Jul 132009

Bid Management Misunderstanding #2: "Bid Rules"

"Rules based bidding is a disaster!" PPC Expert 1
"Are you nuts? Bid rules are essential!" PPC Expert 2

As with the question of Automated Bidding, it turns out both of these folks might be right depending on what they mean by "bid rules" -- a loaded term if ever there was one.

Rules-based bidding earned a lousy reputation in the early days of PPC and for good reason. The types of rules varied wildly from the rudimentary to the complex but suffered conceptually on two different grounds: they were generally based on hunches rather than data, and they were often position-based. Rules often looked like:

  • Bid $1 if the KW hasn't had an order and has spent less than X. Turn it off if it's spent more than X with no orders. Bid $2 if its had an order and its cost to sales ratio is good. Yadda yadda.
  • or

  • For this group of KW Bid to position 3 if it's less than $2; if its more than $2 for position 3 take position 6 no matter what it costs. For this group of KW...
  • or, the pinnacle of rules based bidding: the open auction games:

  • Bid to position 3, unless position 2 is within 5 cents then bid to position 2. If position 4 is within 2 cents (bid jamming) and we could save more than X% by dropping to position 4 then do that...and on and on.Very clever ways of playing the game badly.

In addition to bidding by hunch rather than data, and letting competitors dictate your CPCs by playing position games, rules-based bidding also frequently involves another poor practice: factoring in sunk costs.

Let's say you've just assumed management of a program badly managed by someone else. All things considered you're aiming for break even profitability. For a given KW, the previous manager spent $10,000 on 10,000 clicks which drove 100 orders and $5,000 in margin. Horribly inefficient, yet the answer isn't to turn it off, it's to realize that if 10K clicks drives $5K in margin you can only afford to spend $0.50 per click going forward. The fact that you've lost $5K on that term already is irrelevant. The CFO may go berserk wondering why that bad KW is still running, but the answer is: the traffic is valuable...not as valuable as the other person thought, but it does have some value.

Bidding smartly involves predictive modeling. The goal is to determine what the value of the traffic driven by each different ad is likely to be and then set the bid to the fraction of that value you're willing to spend on marketing. The best foundation for guessing the value of the traffic comes from the data you've observed at the most granular level possible. This gets complicated as the data gets thin, and requires some fancy stats on the way different attributes of the ad (the keyword, product category, sub-category, manufacturer brand, match type, etc) are likely to impact the value of traffic. The algorithms can be incredibly smart, and can involve as many checks as you want: "Don't change a bid more than X or don't bid more than Y without human review" whatever.

However, while bidding based on observed past performance is necessary, it is not sufficient by itself. Oftentimes the smart analyst knows that something is about to happen that will change the performance of a KW or collection thereof in ways that no algorithm can predict, or react to quickly enough.

A big sale is happening on wide screen TVs next week. This will impact the conversion rate, the AOV, the margin %, etc. all of which impact the value of the traffic. The algorithm will figure this out eventually, but is likely to react after the sale is over, missing opportunity on one side and wasting money on the other.

Enter the smart type of "bid rule". The analyst can say: "Past experience indicates that during a sale of this magnitude the value per click changes by X%, so I want to lift bids by X% during the sale to capture more traffic without sacrificing efficiency. And, after the sale is over, the algorithm is going to crunch on data that is somewhat inflated by the sale effect, so I might have to push the numbers generated by the algorithm down a fraction to stay within the target."

These types of rules overlays are essential in retail advertising to account for short term offers, fluctuations in stock levels, seasonal effects (Does the algorithm know which Friday is Black Friday, or when your shipping cut-off for Xmas delivery is? or what the July birthstone is?), etc.

Agencies without these rule overlays invariably underbid at the beginning of the Christmas rush and waste money after Xmas is over -- we've seen the data, over and over again.

So are bid rules good, or bad?

It depends on what you mean by rules!


7 Responses to "Bid Management Misunderstanding #2: "Bid Rules""
Marc Adelman says:
Great Post. I have to come up with another way to start a comment, because the "Greatness" of "Great Post" may start to depreciate in meaning if overused. Ok... really amazing post. I feel you have zeroed in on a philosophical pivot point of the PPC world. Should bid rules be a tool or weapon, if you will, that is wielded by a PPC Shaolin Ninja, or should they be more kindred to H.A.L., and bring us swiftly and effortlessly into the new dawn of profit and success. The Ninja approach integrates the efficiency, flexibility, and accuracy of algorithmic bid rules with the intellect of human analytics and strategy. The H.A.L approach is the underlying culprit in the insidious "Set It and Forget It" culture that has infected all too many PPC agencies + consultants and company side PPC teams. Anyway, as we learned in 2001 A Space Odyssey - H.A.L is clearly not the answer and might led to our downfall. I think where RKG draws its strength and distinction is its balanced approach between advanced, flexible algorithmic bid management and advanced human analytics and strategy. (George, I just moved, so you will have to send the check to a new address). But seriously, as you stated, "It depends on what you mean by rules." Do you run the rules..... or do the rules run you.
Thanks for your kind words, Marc, the check is in the mail :-) Another metaphor I've used centers around cars. The defense department created a competition a few years ago with a $1 Million pay off for the winner. The challenge: build a car that can drive itself from point A to point B across a desert without wrecking. The first couple of years none of the contestants made it to point B. In 2007 or 2008 someone finally made it and won the prize. They, obviously, spent millions and millions of dollars in R & D to get this to work, and it putts along at 30 mph. I'd be willing to bet that a smart person in a Range Rover will ALWAYS be able to outperform the "winning model". Indeed, a Range Rover from the 1930s would outperform the 2008 Winner as long as it has a smart driver. As we built our system we kept this in mind: build a powerful flexible machine that a smart human can handle and that will be the best solution every time. The military understandably wants drones that can take on dangerous missions without risking lives. I'm pretty sure no one has died as a result of PPC management, so I don't see any need for the robotic solution. Thanks always for the dialog. George
Billy Wolt says:
I personally believe there is a happy medium. While "setting and forgetting" is always a bad strategy, using rules to handle the bulk of your accounts while using Marc's "ninja" approach on the groups of keywords that add the most value, would probably be best.
Hi George, Great post. You're right - rules are often biased on historical performance so they can be inefficient when dealing with fluctuations, such as in your TV sale example. Rules also place too much emphasis on 'converting' keywords, and not enough emphasis on keywords that create awareness at the earlier stages of the buying cycle. I'd like to see rules with higher weightings for metrics such as page views, time on site, bounce rate and returning visits, as they are key indicators of interest generation. If you knew a keyword wasn't converting but people were viewing on average 38 pages and spending 3 hours on the site, you wouldn't want to pause the keyword, but the robot will. Marc - I like your 'ninja approach' of mixing automation with human intellect. Does that mean bids aren't actually changed, but advised?
Thanks for your comments Billy and Alan. My point is the algorithms can take into account any metric you want to feed it, including the "buying cycle" data you mention, Alan. By letting machines do robotic tasks you free the smart human analyst to study the data, study the user search strings for additional KW and negatives, and study and fine tune the rules themselves. Just as the race car driver doesn't need to climb under the hood with a wrench to shift gears, accelerate, or turn the wheels, having an analyst "getting their hands dirty" manipulating bids manually doesn't make a lot of sense to me.


Check out what others are saying...
[...] the next post I’ll talk about different approaches companies take algorithmically, and the confusing use of [...]
[...] Bid Management Misconception #2: “Bid Rules”, Rimm Kaufman [...]

Leave A Comment