Gianluigi - I think that in order to understand what it really means to apply modern portfolio theory to keyword management, it's important to understand the history of how portfolio optimization first entered the SEM lexicon and practice. Efficient Frontier (my firm) was solely responsible for that (as I'll show below), so I'll run through both a short history of how portfolio optimization was introduced, as well as what the correct definition of portfolio-based keyword optimization is, and finally some things to look for in a solution that purports to leverage portfolio algorithms. If you only read one thing, though, look at the 2-keyword bidding example linked to
here and below, as it makes very clear what portfolio optimization is & isn't; Shimsand is, in my strong view,
completely and utterly mistaken when he says that a portfolio approach doesn't look at the performance of individual keywords. In fact, the beauty of portfolio algorithms is that they look at both keywords individually and the opportunity for getting more out of the overall keyword portfolio by bidding more or less on any one given keyword in order to free up $$ to better allocate to other keywords (by better I mean getting more volume and/or margin out of the overall portfolio than you would get by constraining all keywords individually to a given goal/constraint).
I blogged on this last week, so the material comes from
here. As you'll see, I feel very, very strongly that because of EF's success in the market applying portfolio algorithms, the term 'portfolio' has been adopted by many search marketing firms as the SEM equivalent of the 'organic' label in food = something that used to mean something but now doesn't because it's meaning and implementation has been diluted to a shadow of what 'organic' used to mean in farming. And for those SEM firms like Did-It who attempt to denigrate EF's portfolio algorithms, they simply don't know what they're talking about.
Back in late 2002 when Efficient Frontier started doing SEM work for clients, it was the only firm talking about applying quant hedge fund portfolio algorithms to the optimization of paid search campaigns. The few SEM firms who were more than a couple guys in their pyjamas made no reference to portfolio optimization, sticking instead to the rules-based mantra that their products are based on.
Some SEM competitors (all of them rules-based) decided that, rather than just sit there while being beat by EF in competition left & right, they would start talking about portfolio optimization as well in the hopes that advertisers wouldn't know the difference between:
a) portfolio algorithms that use historical cost and revenue data to model all yield scenarios across the keyword portfolio and automatically, consistently pick the optimal set of bids (Efficient Frontier's approach); and
b) portfolio optimization as practiced by rules-based SEMs(Did-It, Performics, Inceptor, 360i and AtlasOnePoint). When anyone other than Efficient Frontier talks about applying a portfolio approach to keyword management, what they mean is simply applying a rule(s) to a bucket of keywords and hoping that a human can tease a little more performance out of a few head-end keywords by hand.
As shown in this
two-keyword bidding example, applying 1+ rules to a bucket of keywords is the perfect way to doom your paid search campaign to mediocrity, yet that's exactly what the above firms (and others) are trying to pass off as portfolio optimization.
Falsely claiming portfolio optimization capabilities hurts the SEM industry the most because it sets improper expectations that result in advertisers giving up hope that better ROI can be had from their campaigns.
Anyone who claims to be capable of applying portfolio theory to the optimization of keyword sets should be able to provide proof. For starters:
Predictive Modelling Capabilities - the natural result of a set of algorithms that can leverage historical and actual impression/click/cost/revenue/margin data across a large keyword set is - surprise! - predictive models that can show the advertiser all of the possible efficient operating points for that keyword set. If an SEM firm claims portfolio optimization capabilities, ask to talk to a client of theirs and ask that client a) if they have access to keyword portfolio ROI models specific to their account; b) if those models are accurate; and c) if those models can look backwards and forwards.
Accurate Google AdWords Click Model - applying portfolio algorithms to PPC is useless unless your algorithms can see through Google's opaque marketplace. Unlike Yahoo Search Marketing (at least until Panama), AdWords doesn't tell you what CPC will yield what bid position, and if you simply take Google's Traffic Estimator at its word you'll be basing decisions on data that's off by 40%+, 70% of the time. Ask the SEM firm how they estimate traffic on Google, and if they say "Google has a handy traffic estimator", run away and don't look back. Now that I'm telling you this, though, expect all SEM's to start saying the same thing - so go to the next level of proof and ask one of their clients whether or not they get access to keyword-level *and* portfolio-level traffic forecasts, and whether those forecasts have been accurate over time. I'll bet if you do this that you get blank stares back from the SEM firm.
Efficient Learning Methodologies - if an SEM is truly leveraging portfolio algorithms, then in addition to getting more volume and/or margin out of a campaign, they should be able to learn about new keywords, tail keywords, and under-explored keywords efficiently - meaning spending as little as possible and in a controlled fashion to learn as much as can be learned about those keywords. What does proof of this look like? Well, for starters they should be able to explain what their learning methodology is, how it's built into their system, and how they leverage sparse data sets. If they can't do that, show them the door or expect any learning they get on your keyword portfolio to be incomplete... or expensive.
Effective Recency Strategy - in keyword optimization, recency is the notion of weighting recent data more or less heavily than historical data to take into account seasonality, inventory, promotions, and market volatility, among others; practically, it should also take into account inaccurate or incomplete historical data that if taken at face value would lead any optimization astray. Here again, don't just throw the SEM firm a softball; instead ask the open-ended question "What does recency in search marketing mean to you and how do you take it into account when optimizing campaigns?" If their answer isn't satisfactory, head for the hills.
Rules Are For Fools - if an SEM says they can do portfolio optimization but has a self-service product demo filled uniquely with references to keyword or campaign-level rules, consider this the SEM equivalent of the bait and switch and tell them to get back to you when they've made up their mind.
I know a lot of this sounds pretty complex, and I imagine some of you might think that it's unreasonable to expect the SEM salesperson you're talking to to know enough to speak at this level of detail - heck, the same might even be said of most online marketing professionals. I know that for me, learning about what portfolio algorithms are and what they aren't has been a multi-year learning process.