That consumers are rapidly changing the way they interact with advertisers is no secret – but the rapidity in which they are doing so is quite surprising. As I looked into the data from the last retail season three trends stood out.

Christmas Day Is The Early Predictor Of Tablet Traffic For The Next Year

Tablet impression-share has a distinct pattern. It spikes on the weekends when people are home and drops in the weekdays when people are on desktops in their offices. This  five low, two high pattern is clearly seen in the chart below which shows the tablet impression share of paid clicks for a subset of retail advertisers.

However, note the spike around December 25 , Christmas Day. Tablet impression-share spikes and remains high until Jan 1. I noted the same trend with 2011 data. We found similar trends in the UK, too.  This spike reflects the fact that many users get tablets as gifts during the holiday season, specifically around Christmas day.

What’s more, the impression-share around this time, sets the baseline for the next year until the next holiday season rolls around. Thus, while tablet traffic is growing every month, the bulk of the increase happens around Christmas.

Tablets Show A Very Distinct Hour Of Day & Day Of Week Pattern

We live in a multi-screen world with consumers using different devices at different times of the day and day of the week. The charts below reflect this trend.

During a typical weekday, desktop traffic is high during the working hours, notches down around 5 p.m. Most people leave work, and it picks up again early evening. In contrast, tablet traffic is low for most of the day but shows a distinct increase around 8 p.m. Smartphones, on the other hand, show no distinct trend.

Conversions by Hour of Day and Device

In contrast, the weekend trends are quite different. No device shows a distinct pattern, and the distinct 8 p.m. tablet hump disappears. These trends reflect when and where these devices are being used – desktops in the office, tablets at home and smartphones everywhere.

Advertisers Have Taken Notice Of The High ROI On Tablets

Tablets have had lower CPCs than desktops despite the higher ROI on these devices. This was because advertisers had not caught on to the rapid adoption of tablets by consumers and their proclivity to convert on these devices. It now appears that marketers have taken notice of the trend.

In a sample we tracked, tablet CPCs were 17% lower than desktops in Q1 2012. By Q4, they were only 10% lower.

Key Takeaways For Advertisers

1.  Tablets will account for 1 in 4 paid search clicks by the end of 2013.

Looking at historical trends and the Christmas day jump , one can anticipate that tablet traffic will account for 25% of all paid search traffic by the end of this year.

2.  It pays to target to devices separately.

The varied traffic patterns on tablets , smartphones and desktops imply that advertisers should target these devices separately on search. Broadly targeting these devices with the same campaign means that one is leaving money on the table.

3.  The tablet opportunity still exists.

Despite the rise, CPCs on tablets are lower than desktops. Given that tablets enjoy higher conversion rates than desktops, marketers should increase the investments on tablet targeted search campaigns.

The above trends highlight the rapidly changing surfing and purchasing behaviors of consumers on the Internet. While it is challenging for marketers to navigate this new ecosystem, it also opens up new opportunities. I hope the above trends enable you to make the most of this new multi-screen , multi-device world.

While search marketers in particular have to overcome several tactical and strategic issues in January to ensure success; this January, I see three key areas where search marketers would be well advised to focus their energies.

Adjusting To Rapid Changes In Traffic Patterns

For most businesses, the period between December 25 and January 1 is a quiet period when traffic dies down. However, traffic patterns pick up right after New Year’s Day. To complicate matters, advertisers in many verticals see an uptick in traffic due to seasonality.

For instance, consider the chart below that shows traffic patterns of the top categories in the travel vertical, which was made by calculating the traffic volume of several high-volume keywords from the flights, cruises and packages categories. Here, keyword traffic significantly picks up in January.

Seasonality patterns in travel. Note that traffic volume drops significantly in November and December, and picks up sharply in January, especially in the cruises and packages categories.

Similar effects are seen in other verticals, too. In the financial vertical, interest in tax and retirement related products usually picks up in January. Automotive advertisers see a strong pick up in January owing to new cars being launched. While retail, in general, is quiet in the first quarter, niche categories like weight loss products see very strong interest in January, too.

All these factors point in one direction — search marketers must re-evaluate their bidding strategy right after the holidays.

Priming For Super Bowl & Other Semi-Predictable Traffic Jumps

As seen below, a successful Super Bowl advertisement can generate a huge jump in search traffic for an advertiser. This is both a challenge and an opportunity for the search marketer and presents a budgeting dilemma for January.

Google Trends data for a Super Bowl advertiser. Note that while the spike is predictable during Super Bowl day, it is hard to predict the size of the spike.

Under-investment in online channels like search can leave the audience “under-primed” for the Super Bowl ad, leading to lower response rates. On the other hand, spending very high amounts in the online channel, can eat into the budgets for the remainder of the year.

This is also a dilemma for advertisers who see predictable spikes. In these cases, a successful campaign entails careful planning and execution in the following areas:

1. Appropriate budgeting in both the online and offline channels for the entire campaign
2. Successful flighting of the advertisements in all channels before the high spike event to prime audiences
3. Capturing the awareness and demand generated during the high spike event to take consumers further down the funnel.

The third point, is crucial and often missed.

For instance, consider an advertiser with business model where the final conversion is online (for instance, an e-commerce advertiser). If the advertiser invests in a Super Bowl ad, but is under-invested online, then a savvy competitor could capture some of the generated demand by heavily investing online during and immediately after the Super Bowl.

Macro Economic Factors

Macro-economic factors can affect consumer and advertiser behavior in search quite dramatically. For instance, during the 2008 recession, the monthly advertiser spend in the financial vertical was strongly correlated to the S&P 500. Upswings in the economy are reflected by greater consumer spending, which advertisers see as higher ROI.

As a result , ad budgets increase and CPCs rise. During downswings, the opposite happens. In my experience, owing to the measurability of the medium, paid search is very sensitive to economic changes. Experienced search marketers do this and quickly adjust their bidding strategies and budgets when they believe the economy is going through some significant changes.

However, with talks of the fiscal cliff (now settled), we were in a period of economic uncertainty and not in a period of upswing or downswing.  Hence, search marketers would follow their campaign performance very closely in January and take immediate action when they see a trend in campaign performance.

EM spend in the financial vertical in the Adobe digital index and the stock market index for the 9 months through the 2008 recession. The correlation was close to 85% for this period.

Parting Survival Tips

• Adjust your bidding strategy to account for traffic shifts from December to January. If you have not changed your bids from late December, then you are at risk of spending more than your budget, as the same bid would get you more than the expected traffic.

• If you are in a period of high seasonality, build a plan to pace your budgets. For instance, if you anticipate a high spike event in late January or early February, then back loading your budgets might be a suitable strategy. I covered high seasonality forecasting in detail here.

• If you have a Super Bowl ad running in early February, work with the teams in other channels and build a media plan to prime audiences, generate demand and then effectively capture that demand when the Super Bowl ad has run. Make a plan to attribute the conversions effectively. This is a vast topic in itself, but I have presented the basics here.

• Economic news, forecasts, stock markets and new legislation, etc., can have a significant impact on your campaign performance. This period is a particularly sensitive one; hence, it pays to keep a close eye for such events.

I hope these tips help you navigate January and wish you success for the month and the rest of the year.

Perhaps people interact with them earlier in the sales funnel because these ads have images. If so, do they assist regular search ads? Further, do they contribute substantially to a search marketers campaign in terms of both volume and ROI?

I analyzed several PLA campaigns and found that advertisers would do well to measure PLA campaign performance from a multi-click attribution lens. But first, how much are retail advertisers spending on PLA?

PLA Spend Is Currently 8% Of Google Spend For Retailers

An analysis of PLA spend for several dozen retailers reveals that the mean percentage of PLA spend, as a fraction of total Google spend is about 8.4%, the median being 7.1%. There were some retailers where PLAs represent over 30% of spend.

Thus, for any retailer PLAs can no longer be ignored and needs the same attention (perhaps more, owing to the complexity of the auction process) as a regular search campaign needs.

Multi-click Assist Funnels Play A Big Role For PLA Campaigns

Assisted paid search funnels are those funnels where the user starts with one search term but converts on another search term after 2 or more searches. When I looked at paid search assist funnels in the past, assisted funnels formed only 10% of all funnels. This is quite different for PLA campaigns.

The graph above shows, the percentage of conversions that were assisted when a PLA campaign was involved versus those that were not.

For all three advertisers the percentage of assisted funnels was substantially higher when a PLA campaign was part of the funnel. This begs the question – how much does last click attribution affect PLA performance measurement?

Last Click Attribution Typically Under-reports PLA Performance By 15%

When analyzing funnels containing PLA clicks for several advertisers I found that funnels that begin with a click on a PLA ad, end with a conversion on regular search ad about 15% of the time.

In other words, it means that last click attribution would miss about 15% of all PLA conversions.

Note that for some advertisers, such as Advertiser 3, the number is much higher.

Key Takeaways

1. Efficient management of PLA campaigns matters more than ever before. With over 8% of spend on Google going to PLA for retailers, it has become very important for search marketers to manage their PLA campaigns well. This not only includes feed management but also campaign and bid management.
2. PLA campaigns work earlier in the search funnel than regular search ads. Since about 1 in 4 PLA funnels are multi click funnels, consumers on average are clicking on PLA ads earlier in the conversion path as compared to regular search ads, where multi click funnels form only 10% of all paths.
3. Multi click attribution really matters for PLA. Since 15% (often more) funnels that begin with a PLA ad end with a non PLA ad, looking at PLA performance from a last click perspective will underreport its performance by the same amount. To get a true understanding of the performance of PLA campaigns advertisers really need to look at PLA campaign performance with a multi click attribution lens.

An analysis of mobile traffic of a cross section of advertisers reveals up to 25-30% of all paid search traffic is now mobile. It is now more important than ever before to pay attention to your mobile targeted search campaigns. In this post, I shall talk about three key trends that every search marketer would do well to know.

Trend #1: Mobile Traffic Is Rising Fast… Faster Than You Think

The chart below shows the breakout of mobile clicks on Google between smartphone and tablet as a fraction of total traffic for a sample of our customers. While mobile has a small share of paid clicks in B2B, it comprises between 25-30% of clicks in the automotive and retail sectors.

That’s over one in four paid clicks! It is also interesting to note that while the bulk of the increase in mobile traffic is coming from tablets, a significant portion is also coming from smartphones.

Trend #2: Tablets Currently Have A Much Higher ROI Than Desktops

In a rational marketplace, the CPC of an ad is proportional to the ROI on the ad. In other words, if an ad unit has a good ROI, marketers are willing to pay more for the ad, and hence, its CPC increases. However, our research also shows that when it comes to tablet traffic, the market is not rational.

As smartphones don’t convert as well as desktops, advertisers are unwilling to pay a higher CPC on them. As a result, smartphone CPCs are about half of desktops. This CPC normalization means that ROIs on smartphones and desktops are almost identical.

However, in the case of tablets, CPCs are 30% lower  even though they convert 20% better than desktops. Thus, ROI on tablets is 70% better than desktops. This represents a significant opportunity for the savvy marketer.

Trend 2: ROI on tablet devices is much higher than desktops while the CPCs are disproportionately lower.

Trend #3: ROIs Can Vary Significantly By Mobile Device

Earlier this year, Mark Ballard from RKG showed that ROI varied significantly by mobile device,  i.e., the ROI on iPad, Kindle , Android tablet, etc. varied significantly. Our research shows a similar trend as we found that the ROI from iOS users is about double that of Android users.

The reason why the ROI differs so much is because of demographics, user experience, form factor and the context in which these devices are used.

ROI on ioS devices is 2x better than Android. A big part of the improved ROI can be explained by the higher conversion rates.

Tips to make your mobile search marketing program effective:

1. Analyze mobile traffic trends for your own campaigns. As the first chart shows, mobile traffic can vary significantly by the type of business one is in. In general, retail B2C has the highest proportion of mobile traffic, while B2B and Financial advertisers see the least amount of smartphone and tablet traffic. If you find a significant proportion of traffic coming from mobile devices, you need a mobile-centric paid search strategy.
2. Build tablet and smartphone specific campaigns. While it is tempting to take an existing campaign and target all devices, it is best to build tablet- and smartphone-specific campaigns despite the extra work. Since the CPCs and conversion rates are different, ads of the same keyword must be bid differently by device.
3. Go beyond the brand keywords. Many marketers think that the ROI on mobile devices is poor and only build brand keyword campaigns to maintain some presence on mobile devices. However, I have seen several instances of ROI-positive campaigns on smartphones driven by non-branded, mobile-targeted campaigns. This is something every marketer with substantial mobile traffic should try.
4. Try OS and device specific targeting. As the ROI on different mobile devices varies, it is worth trying OS and device specific targeting. Of course, this only makes sense when there is a substantial amount of mobile traffic, or else one ends up splitting a small fraction of mobile traffic with little to no incremental gain.
5. Build mobile friendly sites. Sounds obvious, but it’s remarkable how many marketers have not built mobile friendly sites. Quite understandably, among the many priorities marketers had, mobile-friendly sites were low on the list, given the small percentage of mobile traffic they saw in the past. But, with the rapid change in traffic patterns developing, these sites have become imperative.

I hope these tips help you build a successful mobile paid-search strategy and help you reach, engage and convert consumers in the new multi-screen, multi-device world.

For instance: is attribution always useful, what are the common attribution techniques, which is the best attribution model, what improvements in efficiency can one expect from the right attribution techniques? In this column, I shall attempt to answer some of these pressing questions.

Question 1: When To Invest In Attribution?

Many advertisers want to investigate cross-channel effects with multi-channel attribution. However, they use one tracking system for search, one for display and a third for email marketing.

These systems are often not compatible with each other, and it is often very difficult, if not impossible, to combine data for multi-event funnel analysis. It is, as they say in Cantonese, “Like a chicken talking to a duck.”

The first step, then, is to make sure you have a system that can track across all channels – search, display, SEO, social, etc. It is only then that one can apply an attribution rule. In the absence of a unified tracking system, attribution becomes difficult, and one has to resort to econometric techniques that are time consuming and labor intensive.

Still, even when you have a good tracking system in place, it might not be worth your time investing in multi-click attribution analysis. Consider the following examples:

Image 1

In the first example, 88% of funnels are single-click funnels. Hence, any multi-click attribution rule will only affect 12% of the funnels. The overall impact would be minimal. In the second example, 42% of the funnels are multi-click.

Hence, these attribution rule changes would have a significant impact on how revenue is measured and how keywords, campaigns and channels are optimized. As a general rule of thumb, I suggest that for any business where over 15% of the funnels are multi-click attribution, rules should be considered.

Question 2: Which Is The Best Attribution Rule?

There are several “generic” attribution rules available in attribution platforms – first click, last click, even weights, last more, parabolic, etc. I have shown these graphically below. Is there a method that ensures the “best” media mix? No.

In fact, any claim of a best method should be treated with suspicion. I have stated before, rules based attribution are not the optimum solution for the media mix problem.

However, from the point-of-view of campaign management and insight generation, especially regarding search, there are some factors to be considered.

First, when moving from last click, several keywords (typically brand keywords) which seem to generate high ROI will appear to generate a lower ROI with other attribution methods. This can be unnerving to management.

In these cases, a more phased approach for going from, say, last to first, or last to even, is to use a weight last more attribution. Second, a benefit of even distribution is that revenue can be attributed to more keywords that, in turn, encourage us to bid them to higher positions. Thus, keywords that are assisting and have been bid down because of their seemingly bad ROI would get a greater opportunity.

In the example below, changing the attribution rule from last to even distributes more revenue to keywords at lower positions. Hence, they would have a higher ROI in the new rule and would be bid higher.

Gainers and Losers with attribution rule change: In this example by moving from last click to first click, keywords at lower position would get more attributable revenue while those at higher positions would get less revenue. As a result, bidding with even click would encourage longer tail keywords.

Question 3: Which Attribution Rule Will Get Me The Right Media Mix?

As I mentioned before, rules based attribution are not the best solution for answering the media mix question. For one, attribution reallocates revenue (or any other response metric) to events that happened in the past, based on a certain way of counting revenue. It cannot answer what would have happened had you changed the counting system (say from last to first click).

Secondly, for most enterprise level marketers, the bulk of the budgets are spent offline in TV, print, etc. Factoring these channels in the media mix question requires econometric methods coupled with a statistical attribution approach.

So, are rules based attribution rules irrelevant for media mix questions? Not quite. They indicate where allocation of budget should be shifted. In the example below, we calculated revenue by even distribution and then by last click and quantified the assist effect as

If the Assist is positive, the channel is assisting, if negative, the channel is assisted. Positive assists should get more budget and negative assists, less budget. Note that the exact budget cannot be determined with this method. At best, we only have a partial directional answer to the media mix question.

For most enterprise search marketers, attribution will be increasingly important in the coming years. However, amidst these conversations, and numerous claims made by their proponents, it is important to know what commonly used attribution techniques can and cannot do. I hope these examples gave you a deeper understanding of this often-misunderstood subject.

Impression Trends for “tax software”. Note the sharp peaks at the end of January when W2 forms arrive and the day before tax day (mid April

For a SEM campaign manager, this situation is a mixed blessing. On the positive side, one knows which times of the year are important and can plan well in advance for it. On the negative side, the high seasonality period is a critical one as the performance of every single day matters.

If one can forecast performance well ahead of high period, one can plan various elements of the search campaign such as daily budgeting, revenue and order goals,  creative roll out, etc. Further, accurate forecasting alleviates a lot of the stress for the search manager as the right expectations are set throughout the organization.

Although advanced econometric methods when used with sophisticated forecasting algorithms can give very accurate forecasts, these techniques can be complicated as several factors need to be considered – the growth of the business, the competitive landscape, macro economic conditions.

However, when constrained with time, I often use the following three step forecasting method that gives very good results.

The Three Step Forecast Model

Step 1:  Build a baseline Ad Budget to Revenue Elasticity curve

The first step is to build a curve that estimates the expected revenue for different levels of spend just before the season starts. If you do not have an algorithmic way of doing this, then you can build such a curve by calculating average revenue for different levels of spend in the month preceding the high season in the past year.

The baseline which represents the market conditions, just before your promotion starts should represent market conditions just before the high season. An example curve is shown below:

Step 2:  Adjust Elasticity Curves with a 4 Factor Model

When the high season kicks in, the elasticity curve shifts because there is more demand for the product. Let us capture this effect mathematically. Consider the following equations that capture spend and revenue:

Where RPC= Revenue per Click

Note that the above equations have only 4 terms ; CPC, RPC, Impressions and CTR. When market conditions change these 4 factors change and as a result each point on the curve shifts.

Consider a point on the baseline curve (R_baseline,C_baseline). If this point shifts to (R_new,C_new) then the new point can be expressed as:

Thus, if I know the change in impressions, CPC , RPC and CTR on a given day versus baseline, I can calculate the new elasticity curve for that given day.

Since you have data from the previous years and access to tools like Google trends (that captures impression volume), you can build a table of daily deltas of the 4 factors and use it to build the daily elasticity curves.

Note in the example below you can see that days 9,10,11 are the peak sale days. Also, CTRs are assumed to be constant during the high season, which is a fair assumption unless you are planning to do TV advertising during this period. If that’s the case, CTR deltas should also be estimated.

Step 3:  Building out the forecast

The hard part is now done as you have daily forecasted elasticity curves for the high season period. Now it’s a matter of selecting the daily budget during the forecast that meets your goal.

For instance, if your goal is to ensure an ROI of 300% per day then based on the above data you get a spending schedule that looks like this:

Final Notes

1. The success of this method depends a lot on the elasticity curve. Hence, make sure you are confident that the curve is reasonably accurate. You can check the validity of the curve by checking if the curve predicts revenue well for different budget levels at times close to your coming high season.
2. Always correct your forecasts during the high season by normalizing the forecasted versus actuals for the previous days.
3. All calculations here have been done with Revenue as my KPI. However, this technique is not limited by the KPI. One can build the model with KPIs such as orders, margin or even a weighted combination of KPIs.
4. In the above example, we have tried to ensure that the ROI is above a threshold every day. We note that if the goal is to maintain a certain ROI for the entire high season, then an optimization approach is more appropriate than the approach above where it often pays to garner more volume at a lower ROI during the peak days and make up for it by spending less at higher efficiency in the lower volume days. However, this opens up the whole new topic of budget pacing which I shall leave for a future post.

This is a significant change for retailers as it means that a significant percent of traffic that has until recently been free until recently, is now paid. Thus it is imperative for the retailer to focus on effectively managing and optimizing Google Shopping campaigns.

How Google Shopping Works: A Very Short Summary

Google Shopping works differently from regular search ads. In the case of search ads, campaign, ad group, keywords are managed in Google Adwords where the advertiser can change bids.

To run a Google Shopping campaign, the advertiser has to also set up a product feed with Google Merchant Center. The feed contains a variety of product parameters including product id, image URL, price, availability etc.

On the Adwords side, the advertiser has to set up PLA campaigns and ad groups that contain product targets (as opposed to keywords) and offer text. These targets are linked to the GMC account which reference specific parameters on the product feed. However, like regular search campaigns, the advertiser has to place a bid to participate in the PLA auction.

When a user types a query, Google will map the query to a specific product/target combination in the PLA campaign in Adwords and will return a product listing ad based on the parameters of the selected products in the GMC feed.

Note that there are 3 distinct bits to managing your Google Shopping program:

1. Feed management
2. Bid management
3. PLA campaign management

I have observed several blogs discussing the importance of feed management but the other two points above have seldom been talked about. Are bidding and PLA campaign management important? Absolutely. We have observed some advertisers gain upto 80% lift through a mix a smart bid and feed management .

Five Best Practices In Bid & PLA Campaign Management

1.  The more granular the better

Since queries are mapped to individual PLA adgroup/target combinations in Adwords, it’s better to get as granular as possible. For instance, consider two different types of PLA structures below:

Ad group: Lawnmowers

Target>product_type=gas mowers

Target>product_type=electric mowers

Ad group: Pants

Target>product_type=levis AND  type=boot cut jeans AND label=mens

Target>product_type=joe’s jeans AND  type=printed denim AND label=womens

The second example has a higher degree of granularity which gives you control on the query-product listing mapping so that the right product is shown to the right query.

A step further would be to bid at the SKU level which is about as granular as one can get within their product feed. Bidding efficiently at the most granular target level will be the best way to maximize the potential of your feed.

I have noted that several blogs are against SKU level targeting, due to the complexity of campaigns and the difficulty in bid management. However, this is no different from long tail bidding in search.

One can use a variety of techniques such as hierarchical bid modeling to effectively bid on the PLA long tail. Indeed, in our experience, effective granular bidding can deliver significant performance lift, sometimes as much as 80%.

2.  Actively manage your bids

In many ways, PLA campaigns work like Content campaigns as the mapping of queries to specific listings is broad and theme based. Further, akin to content campaigns, bids are at the target level.

Neglecting bid management could prove disastrous as it could mean that you would be paying too much for your traffic or lose on highly qualified traffic as you are bidding too low.This again, is no different from regular search or content campaigns – active bid management is crucial for success.

3.  Always have a catchall target

Feeds are typically quite large and may change based on modifications to product classifications or new products added to an advertisers’ catalog.

Including a Catchall (or All Products) target within your PLA campaign will ensure that Google will be able to match search queries to relevant products in your feed.

The potential danger in omitting a Catchall target from your campaign is that search queries that would have otherwise not been able to be mapped to an existing target would be neglected and an advertiser’s product would not be shown.

One of the most important parts of a Catchall is to ensure its bid is less than that of the rest of the targets. Bid is one of the most (if not the most) important part of the auction process and having a higher bid on the Catchall can interfere with how queries are matched to their appropriate targets.

4.  Use negatives

It is imperative that advertisers utilize negatives to properly funnel the right queries to the right targets.

For example, consider an advertiser that sells small kitchen appliances and has a product target for coffee makers. A user then enters a search query for an 8 cup French press. Since Google performs a sort of broad match to the query it may serve the Catchall target rather than the coffee makers target.

As Google’s matching becomes more refined we believe that this sort of behavior will not be as prevalent, but it will help to add negatives to your Catchall that represent existing targets in your PLA campaign.

5.  Constantly monitor queries that are being matched to targets

As more advertisers start to invest in PLAs we believe that the competition for search queries will rise. Since the auction for PLAs will be completely separate from search, we believe that this is a great opportunity for new (and even existing) PLA advertisers to start off on the right foot.

Google allows advertisers to easily view which search queries are being matched to specific targets. Using this information can assist with placing negatives (as mentioned above), adding labels to specific products.

For example, if an advertiser only uses product level targets and sees that queries contain specific manufacturers then it may make sense for them to use labels in the feed to add another layer of specificity to their bidding strategy, and of course, input negatives based on irrelevant queries. These steps will be increasingly important as advertisers look to maximize their investment in PLAs.

While the current transition of Google Shopping represents a challenge due to the complexity of feed and bid management and the opacity of the query to product mapping, it is also an opportunity. I hope these tips help you get the most out of your PLA campaigns.

I wish to thank James Varughese, one of our crack analysts in helping me compile these best practices.

In display or social advertising, the marketer can reach customers with different offers  based on demographics or behavioral signals. This is not so in search where one only has the query searched as the fundamental mode of segmentation. This is rather limited as many different types of customers often search with a similar query.

For instance: someone searching with the term “shoes” might be a bargain hunter or a high value customer.

Hence, for the most part we can only create different offers, landing page experiences based on obvious search pattern differences, or a comparison-centric landing page for someone typing “discount shoes” vs. a brand-centric page for someone typing “Ugg boots”.

However, there are some subtle variances in search patterns that a savvy search marketer can leverage. I am going to share one of these with you today – singular vs. plural searches.

Customers searching with singular term are more often closer to a conversion than a customer searching with a plural term.* I hypothesize that this is because someone searching for a product in plural is both earlier in the sales funnel as well as likely to bargain hunt. There are a couple of analyses I use to back up my hypothesis.

First, when I looked at keyword performance data of over 40k keywords across several product lines, I found that the revenue per click of the plural search terms was significantly lower than their singular analogs.

In the above chart, we find that about 47% of spend goes to the singular terms, but they contribute to over 60% of revenue. The key reason for this is that the RPC (return per click) on these keywords is about 80% higher than the plural terms.

So should we write plural terms off? Not so fast! As mentioned before, searchers on plural terms are earlier in the sales funnel, and plural word searches are often. To back this claim, let’s do an attribution analysis of sales funnels. We use the following methodology.

Consider the following hypothetical sales funnel:

In the first calculation, only keywords that are involved in the final step (last click) of the sales funnel are given conversion credit. Hence, the keyword mortgages does not get any credit.

In the second calculation, all keywords involved in the sales funnel share the conversion credit i.e. its an equal click attribution. Considering the total of all sales funnels the keyword is involved in, when a keyword has more multi-conversion credits (the latter calculation) than final stage credits (the former calculation) it is a contributor.

If the situation is reversed, it is a beneficiary. Next we quantify this assist effect as:

∆ Rev=Last-EqualLast

Thus if a keyword is assisting then last is less than equal or ∆ Rev <0 and vice versa. When applying this methodology to an advertiser in the mortgage vertical we see the following pattern:

In all these cases, the singular keywords are benefactors i.e. on average they occur at the end of the funnel compared to their plural analogs.

Takeaways For The Marketer

• Consumers earlier in the sales funnel often search for products in the plural. The ad copy and landing page experience must reflect this mindset.

• Plural keywords will often seem less profitable than their singular analogs. However, when viewed from a multi-click perspective, they are more often than not driving demand to other keywords in the assisted funnel. Hence, use attribution technology to understand the full picture.

• Finally, while it may seem tempting to only have singular keywords for a small campaign, one should have both singular and plural versions of the keywords. As the first graph showed, in any search campaign of scale  plural keywords drive substantial percentage overall demand.

Thus, it is not a question of having plural keywords but of having the right bidding strategy and technology that aims to maximize a campaigns profitability while keeping volume needs in consideration.

That’s all y’all !

*Editors Note: Typo / Statement clarified per user comments since originally published erroneously.

Some people hate the very name of statistics, but I find them full of beauty and interest. Whenever they are not brutalised, but delicately handled by the higher methods, and are warily interpreted, their power of dealing with complicated phenomena is extraordinary. They are the only tools by which an opening can be cut through the formidable thicket of difficulties that bars the path of those who pursue the Science of man.

— Sir Francis Galton Natural Inheritance (1889), 62-3

Data analysis can be overhyped. We are taught to believe that with the right data analysis, we can understand everything around us and make rational decisions in our best interests. We are even shown how to do this in our schools and universities where we analyze datasets and are asked to infer insights.

The trouble is, the datasets we are taught on are unrealistic. Unlike toy datasets, real marketing data is often incomplete, sparse and some times even wrongly entered. So while it would be possible to come up with crystal clear conclusions if the data were robust, we are often forced to make decisions with partial or small datasets.

In these situations, we must combine the analysis with some heuristics and gut checks to come up with our conclusions. This is best understood with an example.

You are asked to find the quarter on quarter change in spend for  advertisers based on a small sample. (The data has been randomized and is not representative of any real advertiser.)

The simple thing to do here is to take the average across all advertisers and measure change i.e. measure the change on the total. The change in this case is -63%. So is this reflective of the marketplace? Let us explore this data a bit further.

Step 1: Directional Trends

Are we sure that the overall trend is positive or negative? If we breakdown trends by advertiser, we get this chart:

Note that 9 out of 11 advertisers show a drop in spend. Further, we know from our knowledge of the retail vertical that spend usually drops in Q1 wrt the Q4 holiday season. So we are confident that the spend trend is negative.

Step 2: Identification & Treatment Of Outliers

A closer look at the sample reveals that in Q4, over 50% of spend came from advertiser 9, the same advertiser dropped spend by 83%. In such cases, it is useful to measure the median change across the sample. Further, we can also measure the change without Advertiser 9.

Clearly, Advertiser 9 is biasing our results. Another way to check the impact of the advertisers is to measure the average drop without each advertiser present in the sample. So we measure the mean drop in spend without advertiser 1,2,3 etc.

Note how stable the estimates are for all but advertiser 9. Most estimates hover around -60% but without advertiser 9, it drops to -31%. Clearly advertiser 9 is biasing the results negatively.

Step 3: Cross Checking The Data

If possible, we should try to come to our estimate by other means. So we try another approach. If we order advertisers by spend and then calculate the change in spend cumulatively, what trend to we see?

Two things become very apparent:

1. Without advertiser 9, the sample represents 77% of spend in Q1 and the change in spend between Q4 and Q1 is -40%.
2. As we add larger advertiser the change in spend generally becomes increasingly negative. This indicates that larger advertisers dropped spend more than smaller advertisers in general; something that we might want to investigate further.

Step 4: Coming Up With The Estimate

This is the hardest and most controversial part. Here are the estimates that we have:

Given all these estimates, I would feel fairly comfortable in saying that spend has declined between 25% and 35% between quarters. Of course, I am mixing estimates with my gut feeling here and here is where it gets subjective.

In conclusion, when working with partial or sparse data, try to see the data from different angles rather than just the overall average.

Further, gut check your analysis with your domain knowledge. This is hard when your analysis shows something unexpected. If it is an unexpected number, it might really be a new insight or it might be a wrong conclusion.

At these times, your gut can really lead you to the right path or completely astray.

Hence, check your conclusions in different ways. The difference between an average analyst and the best ones, are that the best analysts know when to trust their instinct and when not to. That cannot be taught; it comes with experience.

This is a tricky question to answer for the following reasons:

1. TV campaigns are typically branding campaigns while search is typically DR focused.
2. There is typically a halo effect in TV campaigns, e.g. when you run a TV campaign you typically see its effect last on search for a duration longer than the TV campaign.
3. If the online and offline marketing teams are separate (they typically are) then the debate can be fractious.

While the econometric modeling is a vast subject, I want to present a simple method to help you answer the basic questions. We shall use the data of a direct-response advertiser that uses TV to drive people online and finish the sign up process there.

Simply put, econometric models  aim to quantify the effect of several variables into a metric of interest. The basic assumption of econometric models are that the variance (fluctuations) seen in the output are due to changes in certain input variables.

In our case, we shall assume that the controlling variables for the number of brand impressions on search are a linear function of search spend and TV spend.

Since non-brand searches are influenced by many factors (not least the search engines matching algorithms), we shall only consider branded keyword metrics in this exposition.

Further, since SEM spend on brand terms is almost constant (all keywords here were bid to position 1 for the duration of the campaign),the baseline search impressions (search impressions without TV ads) can be expressed as a function of SEM spend:

Since SEM spend is constant, we can then define the impression jump as:

This simplifies to:

This is a simple one variable regression and can be plotted as shown:

The graph shows a strong connection between search impressions and TV spend. For instance: it shows that when TV spend is $600,000, then the impressions in search jump by 70%.

At $1.4 million in TV spend, the impression volume jumps by 200%. The R^2 of 0.86 indicates that 86% of the change in branded search impressions in this example are explained by changes in TV spend.

Limitations:

1. This is a very simplified econometric model that does not account for time lag factors. If you are interested in more sophisticated time series models I would suggest looking into methods such as Holt-Winters, ARMA, ARIMA models.
2. I have chosen only two factors: TV and Search spend to explain the fluctuation in search impressions. There might be many other factors that I have not accounted for to build my model. An econometric model for a large brand typically includes several macro-economic indicators as well as other marketing channels such as PR and print. Variable selection for these advertisers is an art and science in itself. As a result, while the fit of my model is reasonably good, my prediction accuracy over longer periods with the model might be poor if a yet unaccounted for factor becomes meaningful.
3. You might wonder why I chose search impressions and not search conversions as my metric of interest. The reason is I typically find impressions regress much better than conversions when looking at the offline –online conversion. To connect conversions to offline spend, I would need to build a more robust model.
4. This model has a basic problem in that it assumes that the properties of data fed into it have the same properties regardless of time. In other words, it assumes that the effect TV and search spend will have on search impressions is the same if the day is a prior month’s or a prior years. It also does not account for seasonal factors. While this assumption works when analyzing data from a stable period, it is clearly not true when markets change due to macro economic factors as well as due to seasonality. A sophisticated way to over come this is to use time series model , where time is a factor in the models.

I hope my exposition helps you get started in analyzing the offline-online effect or other places where detailed granular data is not available. While not the most accurate or the most comprehensive model, I have had success good success with this simple method in connecting the dots between the online and offline world when the dataset is from a stable time period. I hope you find this simple model useful, too.