THE RKGBLOG

Paid Search: Man vs Machine

My monthly paid search column at Search Engine Land, in case you missed it:

The question is often debated in the industry: “Which is more important the bid management tools used or the person who does the managing?” Sometimes the argument becomes self-serving, with artisans who manage their programs manually saying it’s all about the person, and tool providers arguing that it’s all about the algorithms, analytic power and automation.

For large, complex, tail-heavy accounts, this is like asking: “Would you rather be operated on by a skilled surgeon using stone tools, or a Neanderthal using the latest equipment?” The first may be preferable, but neither option is attractive.

However, the fact of the matter is: not all programs are large, complex and tail-heavy. Indeed, the roots of many disputes in our industry are a product of the vastly different types of challenges and accounts on which we’ve worked. What is most important to some programs is least important to another, hence digging into the man vs machine debate with a bit more precision may be a worthwhile exercise.

The Human Element

Human’s are much better than machines at many elements of the paid search game, including:

  • Keyword selection;
  • Match-type and syndication layering;
  • Negative association anticipation and selection;
  • Landing page selection/testing;
  • Copy writing;
  • Anticipating seasonal, geographic, and promotional effects;
  • Classification schemes, recognizing and predicting meaningful keyword commonalities;
  • Establishing and evaluating success metrics;
  • Troubleshooting;
  • Nuanced understanding of attribution, the characteristics of channels, and the ways that both last-click attribution and cheesy “assist” counting mechanisms can torque the truth.

The Power of Machines

Machines outstrip humans in some areas of the game as well:

  • Speed, speed, speed. Automation allows a person to move mountains in a hurry.
  • Precision. Programmers may make mistakes, but machines execute solid programs pretty much flawlessly.
  • Data crunching. We have a number of folks on staff who can do pretty high-level math — present company excluded — but those aren’t uniformly available skills. Machines and software packages can execute enormously complex calculations across massive arrays of ads and engines and do so hourly without making a mistake.
  • Pattern recognition. Human’s are key for providing flagging mechanisms for algorithms to watch, but the machines are often better than human’s at identifying signals in the data and separating them from statistical noise.

It’s worth noting that not all paid search automation software is created equal. I once heard a computer described as “the fastest, most diligent, but stupidest clerk you’ll ever meet.” Indeed, James Zolman was spot on when he pointed out that many so-called bid management platforms simply execute rudimentary rules set by humans (eg “if over efficiency target by X then decrease bids by Y”, or “bid this collection of keywords to position 6″). More valuable software with real algorithmic power is a rarer breed.

Of course, not all paid search managers are equally good at all facets of the game either. Smarter, better trained analysts will do the human work better than the average Joe.

Quality software allows smart human users to do tremendous volumes of work, quickly and flawlessly, and takes away much of the routine scut work from the plate of the the paid search manager. This has the important side-effect of making their job much more interesting, allowing them to do more of the brain work that humans do better than machines, and improving retention rates of paid search managers. The better the machinery, the better caliber of paid search manager you’re likely to attract and retain.

However, both the human manager and the software used cost money. These costs can exceed their value depending on a number of factors:

  1. How much money can be spent wisely on search? If everything is handled perfectly, what’s the benefit? If the answer is “not much” because the market niche is too small, having both brilliant software and top notch analysts on the case may be too expensive.
  2. Is the tail material? If 95% of the program is in the top 100 keywords, then the value of automation is greatly reduced. As the value of the tail increases the value of automation and smart algorithms increases.
  3. Is there much traffic volume? Sophisticated algorithms require a certain volume of traffic before they become more valuable than intuitive judgment. If the data is too sparse (regardless of CPC) smart algorithms simply don’t help much.
  4. Is there much ongoing maintenance to be done? The ongoing value of human oversight may be reduced if the product/service offerings are static, if there isn’t much seasonality, if there aren’t many promotional changes to implement, if geography is irrelevant, etc.
  5. Are there many anomalies to address? The fewer exceptions to the rule there are due to seasonality, events, the news, competitive pricing, inventory shortages, etc. the more accurate will be the assumptions of a good bid management system and the less value there is in smart human data analysis and anticipatory bidding.

For large, complex, dynamic programs both sophisticated technology and a sophisticated manager(s) are essential to generate the best results. However, as the above implies, in other cases what matters most depends entirely on the context.

The essential element in almost any case is having the core program build out properly from the ground up, and this is only done well by humans. For tiny accounts, the cost of that build out may even be prohibitive, and Google Boost may be the right solution for those folks, but for anyone spending real money investing in the smart skilled human to do the work well is the one piece that no amount of hardware or software sophistication can replace.

Beyond that build out, whether the algorithm or the person is more important will depend entirely on context.

Comments
14 Responses to “Paid Search: Man vs Machine”
  1. Jc says:

    “…but those aren’t uniformly available skills.” Would you say that they are “normally” available?

  2. Hi George,
    I agree with most of what you said in principle but I like to see it as a Man OVER Machine approach. The algorithms and technology piece takes care of time consuming, tedious and complex data crunching type optimization tasks. The savings in time and cost in turn translate to requiring fewer people to manage a large campaign effectively. Once a smart human figures out how to do something efficient in a scalable manner, there is little reason to have a human to it over and over again. Of course, lets not forget that

    “Artificial intelligence is no match for natural stupidity.”

  3. Mark says:

    You make a lot of good points in this post. Automation can help with stability and identifying data patterns but in especially volatile search marketplaces in can end up limiting results by settling in positions it can clearly establish as efficient.

    I especially like you point about tail end keywords, because when you don’t have a lot of them, this is when automation can be especially limiting. If your account performance is dependent on 10 KWs, you could be losing a lot by having a conversion based automated strategy which isn’t pushing its limits every day.

    The other point that should be made here is that the industry has evolved to a point where there are some very viable in between options. Specifically, “saved searches” in Kenshoo’s “advanced search” feature allow the user to set specific criteria and thresholds, and apply bid changes across the entire account based on compartmentalized performance thresholds (i.e. if cpa> 10, actions>20, pos worse than 3, qs better than 5, increase bids by 30%). This means that manual bid optimization can be just as effective and easily implemented as automated strategies, and can maximize output in volatile markets where automated strategies can’t.

    I agree that this approach would not work for “tail-heavy” accounts such as large retailers, where it may be more important for the algo to establish patterns. At the very least, it closes the gap.

  4. Hi Sid, I totally agree. For the large, complex, dynamic accounts both are essential, and the technology helps us hire and retain super-sharp analysts because they don’t have to do the routine, repetitive stuff. It’s a waste to have talented people perform algorithm-friendly tasks.

  5. Hi JC,

    That’s a great question. Hiring someone with a PhD in statistics is certainly doable, but what we’ve found is you have to have more than just math chops to do this well. Understanding the marketing context is an essential ingredient, and finding high quality mathematicians who also understand that context is less straight-forward. My point was more that you need the heavy horsepower to design a flexible algorithm, but that option isn’t available to everyone — not enough to go around — and moreover, that person should not have to do the calculations manually for each account — that’s where machines excel. Outsourcing can make sense, SaaS licenses for top drawer products can make sense, it just depends on what you have available, and what you can acquire readily.

  6. Mark, thanks for your comments!

    There certainly are instances where traffic value fluctuates too rapidly making reactive algorithms too slow to the game. Our system allow analysts to build any sort of conditional logic they need to either modify the algorithmically developed bid, or go it alone. We would make the case that position-based rules are almost always the wrong way to go as they give control over to competitor’s behavior, but sometimes we’re chased down that path against our better judgment, too.

  7. Jc says:

    Haha George I was just making a statistics joke, but I’m glad you were able to take something away from it :)

  8. Heh heh, ohhhh, I get it now! I’m a bit slow on the uptake on Mondays :-)

  9. Philip Rosen says:

    Excellent post, George. You really hit on one of the key dilemmas facing paid search managers today. I wanted to add my two-cents by broadening the discussion beyond the two options you mentioned: human or bid-management systems. {Sales pitch deleted}

    Our belief is that paid search (and paid advertising in general) cannot and should not be fully automated–only parts of it will be. Frankly, the automated bid management tools have brainwashed us throughout the years to “automate” but in real life, any SEM and/or media manager out there will tell you how much tedious, manual work they still need to do (e.g. using Excel to manipulate and analyze campaign data to come up with actionable insights)

    Next generation SEM tools need to empower the human, not to replace him or her. We believe that at least half the time of and SEM manager should be invested in analysis {sales pitch deleted} It is much more important to know where to put your next advertising dollar than to automate that allocation (again, full automation does not work).

  10. Thanks for your comments, Philip, and please pardon the edits. We prefer your commentary to your sales pitch :-)

  11. billy wolt says:

    Humans = proactive

    Machine = reactive

    I’d personally rather prevent fires rather than put them out. No machine is going to build a campaign from the bottom up that is setup correctly for the given advertiser.

    Although, I have seen some very poor work from humans that claim to be ppc experts because they setup their cousins adwords account for his t-shirt store.

  12. That is a very succinct way to frame it, Billy. Well done. You made the same point I did in 1/15th the number of words! :-)

  13. @Billy… I love the distinction. The fire analogy is a great one. Thanks George for getting our wheels turning.

Trackbacks
Check out what others are saying...
  1. [...] This post was mentioned on Twitter by Meta Tags and others. Meta Tags said: Paid Search: Man vs Machine: My monthly paid search column at Search Engine Land, in case you missed it: The qu… http://bit.ly/fPybsw [...]