OMS: Optimizing Paid Search And Customer Acquisition

Let's tackle a question that we all eventually have to deal with. Somebody is requiring us to cut back on an advertising channel. Maybe the CFO did an analysis, and thinks that the ad-to-sales ratio in paid search is too high. She demands that we cut our paid search program by 50%.

The online marketer and web analyst work together to cobble conversion rate reporting, and if available, profitability reporting, attempting to make a case to continue to spend money. And this is a good thing. Your web analytics reporting helps you see what happened in the past. Your reporting shows you that you're losing a lot of money at the margin in your paid search program. It sounds like your CFO is right.

Your CEO, however, will want to know what will happen in the future if the CFO gets her way and cuts your paid search program in half. Your job is to make a case to the CEO to keep spending the money. You have to prove that there is a long-term return-on-investment that must be protected.

This is a perfect application of the Online Marketing Simulation, the "OMS" as I call it. Why not simulate what happens over the next five years if your paid search budget is cut by 50%?

First up, the current plan.

• Year 1 Demand = $19.1 million, Profit = $0.6 million.
• Year 2 Demand = $19.3 million, Profit = $0.7 million.
• Year 3 Demand = $19.5 million, Profit = $0.7 million.
• Year 4 Demand = $19.6 million, Profit = $0.8 million.
• Year 5 Demand = $19.6 million, Profit = $0.8 million.

This information alone would be good for your CFO and CEO to see. When is the last time you showed your Sr. Management team where your online business is heading, from a sales and profit standpoint, over the next five years? In the old days, you'd put your finger in the air and guess that a 35% sales increase would happen, and then you'd be praised when you had a 45% sales increase. Unfortunately, for most of us, those days are gone.

Ok, now we plug the 50% reduction in the paid search budget into the OMS.

• Year 1 Demand = $18.2 million, Profit = $1.0 million.
• Year 2 Demand = $17.5 million, Profit = $0.8 million.
• Year 3 Demand = $16.8 million, Profit = $0.6 million.
• Year 4 Demand = $16.1 million, Profit = $0.4 million.
• Year 5 Demand = $15.4 million, Profit = $0.1 million.

Which business would you rather be part of?

So often, our customer acquisition activities are unprofitable, and are the first area that the CFO wants to cut. If you're using your standard web analytics platform, you're going to look at conversion rates and average order values and you'll end up agreeing with your CFO.

If you run your customers through the OMS, simulating the long-term migration of your customer base, you'll arrive at a different answer. The example I illustrate above repeats itself across the data I analyze.

Short-term optimization frequently results in a long-term drain on the business. It is here that the web analytics community are sometimes sold the wrong message. We're told about these glorious optimization tests, we even read about how offline data is integrated into the multivariate tests that result in an optimized outcome. It all sounds really good.

Until we do a better job of simulating the long-term impact of our decisions, we won't know if we're actually optimizing our business, or if we're hurting the future of our business. It is time for the online marketing and web analytics community to take the next step!

OMS: They Bought Via An E-Mail Campaign. Now What?

You have a customer who purchased online a couple of years ago. Since then, you've sent this customer an opt-in e-mail message, twice a week, 200+ in total.

Today, for whatever reason, that customer is ready to purchase something. Through no fault of her own, the customer receives another e-mail campaign from us. The customer has a choice. Should she click through the e-mail message and purchase? Or should she just key in your website url and purchase?

Does it even matter?

Let's go to the Online Marketing Simulation and find out.

First, we simulate 1,000 2x buyers, first purchase online, second purchase online, both purchases = $150, both purchases from Merchandise Division #3. This will be our benchmark.

• Annual Repurchase Rate = 41%.
• Demand: Year 1 = $104,000, Year 2 = $68,000, Year 3 = $48,000, Year 4 = $38,000, Year 5 = $33,000. Total = $291,000.
• Online Buyers via Offline Source: Year 1 = 124, Year 2 = 75, Year 3 = 49, Year 4 = 37, Year 5 = 31.
• Online Buyers via E-Mail: Year 1 = 162, Year 2 = 88, Year 3 = 55, Year 4 = 41, Year 5 = 34.
• Online Buyers via Search: Year 1 = 32, Year 2 = 20, Year 3 = 13, Year 4 = 10, Year 5 = 8.
• Online Buyers, Pure Web: Year 1 = 99, Year 2 = 53, Year 3 = 34, Year 4 = 26, Year 5 = 22.

Now, we'll simulate what happens if these 1,000 customers instead place their second order via E-Mail.

• Annual Repurchase Rate = 43%.
• Demand: Year 1 = $102,000, Year 2 = $79,000, Year 3 = $63,000, Year 4 = $54,000, Year 5 = $48,000. Total = $346,000.
• Online Buyers via Offline Source: Year 1 = 129, Year 2 = 85, Year 3 = 62, Year 4 = 50, Year 5 = 43.
• Online Buyers via E-Mail: Year 1 = 204, Year 2 = 117, Year 3 = 80, Year 4 = 61, Year 5 = 51.
• Online Buyers via Search: Year 1 = 26, Year 2 = 17, Year 3 = 13, Year 4 = 10, Year 5 = 9.
• Online Buyers, Pure Web: Year 1 = 100, Year 2 = 63, Year 3 = 45, Year 4 = 35, Year 5 = 30.

If the customer converts to an E-Mail purchase, then future value is increased by about $50,000 ... or $50 per customer, so that's a good thing (your mileage will vary). Now look at a sampling of the key online micro-channels. Customers buying from offline sources are not significantly changed. Customers, however, become much more likely to buy via E-Mail (duh).

But there's a more interesting outcome when we look at merchandise divisions. Let's look at the outcome for Merchandise Division #5.

• Pure Web Buyer: Year 1 = 86, Year 2 = 73, Year 3 = 55, Year 4 = 45, Year 5 = 40.
• Pure Web + E-Mail Buyer: Year 1 = 171, Year 2 = 111, Year 3 = 84, Year 4 = 69, Year 5 = 61.

This is why we focus on the Online Marketing Simulation, the "OMS". The simulation tells us how customers are likely to behave in the future because of an action that happened in the past. Combined with Web Analytics, OMS yields very interesting insights, insights that help business leaders make decisions today that profitably influence the future.

In this case, when we encourage a customer to purchase from an e-mail campaign, we change the future merchandise preference of the customer. The executive in charge of Merchandise Division #5 should be partnering with the e-mail marketing team, given the synergy identified in the OMS run.

An OMS analysis complements your Web Analytics enviornment. With Web Analytics, you are easily able to look back in time. With OMS, you get to see the future.

OMS: Keyword Optimization

Any online marketer and web analytics practitioner looks to maximize the performance of her business. Typically, this is done by analyzing campaign performance. The marketer and web analytics practitioner work together to make sure that current activities are optimized.

The Advanced Web Analytics Practitioner knows that all current activities cause changes in future activities.

For instance, a set of keywords may yield lower costs, higher conversion, and as a result, an increase in short-term profit. From an SEO, Online Marketing, and Web Analytics perspective, this is all good.

From an Online Marketing Simulation (OMS) standpoint, it may be good, it may not be good!

In my dataset, Merchandise Division #4 represents a product line with more expensive price points. Merchandise Division #3 is an extension of Merchandise Division #4, with less expensive price points. Not surprisingly, customers flock to Merchandise Division #3!

So, let's run simulations of future performance on 1,000 customers. Remember, we categorize customers based on combinations of future value, advertising channel preference, physical channel preference, and merchandise preference. This yields anywhere between maybe 80 and several thousand segments. Once done, we apply twelve-month sales and profit value to each segment, we migrate customers to their new segment, then we replicate the process four more times, with the ultimate destination being the segments that these 1,000 customers will reside in five years from now.

The first simulation is for a paid search customer buying from Merchandise Division #3, the division with lower price points.

• Annual Repurchase Rate = 24%.
• Demand: Year 1 = $51,000, Year 2 = $35,000, Year 3 = $28,000, Year 4 = $25,000, Year 5 = $23,000.
• Merchandise Division #3 Buyers: Year 1 = 128, Year 2 = 72, Year 3 = 52, Year 4 = 45, Year 5 = 42.
• Merchandise Division #4 Buyers: Year 1 = 113, Year 2 = 65, Year 3 = 48, Year 4 = 41, Year 5 = 38.

Ok, now let's run simulations of future performance on 1,000 customers buying from Merchandise Division #4 via paid search --- this division has similar product, but higher price points.

• Annual Repurchase Rate = 42%.
  Demand: Year 1= $123,000, Year 2 = $76,000, Year 3 = $52,000, Year 4 = $41,000, Year 5 = $35,000.
• Merchandise Division #3 Buyers: Year 1 = 169, Year 2 = 103, Year 3 = 70, Year 4 = 54, Year 5 = 46.
• Merchandise Division #4 Buyers: Year 1 = 258, Year 2 = 137, Year 3 = 87, Year 4 = 66, Year 5 = 55.

The typical web analytics practitioner partners with the online marketer, seeking to optimize conversion rates and transaction profitability. Once the transaction is complete, the typical web analytics practitioner and online marketer moves on to the next conversion.

In this case, which customer are you willing to pay more to acquire? I'm willing to pay a fortune for the customer who purchases from Merchandise Division #4. I'll gladly sub-optimize my short-term business in order to acquire a customer that will spend more in the future in both Merchandise Divisions!

This is what Advanced Web Analytics via the Online Marketing Simulation (OMS) is all about. We want to understand how our optimized short-term decisions impact the long-term health of our business.

Keyword Optimization requires a view of long-term performance, in order to be successful. The OMS environment can point you in the right direction.

Presentation On Web Analytics And OMS

Download this presentation on how Web Analytics and the Online Marketing Simulation work together to give you a peek into the future of your online business.

There are more than three-dozen slides in the .pdf file, chocked-full of ways that I'm asked to apply the OMS to the challenges CEOs face when trying to understand where their online business is heading over the next five years. The future sales trajectory of the online channel has become a really hot topic, now that organic online growth is drying up.

If you're technically adept, you can take the framework I outlined and create your own OMS. Or, give me a holler, and I'll be glad to build an OMS that outlines the future of your online business.

Let me know what you think of the presentation --- especially those of you in the Web Analytics community, what are your thoughts?

OMS: Shopping Cart Abandonment

Web Analytics software and practitioners do a very nice job of illustrating the rate that various segments of customers abandon shopping carts.

The next step we can take on this journey is to understand what the consequences are of an abandoned shopping cart. In other words, given that a customer abandoned a shopping cart, we need to measure what happens next, and then quantify the sales and profit impact of what happens next.

Recall the OMS framework authored over the past few weeks ago. We can add variables to this framework. Add dummy variables that tell whether the customer abandoned a shopping cart in the last week (1 = yes, 0 = no), last month (1 = yes, 0 = no), or even last year (1 = yes, 0 = no). Add these variables to your pre-post datasets, and run them through your factor analysis, integrating shopping cart abandonment with channel and merchandise preferences. Enter the variables into your regression models, too, while you're at it. Essentially, you make shopping cart abandonment part of the 100 to 1,000 segments that forecast future customer behavior.

Or define the information however you like ... no rules here!

At this point, you'll be able to run simulations that show how customers evolve, based on past shopping cart abandonment activity.

For instance, we can compare four customers. We'll assume that prior to last month, all other attributes are equal:

1. Customer purchased merchandise online last month.
2. Customer abandoned a shopping cart last month.
3. Customer visited website last month, didn't put anything in shopping cart.
4. Customer didn't bother to visit website last month.

For each of the four customers, we run a five year sales simulation, based off of past customer behavior. For 1,000 simulated customers, you might find an outcome that looks like this (your mileage will vary):

1. Purchaser: Year 1 = $100,000, Year 2 = $70,000, Year 3 = $50,000, Year 4 = $35,000, Year 5 = $25,000, Total = $280,000.
2. Shopping Cart Abandoner: Year 1 = $85,000, Year 2 = $65,000, Year 3 = $45,000, Year 4 = $35,000, Year 5 = $25,000, Total = $255,000.
3. Visitor, No Cart: Year 1 = $60,000, Year 2 = $40,000, Year 3 = $30,000, Year 4 = $20,000, Year 5 = $15,000. Total = $165,000.
4. No Visit: Year 1 = $40,000, Year 2 = $25,000, Year 3 = $15,000, Year 4 = $12,000, Year 5 = $10,000. Total = $102,000.

See, within the OMS context, we get a different level of business intelligence. In this example, an abandoned shopping cart costs us $25,000 of demand per 1,000 customers ... a loss of $25 per customer, over five years.

But worse, look at the website visitor who doesn't even make it to the shopping cart. In this case, the unconverted customer (no shopping cart, no purchase) spends $115,000 less per 1,000 customers ... a loss of $115 per customer, over five years.

Of course, even the unconverted visit is worth something. We get an incremental $63,000 per 1,000 simulated customers when a customer visits, vs. no visit at all. In other words, OMS is projecting in this instance that there is a downstream value to every action on a website, and that value can be calculated via five-year simulations. In this case (and remember, your mileage will vary --- use the principals here to estimate the numbers for your own business).

• An visit is worth an incremental $63,000 per 1,000 customers over five years, $63 per customer.
• An incremental item into the shopping cart adds $90,000 per 1,000 customers over five years, $90 per customer.
• An incremental purchase adds $25,000 per 1,000 customers over five years, $25 per customer.

In Web Analytics, we look back in time to report what happened. In OMS, we look forward, simulating a likely future outcome based on what happened in the past. Combined, Web Analytics and OMS make good business sense, and provide the answers a CEO frequently looks for.

If this style of shopping cart abandonment analysis makes sense to you, contact me for details on an OMS project!

OMS: Up-Selling and Cross-Selling

Up-Selling and Cross-Selling merchandise to a customer is an established e-commerce best practice, right? Retailers love to capture the additional margin dollars at the end of a transaction by offering the customer the opportunity to add an item to her order.

With a standard Web Analytics software tool, we can identify how customers respond to a cross-sell or up-sell opportunity.

With the Online Marketing Simulation (OMS), we can see what the downstream impact is of a customer who responds to an up-sell or cross-sell opportunity.

Let's simulate 1,000 new customers who purchased three items for $50 each via paid search. In my dataset, I have eight merchandise divisions. Let's assume that the customer purchased three items from merchandise division #3. Here's a five year simulated run for this customer:

• 12 Month Repurchase Rate = 24%.
• Future Demand: Year 1 = $51,000, Year 2 = $35,000, Year 3 = $28,000, Year 4 = $25,000, Year 5 = $23,000.
• Merchandise Division #1 Buyers: Year 1 = 22, Year 2 = 15, Year 3 = 12, Year 4 = 11, Year 5 = 11.
• Merchandise Division #2 Buyers: Year 1 = 26, Year 2 = 18, Year 3 = 15, Year 4 = 13, Year 5 = 12.
• Merchandise Division #3 Buyers: Year 1 = 128, Year 2 = 72, Year 3 = 52, Year 4 = 45, Year 5 = 42.
• Merchandise Division #4 Buyers: Year 1 = 113, Year 2 = 65, Year 3 = 48, Year 4 = 41, Year 5 = 38.
• Merchandise Division #5 Buyers: Year 1 = 66, Year 2 = 42, Year 3 = 34, Year 4 = 30, Year 5 = 28.
• Merchandise Division #6 Buyers: Year 1 = 20, Year 2 = 15, Year 3 = 13, Year 4 = 12, Year 5 = 11.
• Merchandise Division #7 Buyers: Year 1 = 28, Year 2 = 19, Year 3 = 15, Year 4 = 13, Year 5 = 13.
• Merchandise Division #8 Buyers: Year 1 = 36, Year 2 = 25, Year 3 = 19, Year 4 = 17, Year 5 = 16.

Ok, those metrics don't have much meaning unless you have something to compare them to. Now let's assume that you, the online marketer, were able to cross-sell or up-sell this customer one additional item, a $50 item in Merchandise Division #7.

How does this one additional item, a $50 item in Merchandise Division #7, impact the future trajectory of this customer? We'll pop the results into the OMS ... let's see what the simulation tells us!

• 12 Month Repurchase Rate = 41%.
• Future Demand: Year 1 = $104,000, Year 2 = $68,000, Year 3 = $48,000, Year 4 = $38,000, Year 5 = $33,000.

Let's just stop right there. The simulation suggests that a new paid search customer with this one additional item from a different merchandise division is instantly worth between 50% and 100% more. Clearly, your mileage will vary, some of you will experience no incremental long-term value, some of you will experience double or triple this outcome. The goal, of course, is for you to strategically think whether this issue has applicability to your business.

Here's how customers purchased from the eight merchandise divisions in my dataset.

• Merchandise Division #1 Buyers: Year 1 = 37, Year 2 = 28, Year 3 = 21, Year 4 = 18, Year 5 = 16.
• Merchandise Division #2 Buyers: Year 1 = 39, Year 2 = 33, Year 3 = 25, Year 4 = 21, Year 5 = 19.
• Merchandise Division #3 Buyers: Year 1 = 227, Year 2 = 123, Year 3 = 79, Year 4 = 61, Year 5 = 53.
• Merchandise Division #4 Buyers: Year 1 = 225, Year 2 = 119, Year 3 = 78, Year 4 = 61, Year 5 = 52.
• Merchandise Division #5 Buyers: Year 1 = 86, Year 2 = 73, Year 3 = 55, Year 4 = 45, Year 5 = 40.
• Merchandise Division #6 Buyers: Year 1 = 36, Year 2 = 28, Year 3 = 22, Year 4 = 19, Year 5 = 17.
• Merchandise Division #7 Buyers: Year 1 = 44, Year 2 = 35, Year 3 = 25, Year 4 = 21, Year 5 = 18.
• Merchandise Division #8 Buyers: Year 1 = 55, Year 2 = 43, Year 3 = 32, Year 4 = 26, Year 5 = 23.

Pay close attention to Merchandise Division #7. This was the division where the up-sell / cross-sell item was purchased from. There is some improvement in the number of buyers in this division. Now pay attention to Merchandise Divisions #3 and #4 --- these divisions are expected to get a significant bump in customers.

Yes folks, your actions in one area of your business cause unexpected changes in the business performance of another area of your business. In this example, the cross-sell of one item from Merchandise Division #7 causes Merchandise Division #4 to experience a significant improvement in performance within this customer segment (in fact, almost all divisions experience improvement). Merchandise Division #4, of course, had no role in either simulated transaction --- it simply benefits from something that happens elsewhere in your business.

Would you make different business decisions if you knew how this dynamic impacted your e-commerce business?

This is what Advanced Web Analytics, specifically, the Online Marketing Simulation (OMS) is all about. We're looking to see how the decisions we make today impact the future of our business. Most Web Analytics applications look backward, measuring what happened in the past. The OMS environment looks forward.

OMS: Nuts And Bolts, The Algorithm!

I promised some details!

Here's what I like to do. I'll take client data, and create a set of attributes that are of interest to the management team. In the dataset I'm currently working on, here are the attributes:

• Recency: Months Since Last Purchase.


• Demand12: Demand Spent In Past 12 Months.


• Demand99: Demand Spent 13+ Months Ago.


• Price_Item: Average Price Per Item Purchased.


• Items_Order: Average Number Of Items Per Order.


• 24 Channel Attributes: This business has twelve purchase channels (i.e. paid search, affiliates, etc.). I create 1/0 (1=yes, 0=no) indicators that tell if the customer purchased from that channel in the past 12 months, and if the customer ever purchased from that channel 13+ months ago. Dollar values can also be used instead of 1/0 indicators.


• 16 Merchandise Attributes: This business has eight tabs, if you will, across the top of the homepage, representing eight merchandise divisions. I create 1/0 (1=yes, 0=no) indicators that tell if the customer purchased from that merchandise division in the past 12 months, and if the customer ever purchased from that channel 13+ months ago. Dollar values can alos be used instead of 1/0 indicators.

So, this dataset has 45 variables, one row per customer. The file contains data through today.

Next, I need to reduce the dimensionality of the database. There's literally an infinite number of ways to combine the 45 variables, right? Somebody could have last purchased 1 month ago, spending $64.95, whereas another customer could have last purchased 1 month ago, spending $61.95.

I do this via a combination of Logistic Regression (Response), Ordinary Least Squares Regression (Spend), and "Factor Analysis" (Merchandise and Channels). Yes, I realize this is geeky.

The combination of Logistic Regression, Ordinary Least Squares Regression, and Factor Analysis result in a series of "strategic segments". Each segment is a combination of customer quality, channel preference, and merchandise preference. Some of the segments are very responsive, some are not responsive. Some buy from all merchandise divisions, some only buy from one merchandise division, and have a specific channel preference.

For smaller companies, I limit the number of segments to 100 or less. For bigger companies, I'll use 1,000 or more segments ... it all depends upon how many customers end up in each segment.

Once I determine what segment a customer resides in, I create a brand new dataset, replicating every variable for every customer as the customer looked exactly one year earlier (in social media, you might use a week timeframe instead of the yearly timeframe we use in e-commerce). I create the same segmentation strategy, and assign a segment id to each customer based on the way the customer looked last year.

Now each customer is assigned to a segment from one year ago, and a segment today. I aggregate this dataset down to every last-year / this-year segment combination (100 x 100 = 10,000 segment combinations). When simulated over five years, there ends up being 100 x 100 x 100 x 100 x 100 = lots of combinations!

Once I have the 10,000 segment combinations, I can take a sample customer (i.e. first time buyer purchasing an iPod via paid search), and simulate how that customer will migrate and evolve over the next five years. I can see what merchandise that customer will buy in the future, I can see the channels the customer will purchase from in the future, and I can calculate the incremental demand and profit the customer will generate.

Best of all, I can compare this customer vs. any other customer, to see how customers will evolve. Will a paid search iPod buyer evolve differently from an e-mail inkjet printer buyer? Will either customer use a retail channel in the future? Am I unwittingly altering the future trajectory of my business by optimizing for inexpensive keywords?

These are the problems that CEOs are asking me to solve for them, problems not easily answered by a typical Web Analytics toolset.

At a 30,000 foot level, that's the nuts and bolts behind the Online Marketing Simulation (OMS) that I've developed.

As we work through future examples, consider the following questions:

1. Can my Web Analytics software tool do this analysis?


2. Can my Web Analytics analyst do this analysis for me?


3. Can my Web Analytics software vendor do this analysis for me?


4. Can the leading Web Analytics consultants / bloggers do this analysis for me?

If the answer to each question is "no", then the Online Marketing Simulation (OMS) is something you'll want to investigate.

OMS: The Online Marketing Simulation

This is the fourth part of our series on Advanced Web Analytics and Online Marketing Simulations (OMS).

The goal of an Online Marketing Simulation (OMS) is to help us see how decisions that are made today influence the long-term health of our business. We're going to use an analysis process that is not commonly, if ever, used in Web Analytics.

We manage the Online Marketing Simulation (OMS) by linking conditional probabilities, simulating how a group of customers are likely to evolve over the next five years (or if you're analyzing social media, maybe the next five days!).

Let's look at a very simple example. You have three micro-channels in your online business.

1. Online Orders via E-Mail Marketing
2. Online Orders via Paid Search
3. All Other Online Orders

In this simple example, 10,000 customers purchased in 2007 via paid search, not purchasing via e-mail marketing or via any other method of generating an online order. We follow the 10,000 customers to see how they evolve during 2008. Let's assume that the 10,000 customers migrated as follows in 2008:

• 6,000 did not purchase during 2008.
• 1,500 purchased via all other online orders.
• 1,200 purchased via paid search.
• 200 purchased via paid search and all other online orders.
• 700 purchased via e-mail marketing.
• 100 purchased via e-mail marketing and all other online orders.
• 200 purchased via e-mail marketing and paid search.
• 100 purchased via e-mail marketing, paid search, and all other online orders.

Since we are analyzing three micro-channels in this example, all via yes/no indicators, we have 2*2*2=8 possible future outcomes.

The majority of customers (6,000 of the 10,000, 60%) did not purchase during 2008.

Notice that 1,700 customers purchased via paid search during 2008. This is one of the interesting things that we don't take into account when using the web analytics tools from the leading paid and free vendors to measure conversion rates --- we don't factor in how today's actions influence tomorrow's business. In this example, 10,000 paid search customers in 2007 yield 1,700 paid search customers in 2008.

Are you budgeting for future paid search activity that you are causing because of today's paid search optimization activities?

This is what the Online Marketing Simulation (OMS) environment does. We look at the future trajectory of all customers. Instead of looking at three dimensions (paid search, e-mail, all other), we look at a dozen or two dozen or more dimensions. We look at many combinations of prior activity, measuring the percentage of customers who migrate to a future state of activity. And we don't have to look only at advertising micro-channels, we can fold in the merchandise categories the customer purchases from (or views online if you wish). Once each customer is placed in his/her future state, we replicate the process, showing where the customer will migrate in year two, then year three, then year four, then year five.

In the example above, we can estimate how much paid search expense we will incur over the next five years because of today's paid search and conversion rate optimization practices. We can estimate how many customers will purchase via e-mail marketing over the next five years because of today's paid search and conversion rate optimization practices. We can see how one merchandise category will grow or shrink if we change our e-mail marketing strategies. We can sum demand, expense, and profit, short-term and long-term.

In the next OMS post, we'll begin to work through an actual dataset with numerous dimensions, so that you can see how the Online Marketing Simulation (OMS) environment really works. The example will be representative of the type of consulting I do for clients, helping them understand how the online channel will evolve based on today's decisions.

As we work through examples over the next month, ask yourself four questions after each post:

1. Can my Web Analytics software tool do this analysis?
2. Can my Web Analytics analyst do this analysis for me?
3. Can my Web Analytics software vendor do this analysis for me?
4. Can the leading Web Analytics consultants / bloggers do this analysis for me?

If the answer to each question is "no", then the Online Marketing Simulation (OMS) is something you'll want to investigate.

From Conversion Rate to Repurchase Rate to Multiple Probabilities

We're up to the third part in our series on Web Analytics and Online Marketing Simulations (OMS).

My central thesis is that by emphasizing conversion rates, we optimize our business based on the advertising sources that cause a customer to purchase now. By doing this, we create an inefficiency. We overlook customers who yield a positive outcome in what we deem an inefficient manner.

Here's an example.

• We all know that pay-per-click customers convert at less-than-thrilling rates.
• We know that pay-per-click customers can be expensive, maybe costing us $0.10 per click, or $0.40 per click, or $0.70 per click.
• Over time, pay-per-click customers can become e-mail subscribers.
• And e-mail subscribers often have higher-than-average conversion rates if they click-through an e-mail campaign.
• And e-mail marketing is really close to free, having virtually zero variable cost.

If we want to optimize conversion rates, we'll steer ourselves away from expensive pay-per-click programs with low conversion rates, right? At the same time, we'll want to maximize our e-mail marketing program, with low costs and high conversion rates.

If we want to optimize repurchase rates, we'll take a different action. We want pay-per-click customers, because pay-per-click customers become e-mail subscribers. We want to optimize the multi-year process of acquiring an expensive pay-per-click customer who becomes a profitable e-mail customer.

If we optimize via conversion rate, we won't "seed" our business with the pay-per-click customers who become e-mail subscribers with high conversion rates. We optimize our business in the short-term, but create a long-term inefficiency that limits our ability to grow over time.

We have an opportunity to add to our responsibilities. We have an opportunity to measure what are called "conditional probabilities". Here are examples of conditional probabilities:

• What is the probability of a customer becoming an e-mail subscriber, given that she last purchased via pay-per-click?
• What is the probability of a customer becoming a loyal customer, given that she has become an e-mail subscriber?
• What is the probability of a customer buying from multiple channels, given that she has become a loyal customer?

By linking each of these conditional probabilities, we arrive at a customer that generates an optimal amount of profit, over time. Within each step, we may have numerous instances of sub-optimal conversion rates, but those sub-optimal situations result in a customer that is optimally profitable. We combine conditional probabilities with demand and profit calculations, allowing us to simulate the future based on the actions we manage today.

From an Advanced Web Analytics standpoint, we create a table that records customer actions in a prior period of time, and in a future period of time. In the future period of time, we also tag the amount of demand the customer generated in the future period of time. The list of variables below is not exhaustive, and variables can be combined (receive catalog, buy via pay-per-click), creating what I call "micro-channels".

Prior Period of Time (1 = yes, 0 = no).

• Did customer visit the website?
• Did customer put merchandise in a shopping cart?
• Did customer purchase from the website?
• Did customer purchase multiple times from the website?
• Did customer purchase via e-mail?
• Did customer purchase via pay-per-click?
• Did customer purchase via affiliates?
• Did customer purchase via offline catalog marketing?
• Did customer purchase via display ads?
• Did customer purchase via social media?

The same set of variables are replicated for a future period of time, along with demand and profitability (if available) metrics. Obviously, the same customer will not have the same attributes, as customer behavior changes.

The prior timeframe is usually defined as a year, the future timeframe is usually defined as a year. That being said, there's no reason you cannot explore different timeframes, weeks, months, seasons, etc. However, you identify more inefficiencies, more opportunities for profit, when you lengthen the timeframe.

When the dataset is created, the analysis begins. We begin to link the conditional probabilities together, finding customer behavior that leads to long-term sales and profit. In our next blog post in this series, we'll begin to explore how this information comes to life.

From Conversion Rates To Repurchase Rates

In Moneyball, Michael Lewis explained how the Oakland A's, with a very low payroll, optimized wins by finding market inefficiencies. Baseball historically focused on metrics like batting average and runs batted in. Oakland focused on "on base percentage" and "slugging percentage", finding athletes who generated good outcomes in these metrics at lower-than-average salaries.

In other words, the existing best practice was to find great athletes with good batting averages and an ability to drive in runs. Oakland identified a different set of metrics, and then found players who were good at generating these metrics at a low cost. Oakland won a ton of games from 1999 - 2006, using this methodology.

In Online Marketing, we look to optimize conversion rate, and we have the best set of tools we've ever had to do this style of optimization.

But we're not making big strides in understanding how to increase customer spend over time. In other words, we work really hard to increase conversion rates, maybe from 4.1% to 4.5%. But we somehow aren't able to engage customers in a way that increases loyalty. E-commerce sales have largely grown from traffic, not from increases in repurchase rates, orders per buyer, items per order, or price per item. In the future, growth must come from increases in repurchase rate, orders per buyer, items per order, and price per item.

So if you are an online marketer, I'm going to encourage you to evolve your thinking. I'm not going to ask you to abandon all of the metrics and optimization strategies you've historically employed. I am going to ask you to think differently.

Let's start by defining a metric. The name of the metric is "Repurchase Rate". Simply put, this metric is defined as the percentage of customers who purchased last year, and then purchase again this year. Now if your business is not an e-commerce business, then go ahead an think about whatever the "action" is that you want to maximize --- if you are Twitter, you might look at a "Re-Use Rate", how many people use your service again today, given that they used your service yesterday.

Why is "Repurchase Rate" important? Let's look at two customers. Both customers purchased one time during 2007, on December 10, 2007:

• Customer #1: Visit 2/1/2008, Visit 2/8/2008, Visit and Buy 2/12/2008, Visit 7/1/2008, Visit and Buy 7/2/2008, Visit 9/10/2008, Visit 10/1/2008, Visit 12/1/2008.
• Customer #2: Visit and Buy 2/15/2008, Visit and Buy 7/2/2008, Visit 12/1/2008.

Both of these customers have a 100% "Repurchase Rate", and both customers ordered two times during 2008. Both customers last visited the website on 12/1/2008. In many ways, both customers yield the same outcome --- both customers purchased twice during 2008.

But from a "Conversion Rate" standpoint, these customers are very different. Customer #1 has a much lower conversion rate than does Customer #2. Our web analytics tools are often configured to favor Customer #2.

When we favor Customer #2, we favor the actions that cause Customer #2 to come to our website. And as a result, we will spend more money, via optimization, on the actions that generate a lot of customers who look like Customer #2.

So my thesis is this: Why not look for the actions that generate customers who have good Repurchase Rates? By optimizing "Repurchase Rate", a metric measured across a multi-month or multi-year period of time, we find customers who may look bad when measured via "Conversion Rate", but are equally or more valuable to the long-term health of the business.

In other words, there is a market inefficiency that exists when everybody focuses on "Conversion Rate". By instead focusing on "Repurchase Rate", we identify customers who appear to be poor converters, but spend the same amount in the future as do other customer who convert well. The secret is that we can grow our business faster than our competition, because we are optimizing on a different set of measures.

Next week, we'll begin to explore the math that allows us to optimize via "Repurchase Rate". The math will lead us to a simulation environment that helps us understand the long-term impact of short-term decisions.

Web Analytics, Online Marketing, Metrics

Note: Every Monday and Wednesday in August, we'll be talking about Web Analytics and Online Marketing Simulations (OMS). The discussions will build in depth and math as the month moves along.

There's never been a better time to be a Web Analytics expert!

There's never been a more challenging time to be a Web Analytics expert, either.

Do you want to know how customers who visit your site via iPhones convert? No problem! Care to measure the bounce rate of new customers? Have at it! Need to know the cost per conversion of visitors from Wyoming visiting because of PPC? Easy!

Online marketing is calibrated around a metric called "conversion rate". We are able to segment visitors via a veritable plethora of dimensions, using a mouse to drop in metrics based on dimensions we define on the fly. It's easy, it's fun, it is actionable, it increases sales! More important, we're able to create A/B tests, allowing us to optimize our results in real time. Technically, we know more about how customers interact with advertising than ever before. What's not to like about being a Web Analytics expert?

That's one side of the spectrum.

Over on the other side of the spectrum are what I call "strategic questions". Strategic questions are a different animal altogether. Strategic questions are harder to answer, because the answer isn't found by using metrics to optimize conversion rate.

Strategic Question #1: You are a retailer with three stores in Akron, OH. Management is considering closing one store, and is hoping that the other two stores and your e-commerce website will pick up the sales lost if one store is closed. What is the impact on website sales if one store is closed?

Strategic Question #2: Your merchandising team added a new product line in July. This product line is already responsible for 5% of company sales, a huge success. However, the product manager for an existing product line experienced a 25% reduction in sales in July. She believes that the new product line cannibalized her assortment. Other product managers don't agree, because their products experienced sales gains during July. Nobody tested offering/not-offering the new product line to customers. What impact did the new product line have on the old product line?

Strategic Question #3: You implemented five new initiatives. Each initiative increased conversion rate, based on A/B tests, by 10%. And yet, in total, your conversion rate is down 10% vs. last year. Your CEO holds you accountable for increasing conversion rate. How are you going to demonstrate that the conversion rate decrease is not your fault? What offline data do you need to make your case to management?

Strategic Question #4: Management is considering closing down your catalog division. Management wants to know what e-commerce sales will look like in 2015 if there has not been a catalog to support e-commerce sales for a five year period of time. What is your estimate for e-commerce sales in 2015, without a catalog division there to support e-commerce sales?

Here's the trap that the Web Analytics expert is in. All of the big Web Analytics providers (Coremetrics, Unica, Omniture, for example) can help you answer these questions. You can, in theory, import data or link to data or export data or build a data mart and, technically, get to an acceptable answer.

But if your focus is on the powerful combination of conversion rates and optimization, it will be hard to conceive an analytical framework that yields an answer acceptable to a CEO.

As mentioned a few weeks ago, we're going to spend a lot of time in August exploring how our focus on conversion/optimization limits our ability to answer strategic questions. In August, we'll show how we can use a simulation environment to better understand strategic issues in online marketing. We'll explore how we can see the future via a different framework.

If you want to prepare for our month-long discussion, consider these authors, folks who use different metrics to re-define their craft, or explain how the "new" is really borrowed from the "old".

• Basketball: The Wages Of Wins, explaining why scoring does not lead to wins.
• Football: Advanced NFL Stats, illustrating the ways that traditional metrics fail to explain success.
• College Football: Smart Football explains why the "spread" offense is not orginal, is largely borrowed from plays from fifty or more years ago.
• Here's an article about Patrick Ewing that has many similarities to Web Analytics, arguing that you optimize the end result by sub-optimizing components, by having less talented players shoot more so that the best player's performance is optimized. Of course, this is contrary to Web Analytics and Online Marketing theory, but it does help explain why a decade of optimization sometimes leads us to lower conversion rates over time.