Even Loyal Customers Quickly Equalize

Look down the column where "Freq = 15", a column that represents customers who purchased 11-15 times in their history.

At recency = 14 the customer has a 44% chance of buying again. This customer, remember, has 11-15 life-to-date orders.

Now look at recency = 01 / Freq = 02:

• Annual Rebuy Rate = 44%.

These two customers are equal.

• Frequency = 11-15, Recency = 15 months.
• Frequency = 2, Recency = 1 month.

Even your loyal buyers quickly equalize themselves, constantly trending in a negative direction unless they purchase again.

This is one of the main reasons why customer acquisition is so darn important. We think we are dealing with a loyal customer base. We simply aren't. We deal with customers who achieve loyal status and then begin to immediately decay. We constantly need new customers, don't we?

Make Sure You Control For Key Attributes

When you control for recency, frequency, monetary value, channel, merchandise preference, and price ... you'll learn that certain attributes that appear important at an "indicator level" are not important at all, or are important for a very short period of time ... or are still very important. Anything is possible. But you have to control for key factors first.

Let me tell you a story about a client ... this client wanted me to prove that their loyalty program was "working". It wasn't working ... loyal customers in the loyalty program were spending 10% less per year and loyal customers not in the loyalty program were spending 10% less per year and just about every customer segment was off 10% year-over-year. The business just wasn't resonating with customers, and the loyalty program wasn't helping.

So that's what I presented.

Guess what?

The loyalty vendor didn't like what I shared. Go figure!!

So the Loyalty vendor puts together a slide that has a horribly biased bar chart ... the slide shows that customers with a Loyalty Indicator = Yes were 160% (or whatever the percentage was, I don't recall the exact number) more valuable than everybody else and therefore the Loyalty program was generating enormous value. The slide used words like "PROOF" and "VALUE" in all caps, so of course you had no choice but to believe the vendor ... a vendor with a vested interest in protecting the program defended by the "indicator variable".

I ran about 50 separate queries. 

I ran a query segmenting Mens customers from all other customers. Guess what? Mens buyers were more valuable.

I ran a query segmenting Online visitors from all other customers. Guess what? Online visitors were more valuable.

I ran a query segmenting discount/promo buyers from everybody else. Guess what? These buyers were more valuable.

I ran a query segmenting liquidations buyers from everybody else. Guess what? These buyers were more valuable.

In fact ... in all cases (for that client) the chosen attribute suggested customers were more valuable if they aligned with the chosen attribute.

Why did this happen? Because of the highly biased nature of the queries that some vendors and some marketers construct to prove their point. 

• The attribute isn't what matters.
• The fact that good customers do many things means that you can select nearly any attribute and many / most good customers will possess that attribute and therefore the attribute will show that customers are better.

Don't believe me?

Run the queries for yourself on your own data.

Seriously. Go. Now!!

Meanwhile, there are times when indicator variables are truly important. In a recent analysis I saw that an email subscriber variable (1 = yes, 0 = no) showed a 45% lift in a segmentation analysis ... but get this ... it showed a 52% lift in a logistic regression after controlling for recency and frequency.

In other words, I'm asking you to do some work ... be a bit more sophisticated than normal. Your indicator variable might well be meaningful and important. Prove it by controlling for other factors, ok?

Equalization

Here are annual repurchase rates by recency/frequency combination for a "brand".

Yellow colors are important ... yellow cells represent cases where a customer has a minimum annual repurchase rate of about 35% (which isn't anything to write home about). There are different customers who are essentially "equal" in terms of repurchase rate.

• Recency = 18 months, Frequency = 6 orders.
• Recency = 16 months, Frequency = 5 orders.
• Recency = 12 months, Frequency = 4 orders.
• Recency =   9 months, Frequency = 3 orders.
• Recency =   4 months, Frequency = 2 orders.

All of those customers are essentially "equal" in terms of repurchase rate. Each time you get a customer to repurchase, you buy yourself more time to convert the customer next time.

Notice that not a single first-time buyer in the example above has a 35% or better annual repurchase rate. First-time buyers are unlikely to repurchase again ... this is why Welcome Programs are so darn important. You spent all that money acquiring the customer, why not focus on getting a second purchase out of that customer before the customer goes dormant?

"AV" ... Annual Value

Here's something I run in various projects ... something I call "Annual Value".

You'll have to click on the grid to see it more clearly, as there is a LOT of information in the image.

Structure of the Query:

1. Freeze the customer file as of, say, June 30, 2019.
2. Calculate months since last purchase (recency) and life-to-date orders (frequency) for each customer as of that date.
3. Average how much each recency/frequency combination spent in the next year (July 1, 2019 to June 30, 2020).
4. Depict data in a grid-style format, as illustrated above.

We create an "AV-Table" or "Annual Value Table" so we can understand the general value of customers at different life stages. 

Go ahead and read down the first column ... Frequency = 01. I'll summarize for you here, at least for the first fifteen months.

• Recency 01 = $26.81.
• Recency 02 = $21.28.
• Recency 03 = $13.32.
• Recency 04 = $14.09.
• Recency 05 = $14.42.
• Recency 06 = $14.39.
• Recency 07 = $17.28.
• Recency 08 = $15.73.
• Recency 09 = $10.99.
• Recency 10 = $10.99 (again).
• Recency 11 =   $9.47.
• Recency 12 =   $8.70.
• Recency 13 =   $7.92.
• Recency 14 =   $8.19.
• Recency 15 =   $6.07.

The first-time buyer just drops off of a cliff, doesn't s/he? You lose more than five dollars of annual value if you don't get that customer to purchase again in the first month the customer is on the file. You lose another eight dollars of annual value if the customer drops from recency = 02 to recency = 03. It's almost like a welcome program would really work in this instance.

Ideally you aren't looking at annual demand/sales ... ideally, you subtract out cost of goods sold and net delivery expense and marketing expense. But for a first example, we'll start here, ok?

Ways Of Saying "No"

It was my first consulting project ... P1 ... I completed a multi-hour presentation showing the Executive Team how different brands were performing (they were performing poorly). I explained that there was a significant merchandising issue, I explained that the marketing team approached their craft like it was 1979, I explained how important customer acquisition was going to be going forward.

It wouldn't be the first time I'd give a presentation like this.

It wouldn't be the first time that almost everybody in the room would reject the message.

During a break I tried to find the EVP of Marketing. During my presentation I noticed that he was not paying attention ... at all. I found him, outside, in the designated smoking area. I didn't even have a chance to get the small talk going ... he took a puff of his cigarette, exhaled, and offered me his thoughts.

• "I know why you're here. I'll make this easy for you. I'm not changing. My team is going to do what we've always done. And as for the rest of those people in that building (pointing to the building), they're not changing. Those brands are dying and the merchants don't care and the creative team is going to execute exactly the same way they've always executed and that's just that. You people come, and you people go. The brands and the core employees persist. I'm not hanging my employees out to dry to partake in the flavor of the day."

Later, the CEO would tell me that "these people won't do anything I tell them to do."

Good luck everybody!!!

Yesterday I talked about the pushback I get when I share my "Comp Segment" framework with people (click here). Some of you regaled me with stories of Executives who found interesting ways to say "no" regarding the methodology.

Thirteen years of going this work. Thirteen years of hearing "no".

There are perfectly valid reasons for hearing the word "no". My work might be inappropriate for a brand, not what the company needs at a point-in-time. My methodologies might not truly address what the "brand" is trying to address.

Then there are all of the other types of "no".

A popular version of "no" is "Deflection". 

• "We're not going to do something because gosh isn't Amazon doing all sorts of interesting stuff? And what about Walmart? Who will win that battle? And I wouldn't rule out Target either, they're making inroads. I think it's a battle between those three. We're just the collateral damage associated with their war."

This is the "Woodside Research" crowd ... they will find many ways to tell you no while paying Woodside Research five figures to figure out how Amazon / Walmart / Target say "yes".

Another version of "no" is "Tribal Membership".

• "Don't you get it? The world is headed in an omnichannel direction and if you don't create great experiences in retail you're dead. You just have to be great, you have to be seamless across channels, and the customer doesn't care about channels because to the customer it's just one brand."
• "We aren't executing mail/holdout tests in catalog marketing because we'll lose sales. Besides, our favorite vendor told us that Amazon is going to do another catalog so we know catalogs are important to our future."

No amount of data will get this audience to convert. Take your comp segment analysis and throw it away. They have a hammer, and every problem to them looks like a nail.

A third version of "no" is "Passion".

• "It can't be the merchandise, your analysis has to be wrong. We surveyed our customers and they love what we're selling. Marketing must be doing something wrong."
• "Our marketing strategy is spot-on. Our paid social efforts create significant awareness opportunities. It's a shame the merchandise is so awful."
• "I'm certified in data science and I can tell you that a comp segment analysis is an awful way to look at information. You are a simpleton."

These folks have a passion for their craft (merchandising, marketing, analytics) and they become unlikely to listen or think critically as a consequence. If you are certified in data science, then why the heck can't your methods detect a 20% drop in sales of new merchandise year-over-year? And if you're a great merchant, why did you author a strategy that resulted in a 20% drop in sales of new merchandise year-over-year? And if you are a great marketer, why aren't you trying to help your merchandising team sell the new merchandise by featuring it prominently in your campaigns?

Each version of "No" can be addressed.

• Deflection:  I go right at the individual. What are "you" going to do to keep your brand alive? You're paid to keep your brand afloat, so I want to see what your specific marketing plan or merchandising plan is to grow. These individuals frequently author weak marketing plans and weak merchandising plans, and when the weak plans fail, they have somebody to blame (how can we compete against Amazon, it's their fault?!)
• Tribal Membership:  This is the hardest one to argue against. Tribal Membership requires faith and belief. To move this person, you need to build a bridge from where the person is to where the person needs to be while not destroying the world where the person is. It's hard work. It will take time. But you can improve their business results!
• Passion:  The easiest one to attack. And attack is the wrong word. You aren't attacking, you are trying to get somebody with a vested interest in maintaining current norms to change. There are plenty of merchandising success stories, go find them! There are tons of marketing successes, cheer 'em on!! Praise your analytics guru and encourage the guru to solve the problem you are trying to solve using the tools that the analytics guru cares about.

Have a plan. There are many ways that people will tell you "no". The response to "no" is more important than the change you are trying to enact.

Criticism of Comp Segment

As mentioned a few days ago, there are criticisms of my Comp Segment framework. I've heard 'em since I began sharing the methodology widely ... back in 2012/2013.

Let's hear some of the criticisms/questions:

You only control for purchase frequency. Shouldn't you also control for dollar amount? Why control for 2 orders when some place two orders for $50 each and others place two orders for $125 each?

• Yes, controlling for dollars is even better.
• The problem you face is that smaller companies have fewer customers, so as you add dollar constraints you reduce the size of the audience and therefore increase noise, making inferences more challenging.

Why two purchases in the past year? Why not three? Why not eleven?

• There is no right/wrong answer to this question.
• You want to avoid infrequent shoppers as they're not representative, you want to avoid first-time buyers where possible (as they're influenced by marketing channel used during acquisition), and you want to avoid loyal buyers who are highly influenced by marketing.

Shouldn't large retail brands like Macy's use different criteria than two purchases in the past year?

• Yes, and Macy's has enough customers to eliminate noise.
• Control for both purchase frequency and dollars spent, and if the brand has a loyal customer base, you can move up to 3/4/5ish purchases in the past year.

Won't I get different results if I look at different segments of customers?

• Quite possibly, yes.
• But the "direction" of the results, unless you have a very noisy customer base, will be comparable. And that's what is important.

How do you "know" that the issues identified by the analysis are merchandising issues?

• You analyze comp segment performance for new items, and for existing items. You analyze by merchandise division. If you see differences, you have merchandising issues. If every analysis looks identical, then positives/negatives are marketing driven.

Shouldn't you use machine learning or AI to measure the impact of merchandise productivity, controlling for a myriad of issues?

• Absolutely! Have at it.
• However, if you do 2% of the work (that's how easy this is) you'll get 95% of the answer. For just 2% of the work. Who wouldn't do 2% of the work to get 95% of the answer?

Shouldn't I average results across many different segment?

• Go ahead, do that. Just be sure to not let marketing issues cloud your results.
• Of course, one wonders why you haven't already been doing that for the past decade?

There's nothing you can say to convince me that I should be doing this work. Your work is just plain wrong.

• Then don't do it. Do your work your way. That's fine!

How Do I Compute A Comp Segment Metric?

This topic keeps coming up. In the next post I'll address criticisms of the methodology.

Here's how we calculate Comp Segment performance.

Identify all customers who bought exactly two (2) times between 6/1/2019 and 5/31/2020. 

1. Within this audience and only this audience, calculate the average amount spent between 6/1/2020 and 6/30/2020. If you had ten customers and their spend was 0 / 0 / 100 / 0 / 0 / 0 / 0 / 0 / 0 / 0, then the average would be $10.00.
2. Identify all customers who bought exactly two (2) times between 6/1/2018 and 5/31/2019.
3. Within this audience and only this audience, calculate the average amount spent between 6/1/2019 and 6/30/2019. If you had ten customers and their spend was 0 / 0 / 110 / 0 / 0 / 0 / 0 / 0 / 0 / 0, then the average would be $11.00.
4. Calculation:  (Step 2 / Step 4) - 1. In our example, that would be (10 / 11) - 1 = -9%.
5. Repeat Steps 1/2/3/4/5 for at least 2-3 years.

Across a career spanning more than thirty years, this metric aligns most closely with "why" a business is succeeding/failing than anything else I've run across.

I'll run this metric in total, for new merchandise, for existing merchandise, for items selling above their historical average price, and for items selling below their historical average price. The metric allows me to identify "when" a business began to struggle, it tells me if new/existing items are causing the problem, it tells me if discounting is a problem, and it tells me how/when to drill down into the data for more details.

The definition of exactly two (2) purchases in the past year allows me to get as close as I can get to determining the impact of the merchandise being sold on company performance.

I also create a "Comp New/Reactivated Customer" metric.

1. Identify all customers who bought between 6/1/2020 and 6/30/2020.
2. Subtract all customers who bought the year prior to 6/1/2020 and 6/30/2020.
3. Identify all customers who bought between 6/1/2019 and 6/30/2019.
4. Subtract all customers who bought the year prior to 6/1/2019 and 6/30/2019.
5. Calculation:  (Step 2 / Step 4) - 1.

This metric allows me to see when new + reactivated customer strategies changed. This is the second-biggest issue I find across my client base.

• Comp Segment allows me to see what the merchandising/product people are doing.
• Comp New/Reactivated allows me to see what the marketing people are doing.

Within a short period of time, I have the metrics needed to understand why a business is not meeting expectations.

Build this stuff into your weekly customer database updates ... easy!!

Master Equation

So I run through all of the Elite Program data ... a Chief Marketing Officer didn't like how the writeup reflected his business. He says to me:

• "Instead of comp segment performance, which I don't like because it is biased, is there a Master Equation that combines all elements of important variables to communicate why the business is succeeding or failing?"

Since introducing Comp Segment as my preferred metric to determine if merchandise performance is healthy or not (way back in 2012-2013, no less), this has been a common comment.

Remember:

• Comp Segment = Take all customers who purchased 2x in the past year. Then measure how much these customers spend in the next calendar month. Then compare this year's spend in the next calendar month to last year's spend in the next calendar month. The percentage difference is your "Comp Segment" performance.

Two Key Findings Over The Years:

1. Comp Segment Performance is most indicative of what is happening today.
2. Comp New + Reactivated buyer performance is most indicative of what will happen in the future.

To answer the CMO's question, I performed a Principle Components Analysis of the variables this individual thought would be important.

• Demand Change ... he wanted this to be "predicted", if you will.
• Comp Segment (CompSeg)
• Comp New + Reactivated buyer counts (NewRct)
• Above = TY/LY spend on items priced above their historical average.
• Below = TY/LY spend on items priced below their historical average.
• NewMerch = TY/LY spend on new items.
• ExiMerch = TY/LY spend on new items.
• OrdBuyer = TY/LY change in annual orders per buyer.
• ItemOrder = TY/LY change in annual items per order.
• PriceItem = TY/LY change in annual price per item purchased.

In a Principle Components analysis, the data point closes to "Demand Change" is the variable that most closely aligns to demand changes.

What does the image below show you?

Comp Segment performance and change in Existing Item sales are most closely aligned with Demand changes, year-over-year.

Now imagine what a Comp Segment analysis of Existing Item sales will tell you?

You don't need a Master Equation of key variables. You need to analyze Comp Segment performance, and parse it between New Merchandise and Existing Merchandise if you want to learn why your business is / is not meeting expectations!

Is Your County a Hot Spot?

This map (click here) does a nice job of outlining new COVID-19 cases per 100,000 residents. You can see that Maine has fewer than 3 new cases per 100,000 people ... Arizona has more than 45 new cases per 100,000 people.

Back in 2008 I had a free product called "Zip Code Forensics", where I illustrated the zip codes in the United States that were catalog-centric. Popular industry pundits criticized the work ... "in a world of retargeting at a one-to-one level displaying response at a zip code level seems almost moronic."

But you don't do this stuff to target more accurately ... you do this stuff to imagine what the information means. What happens when people in Maine move back indoors in October in a way comparable to what people in Florida, Georgia, Texas and Arizona did in late May when it became too hot for southern folks to be outside? Can you plan ahead for a shift in behavior? Why would New Mexico buck the odds and have half to a fourth of the prevalence of Texas/Arizona? Similarly back in 2008 you asked yourself "why are rural zip codes catalog responsive while suburban zip codes online responsive and urban zip codes retail responsive? What does it mean to in-store sales when suburban zip codes are shifting dramatically to online shopping?

There were two clear facts that Zip Code Forensics demonstrated.

1. Retail, especially in suburban areas, was in trouble.
2. Catalog circulation was going to dramatically decrease over time given that only rural areas were highly responsive.

Both facts played out over the course of the next ten years. 

Both facts were easy to see when mapped appropriately.

You map this stuff so that you can imagine what could happen in the future ... the map connects the present with your imagination to yield the future.

You just wonder what facts are hiding in your data, in plain sight, waiting to be discovered?

Parsing Long-Term Value

Anytime you work on a long-term value project, you come to an interesting realization. 

1. Customers generate their own value, largely due to love of the merchandise you offer.
2. Marketers are responsible for generating long-term value, largely due to how they manipulate customer value.

Say you split up long-term value over the first five years the customer is active following acquisition. You learn that the customer will spend $100 of future sales and with a 40% flow-through rate and $20 of ad cost over five years, the customer is worth 100*40 - 20 = $20.00 of future profit. 

Then you take the $100 of future sales and assign it to one of four buckets. 

The first bucket? The amount of future value that is organically generated and presumably happens because of brand loyalty and love of merchandise. You find that 35% of future value is in this category.

• $100 * 0.35 * 0.40 - $0.00 = $14.00 profit.
• Ohhhhhhhh .... that's a lot of profit.
• 70% of future value has nothing to do with classic marketing tactics. Ponder that.

The second bucket? The amount of future value that is generated by unpaid social efforts and email marketing.  You measure this, right? Right???? It's darn important because it is practically free marketing, and therefore, is really, really important. You learn that 20% of future value is in this category.

• $100 * 0.20 * 0.40 - $3.00 = $5.00 profit.
• We're up to $19.00 of profit ... and there was only $20.00 of profit to begin with over five years, so that doesn't bode well for what comes next.

The third bucket? The amount of future value that is generated by digital marketing. You learn that 20% of future value is in this category, at a cost of $8.00.

• $100 * 0.20 * 0.40 - $8.00 = $0.00 profit.
• Wooooooo-boy.
• Google + Facebook + Retargeters say "thank you".

The fourth bucket? The amount of future value that is generated by print marketing. You learn that 25% of future value is in this category, at a cost of $9.00.

• $100 * 0.25 * 0.40 - $9.00 = $1.00 profit.
• Wooooooo-boy again.
• Your printers and paper reps say "thank you".

I should recap / summarize what we just learned, correct?

• $14.00 of long-term value generated outside of marketing.
•   $5.00 of long-term value generated by email and organic social activities.
•   $1.00 of long-term value generated by print marketing activities.
•   $0.00 of long-term value generated by digital marketing activities.

You analyze this stuff, right?

What would you do differently if you knew this was happening?? 

• (hint - some version of this is absolutely happening ... and this shouldn't be taken a criticism of digital and/or print ... it's done to get you to think)

Discuss.

Here's a Fun Answer!

You might remember last week ... which feels like 3 years ago during a pandemic but it was just last week. I posted a quiz question (click here) and promised to publish good attempts at an answer.

Well, we have a good attempt at an answer ... from Spiros Parakevas, Dragomir Nedeltchev, and Zlatka Staykova, who work in Data Science at FullBeauty Brands.

The trio wrote up their results ... you can read their methodology/answer here!!

Their writeup is probably useful to your data science team, so why not spend a few minutes reviewing their work and then applying the concepts to situations where you have missing data, ok?

Quiz Question

You could sell 100 units at a price of $29.99 and a cost of goods sold of $9.99, netting you $2,000 of gross margin. That's what your CFO wants you to do. That's what your CMO (Chief Merchandising Officer) wants you to do.

You'd like to sell 160 units at a price of $19.99 and a cost of goods sold of $9.99, netting you $1,600 of gross margin.

Describe the analysis you'd perform to defend your point of view. 

Strategy or Tactic?

You ask your staff to come up with a "Discount Strategy" for your August email campaigns.

• Is this a strategy, or is this a tactic designed to get a customer to buy something?

Your boss asks you to determine who will receive a Friends and Family postcard.

• Is this a strategy, or is this a tactic designed to get a customer to buy something?

The CEO asks you to determine the smartest way to get customers to buy from a new product line.

• Is this a strategy, or is this a tactic designed to get a customer to buy something?

Answer each of the three questions above. Think carefully about what a "strategy" actually is, and compare a "strategy" to the tactics you use to support a strategy.

Predicting Unknown Values

Here's some data from a business that routinely tests different percentage off ideas to various customer segments.

Your job is "fill in the blanks" ... what would have happened if you offered 50% off to a Good customer? What would have happened if you offered 10% off to a Poor customer?

Fill in each of the empty cells. Adjust the cells you don't like that exist in the table. Share your methodology and results with me  ... write up your solution ... I'll publish good predictions, ok?