in the article.

    The bottom line: using a simple heuristic on a complex problem will result in suboptimal returns. If estimates of CPCs et al are not accurate, then the method to optimize for performance will not be successful.

    For more details see our whitepaper (copy and paste) http://bit.ly/4rZqDB

    Justin Merickel
    Vice President, Marketing and New Product Development
    Efficient Frontier