Anil Kamath talks about Portfolio Based Bid Management
Published: June 4, 2007
As the founder of Efficient Frontier, Anil Kamath is the primary architect of the company's core product. He continues to lead the product and technology innovations at Efficient Frontier. Before Efficient Frontier, Anil founded eBoodle, an ecommerce company providing comparison shopping and digital wallet services, that was acquired by Bizrate. At Bizrate, Anil managed the Bay area office and was responsible for Bizrate's contextual advertising product.
Prior to founding eBoodle, Anil was Vice President of International Equity Trading at D.E. Shaw & Co., a hedge fund that specializes in using quantitative models for program trading. Anil worked at Bell Labs and has a master's and a Ph.D. in computer science from Stanford University. He has authored multiple technical papers in the area of optimization and also has several patents.
Also participating in the interview was Michelle Schofield, Efficient Frontier's VP of Marketing.
Eric Enge: Can you tell me what it was that led you to play a founding role in Efficient Frontier? What was it that was motivating you?
Anil Kamath: My background is in the area of optimization, and I have a Ph.D. in this area. I went to Wall Street, spent three years at a hedge fund, trading international equities; came back to the Bay Area, and started a comparison shopping company where I got first exposure to the world of search marketing. I sold the company BizRate/Shopzilla, where we were developing our own search market place for products. So, when I started Efficient Frontier it was a matter of putting my background to good use in some sense. I wanted to bring algorithms and technology to the area of advertising specifically, search marketing which lent itself very well to this because you can control how much you paid for the ads. You have the ability to track and measure in details, and you can have very large campaigns with large numbers of keywords which you can control and target. It was the idea of using algorithms and technology to get the best result for the advertisers that got us started with Efficient Frontier.
Eric Enge: Can you talk a little bit about the problems with rules based bid management?
Anil Kamath: There are plenty of limitations in the rules based approach. You have very little information, because conversions are few and far between. So, if you have a hundred thousand keywords, maybe ten thousand of them will have a conversion, but the other ninety thousand you don't have any information on.
How can you use a rule for that? You also have the situation that the search engines are broad matching keywords, and you are essentially taking a keyword, and then map matching it to other keywords. The result is that you don't have a clear sense of what the landscape looks like; what value you are getting for which keyword. You also don't have information on how it's going to change, based on the geographical targeting. So, rules based approaches definitely have their limitations in the context of market places where a lot of the information that is necessary for these rules to work is not available.
Eric Enge: One of the problems is that there is a strong incentive to build out long tail of keywords. This may work well from a campaign perspective, but with rules based Bid management it just doesn't scale because there is no data.
You may in fact have a keyword that gets one click and happens to provide a conversion. It may be the last click you are going to get for a month but the conclusion is to raise the Bid price through the roof.
Of course, most rules based bid management packages provide some filtering, so they don't do anything until there is a certain amount of data. But, of course for many of these keywords we'd just never get that much data and it would be unmanaged essentially.
Anil Kamath: Yes. It's suboptimal. You can't just Bid a keyword based on its own conversion; you need to look at the elasticity of the Bid as well in terms of how your bid effects your position in the market place that you are participating. What the competition is bidding against you, and how many clicks you'll get more or less based on whether you bid more or less on that keyword. When you take the information of that keyword, and combine it with information that you have about other keywords; you'll have a pretty complex problem. For every dollar that you are budgeting on a Search Marketing Campaign, you need to figure out on what you are going to spend it. Whether you are going to spend it on increasing your Bid on keyword A; or are you going to use it for improving your position on keyword B? So, it's not simply a matter of managing to a CPA on a keyword; its matter of how you spend your budget efficiently, across a portfolio of keywords to get the best result.
Eric Enge: Why is it that the portfolio management is so much better?
Anil Kamath: There is a good algorithm and theoretical basis about how a portfolio methodology can be used to get the best result from an advertising campaign that a simple keyword rule based system won't be able to deliver. With even the most complicated rule based solution you won't get the best possible ROI, which a portfolio based method can get you. The math behind it is pretty complex, but if you use that math you will get better ROI and we have shown that for a large number of customers. It frees you up from the management of individual keywords, so you are not looking at individual ads and trying to make your decisions, and setting your rules on individual ads. You are now looking at the whole campaign or the whole set of campaigns on which you are spending your money and figuring out how to bid on the individual keywords so that the portfolio as a whole gives you the best performance.
Eric Enge: Right. So, you are still setting some goals.
Anil Kamath: Yes, you are setting your business goals and you are freeing yourself from specifying rules for individual keywords. We think of it as translating those business goals into the individual keyword using our portfolio theory to get the maximum ROI for our customers.
Eric Enge: Can we dive a little bit into how it works?
Anil Kamath: There are three main things actually. First you'll have to get data about the market place, and about conversions in the market place. We get data from a wide variety of sources. A lot of it comes from the search engines. A lot of the data comes from our customer's accounts themselves; we also track conversions for our customers. We get information from various sources on keywords themselves as to which keywords are related to which other keywords. This is the first step, getting that data on a regular basis. The second part is taking the data and coming up with a forecast. We use various statistical methods to make these forecasts very accurate. So, when we say forecast, we are saying forecast down to the ad level. For example, this keyword targeted at this geography, with this ad copy will get fifty clicks if it's bid to two dollars at position two. It will generate seventy clicks if it's bid to five dollars, and it's at position one. So, we come up with very granular forecast, based on the data that we have, and using various statistical methods to make these forecasts very accurate. So, the forecasting is the second part.
These forecasts then go into the optimization system, which is essentially a linear programming solver. Think of it as a system which takes the business goals, and figures out how to optimally allocate your money across the different advertising options that you have. We then decide if should you bid four dollars on a given keyword or three dollars, or should you take money from one keyword and put it toward raising the position for another keyword. The forecast and the optimization are what combine to give us the best chances of meeting the business goals that the customer has set. The business goal could be to spend a hundred thousand dollars and maximize my registrations, or maximize my revenue. Or, it could be to get twenty percent margin, while getting you the most revenue possible. The linear programming solver is the thing which takes the forecast and the business goals, and comes up with the bids that meets your goals and maximizes your ROI.
Eric Enge: Isn't there also a value to understanding the conversion potential of the keywords.
Anil Kamath: Yes. That comes in the forecasting. The forecasting has two pieces, one which I described, which is for each keyword predicting what clicks, what position, what CPC you are going to get for various bid levels. The second part is understanding how likely a keyword is to convert into whatever revenue metric the advertiser is interested in managing to, whether it be a registration or revenue or an order or a subscription. Our forecasting does predict how a keyword will convert into that revenue metric.
Eric Enge: One of the key things is that you understand how to group keywords, and what keywords are highly related to one another. So, you can treat things as chunks of data to solve the problem with not having enough data.
Anil Kamath: Yes. For that you need statistics. You use very sophisticated statistical analysis to group keywords, and come up with good forecasts on individual ads.
Eric Enge: Presumably, as you get real data from the campaign, you use that to update your predictive modeling.
Anil Kamath: The forecast gets updated regularly, as new data comes in. They are continuously being updated, based on information that becomes available.
Eric Enge: One of the really interesting things you've done is figured out how to group all the keywords into meaningful data chunks. What has your experience been overall, across your entire customer base in terms of the types of gains that they typically get?
Anil Kamath: It varies from customer to customer. We usually get about 30% to 50% better returns from using our methodology over whatever they were doing before they signed on with us. The degree of improvement depends on the market place. It also depends on how much they have been optimizing themselves, and how much attention they have paid to their campaigns.
Michelle Schofield: Some of our customers report that just moving their campaign over to Efficient Frontier, they begin to see improvement and lift even before we do other optimization activities.
Anil Kamath: This is without changing any ad copy or restructuring the campaigns or any landing page improvements.
Eric Enge: Because they are now doing a better job with the bidding process.
Anil Kamath: Yes. The other aspect which I forgot to mention is the learning aspect. We do a fair amount of learning on long tail keywords, and new keywords. We allocate a certain percentage of the budget towards testing keywords where we don't have enough conversion information.
Eric Enge: You mean you experiment with new keywords, and you see how they do. You do this testing to help you find the optimal mix.
Anil Kamath: Yes. We do this through algorithms essentially.
Eric Enge: How big does a company's spend need to be to benefit from this system?
Anil Kamath: We have two products. One is the enterprise product, where we work with customers who spend about a million dollars a year. That's the full service, where we do other things for them including creating ad copy and generating new keywords. The other product is the express product which is for advertisers who spend less than fifty K a month. It's the same underlying technology, but we limit the kinds of things they can do with the express product. For example, there is a limit of two thousand five hundred keywords in the express product.
Michelle Schofield: They can opt to for an additional fee to increase the number of keywords. Note that we created the express product because smaller advertisers kept coming to us asking us for it. The product makes it easier for them to manage the majority of their campaigns on their own. We also have client services available, but it's primarily setup for someone who wants to take advantage of the technology and the bidding optimization, and wants to be able to run their campaigns more efficiently, even though they are not spending quite as much as the big guys are.
Eric Enge: Does the express product work for someone who is down in the ten, twenty thousand dollar range?
Michelle Schofield: Yes.
Eric Enge: What does the pricing look like, in rough terms?
Michelle Schofield: It's $50 a month to bring up a test campaign and play with it. Then, it's about eight hundred a month to run the full blown campaigns on Google, and Yahoo, and MSN. You can add additional services on top of that.
Eric Enge: At eight hundred a month; you are talking about the expressed product still.
Michelle Schofield: Yes. Pricing for the enterprise product is based on percent of spend, and that depends on the size of the company, the number of keywords, and what services they are using.
Eric Enge: Do you have a way to work with agencies?
Michelle Schofield: Yes. We've made it really easy to manage campaigns for their clients, from small campaigns right up through enterprise products where the agency can private label it, or co-brand it. We'll take on the management of their accounts for them either with the account knowing they were managing it, or still having the agency as the front door to them.
Eric Enge: Thank you Anil and Michelle.
Anil Kamath: Thank you.
Michelle Schofield: Yes, thank you!
About the Author
Eric Enge is the Founder and President of Stone Temple Consulting (STC). STC offers Internet marketing optimization services, including SEO, Social Media and PPC optimization, and its web site can be found at: http://www.stonetemple.com.