Bidding at Launch

My monthly paid search feature at SEL, in case you missed it: Craig Danuloff wrote a very interesting, somewhat controversial piece a few weeks back arguing that advanced bidding strategies should take a back seat to core campaign development, keywords, adgroup architecture, ad copy and landing page assignments for most advertisers. We agreed that good bidding can't save a badly constructed program, but as Bradd Libby correctly observed: you have to start with some bids and doing that well rather than badly is important. As much as RKG has written about paid search bid management, we've never really addressed how to bid and manage a brand new program when no paid search data exists to drive smart decisions. Let's take up that challenge. If you trust Google with your conversion data another good way to wade in is to use their Conversion Optimizer. This will only help with relatively high-traffic campaigns and that by itself limits its value. Nevertheless, by most accounts Conversion Optimizer does the basics reasonably well, with minimal effort. Conversion Optimizer, of course, won't set bids on Bing, and not everyone is eager to share conversion data with Google, so what follows is advice for the rest of us. Step 1: Establish a Success Metric
  • What is the purpose of driving traffic to this site?
  • What action(s) marks a successful visit?
  • What is the value of those actions to your business?
  • What fraction of that value are you willing to spend on marketing?
  • Can you spend an unlimited amount of money on marketing if the target efficiency metrics are hit, or must you budget?
Let's say that for Acme, the goal is to generate qualified leads, and that the success metric is therefore contact form completions. Let's say that on average, 10% of web leads turn into customers, and those customers are worth $1,000 to the business. Your goal is growth and you're therefore willing to plow 90% of that value back into driving new business. We can then calculate that the efficiency target that makes sense is: 10% x $1,000 x 90% = $90/lead More info on establishing a success metric is here. Step 2: Evaluate Risk-Tolerance Done well, paid search is a data-driven game. The more data the more predictable the performance. At the outset, there is no hard data to go on, so the question becomes: how do you want to strike the balance between financial risk and rapid learning? At one extreme: you'll learn the fastest by throwing all the ads to the top of the page with no budget caps in place, and adjusting as the data pours in. These lessons will be learned at a high price, too high for most. At the other extreme: the most cautious approach would be to launch with bids of $0.01 on everything and ratchet up the bids as little as necessary to begin gathering data. This is much safer, but with data trickling in it will take a long time to tune the program, incurring opportunity costs. Neither of these approaches is sensible, but there is much room in between. In Steps 3 - 10 I'll outline a moderate approach capturing some of the fiscal responsibility of the most cautious method but with a faster track to a well-tuned program. Step 3: Estimate Likely Success Rate What fraction of paid search traffic is likely to convert? A bad proxy for this might be your overall website conversion rate. If you have no other option, you can start with that, but hopefully you have better options. This is a poor proxy because a big chunk of your existing site traffic already knows your brand and will therefore convert at a much higher rate than competitive search traffic. A better proxy: if you have some website analytics find out the conversion rate of existing organic search traffic. Clearing out the direct load and email traffic will greatly improve the accuracy of the conversion rate estimate. Better still: parse the organic search traffic into search on your brand (trademark, domain name, etc.) and search on other terms related to your offerings but not referring to your business by name. As before, the people who search for you by name will convert at a much higher rate than will strangers. Since the main value of paid search is attracting the folks who aren't already looking for you by name -- and likely to find you -- this non-brand conversion rate is the best predictor of paid search conversion rates. Let's say this non-brand organic search conversion rate is 5% This suggests that on average you might be able to afford to spend 5% of $90, or $4.50/click. Step 4: Get Granular Averages lie. Can you break this down into its components? Do some categories of terms convert at a higher rate than others? Do leads from some of these categories tend to be more valuable than others? Apply the methodology in step 2 with as much granularity as you can, developing best-guess initial bids. Step 5: Enable Tracking and Reporting You can't play in this space if you can't track the performance. Most Analytics packages have tracking capabilities that will tie costs to results at a granular level. Google Analytics is free, which is a very good price. Step 6: Establish Guardrails Create campaign daily budgets that reflect how much you're willing to lose on the first day when we're really taking SWAGs at bidding. Hopefully our guesses are good enough to deliver some success with the money spent, but you don't know for sure until you see the proof. Campaign daily budgets are absolutely the wrong way to budget a program once it's up and running efficiently, but they serve as useful guardrails as we're learning what works and what doesn't. Step 7: Let 'er Rip Throw the switch in the morning and make sure you'll be in the office the next day to make adjustments. Step 8: Damage Control After you've collected a good blast of data on day 1 it's time for triage. As the saying goes "crap floats". Sort a keyword level report by cost descending and see what the performance looks like for the highest spending terms. If you see major bleeders adjust as needed; use the methodology for estimating initial bids but substitute in the actual observed conversion rates in place of the proxy guess. {Note: if you have significant latency between initial visits and leads you may need to do an additional calculation to account for this. For example, on top of the overall conversion rate, understanding the same session or same hour or same day conversion rate will allow you to scale up your conversion rates appropriately. For example, if half of the conversions you see organically happen within the first hour of the first visit, typically, apply those same stats to your paid search performance to figure out whether doubling the observed number of conversions would make the stinkers look better and the solid performers look like stars} Repeat this process aggregating by AdGroup and Campaign. Ideally, knock out the data from the individual high traffic KW to get a feel for the performance of 'the rest of the AdGroup/Campaign. In these assessments at each level, are there any obvious negatives that you might have missed in the build out? If so, add them and continue collecting data. A search query report may shine a bright light on these. If unsure, try flipping to exact match to see if that dramatically changes the results. If you've put the appropriate time and attention into the build-out, most of these kind of structural issues shouldn't be the issue, but it's worth checking. The other explanation is that the quality of traffic is simply not worth what you're paying and you should pay less -- not nothing -- for it. Step 9: Release the Hounds! Are there KW, Ad Groups, Campaigns performing better than anticipated? Feel free to raise bids if the new calculation (using actual paid search conversion rates instead of predicted values) is better than anticipated. Are there KWs or Ad Groups that aught to work well (you have very strong offerings compared with the competition) but aren't getting any traction because you aren't bidding enough? Try bumping up to the first page minimum and see what happens. Budget caps are essential here: You may find that the traffic quality is better than anticipated and the extra spend for the traffic is justified; you may find that the market is irrational and that the value of the traffic isn't worth the price. Either way, it's important to find out for these keywords and categories that are core to the business. Step 10: Lather, rinse, repeat Repeat steps 8 and 9 as the data builds, looking at longer and longer windows of data as the data grows. Richer data sets give a truer, less noisy picture of the traffic value at more granular levels. Ideally, looking at a couple of months worth of granular data will get the program pretty well tuned. In these solid performing campaigns, make sure you're not hitting budget caps. It's silly to stop spending if you're getting positive ROI. Hitting budget caps in the early stages of tuning is okay. Ongoing it's a disaster. Budget caps are the WORST way to control spend and hitting them routinely is a sure sign of mismanagement. If for some reason you can't spend more than X on a collection of Keywords, reduce bids proportionately until you average about the right spend. This gets you more traffic, leads and sales for the same money than slamming into caps. As I've said before: budget caps are like guardrails: useful safety features, but no substitute for steering. Conclusion: Rocket-Science bid management systems are essential for big spenders in competitive marketplaces; quality search agencies wouldn't have spent years in research and development if it didn't make a difference. But not everyone can afford such a system, and for smaller programs the ROI of sophisticated stats approaches may not make sense. That doesn't mean bid management doesn't matter. It does mean that relatively simple approaches like the one outline above can get you pretty far with very little investment. Hope it helps!
Join the Discussion