Not a few. Not an inconveniently small number. Zero. The Air Force had designed the cockpit for everyone in general and almost no one in particular.
You may be doing the same thing with your customers.
The average customer is 43 years old and 57% likely to be female. She purchases 2.7 times per year. She spends $84 per transaction. She has been a customer for 3.2 years, opens 26% of emails, contacts customer service 0.8 times annually, and reports an overall satisfaction score of 7.4.
This customer is quite active for someone who does not exist.
Individually, each average can be useful. Combined, they create a statistical composite that looks reassuringly specific but rarely resembles an actual person.
A customer can be average in annual spending but far above average in purchase frequency. Another can have average tenure but unusually high service costs. Someone else can match the mean age while differing in product needs, channel preferences, price sensitivity, household structure, and likelihood to leave.
Add enough conditions and eventually nobody qualifies as average on all of them. So while the average is mathematically correct, the business interpretation may still be wrong. And before the end of this column, I’ll give you the steps for an Average Customer Audit of your own.
The Average is a Summary, Not a Customer
I have sat through more than a few presentations in which someone unveiled “the customer” as if the organization had discovered a new species.
The customer values quality, convenience, innovation, affordability, reliability, and personalized service.
An inspiring discovery. Apparently, customers prefer good things.
The problem is not that averages are useless. They are indispensable for financial reporting, forecasting, staffing, and executive dashboards. No CFO wants to begin an earnings call by reading every invoice aloud.
The problem begins when you use a group summary as a design specification.
Suppose a hotel’s average guest stays two nights. That figure helps forecast occupancy. It does not mean the hotel should design its experience around a fictional two-night traveler. The underlying customer base may include one-night business travelers, three-night conference attendees, and families staying a week.
The mean compresses all of that into one tidy figure, but tidy is not the same as true.
Once you mistake the summary for the customer, the errors spread. First into product design, then into marketing, pricing, and service. By the time the average reaches the front line, it has become policy.
How the Average Customer Takes Over the Business
It starts with the product. A product team combines research across the customer base and lands on moderate performance, a moderate price, and a moderate number of features.
The result creates little objection and even less enthusiasm.
Performance-focused customers find it underpowered. Price-sensitive customers find it expensive. Simplicity seekers find it cluttered. Advanced users find it limited. Everyone gets part of what they want, which sounds democratic until you notice that nobody gets enough of it to care.
Market researcher Howard Moskowitz reached a similar conclusion in a more appetizing setting. In the 1970s, Pepsi asked him to determine the ideal level of sweetness for Diet Pepsi. The data refused to produce one perfect answer; different groups preferred different levels of sweetness.
A decade later, in a project made famous through a 2004 Malcolm Gladwell TED Talk, Campbell Soup brought Moskowitz in to help its Prego spaghetti sauce compete with Ragu. He tested 45 variations and found not one universal favorite, but distinct clusters of preference. Some customers wanted a traditional smooth sauce, some wanted it spicy, and a large group wanted extra chunky…a category that barely existed on supermarket shelves.
Prego did not need a better average sauce. It needed several sauces for several kinds of customer. The search for the perfect recipe had been the wrong assignment.
You can remove every sharp edge and discover, somewhat late in the process, that the sharp edges were where the value lived.
The Air Force did not solve its cockpit problem by calculating a more precise average pilot. It built adjustable seats, pedals, and controls. When customers differ in meaningful ways, flexibility usually beats compromise.
Then marketing tries to describe everyone at once.
Once the product has been built for the average, marketing receives the unenviable task of making it sound distinctive. Average-based marketing tends to produce messages for everyone, and messages for everyone have a bad habit of reaching no one.
One customer values convenience enough to pay more; another accepts inconvenience to save money. One wants expert guidance; another wants the company to stop sending emails and quietly process the order.
Those customers do not need different adjectives. They need different value propositions.
Over the years, briefs have crossed my desk with enough customer attributes to describe half the population and enough personality to describe nobody. The resulting campaign is broad, polished, thoroughly approved… and forgettable.
Useful segmentation starts with a decision. Would this group receive a different offer, product, message, or service model? If not, you do not have a segment. You have a label, perhaps accompanied by a stock photograph of “Busy Brenda” holding an artisanal coffee.
Pricing Exposes the Fiction
Eventually, the abstraction collides with money.
Imagine that half your customers would pay $20 and half would pay $100. The average willingness to pay is $60.
Price the product at $60 and you lose the first group while leaving a great deal of money on the table with the second. The average was accurate. The decision was poor.
The same problem appears in cost to serve. One group may buy frequently and require almost no support. Another may generate healthy revenue while consuming enough customer-service time to erase the margin. Averaging them together makes both groups look economically similar when they are anything but.
The response should change the offer, not merely describe the difference. Give performance-oriented customers a premium version. Remove cost and complexity for customers who want the basics. Charge separately for a service used by only a small group. Stop funding expensive benefits for customers who do not value them.
Often the right decision is to stop pretending there is only one market.
Service Makes the Consequences Visible
Suppose the average customer contacts support once per year.
That sounds manageable. It could also be almost meaningless.
Perhaps 80% never contact support, 15% call twice, and 5% call twelve times. Staffing around the average misses the actual workload, but it also misses what is happening to customers.
I will neither confirm nor deny that I once worked for a large telecom company that routed customer service calls based on customer value and service history. Valuable customers who rarely called for help went straight to a rep. The next tier down got an IVR, but easy access to switch to a rep. At the very bottom were negative-value customers condemned to IVR hell, in the hope they would defect to our competitors and be someone else’s headache.
More than once, a business has proudly shown me an average service metric while two different realities sat underneath it: a large group receiving efficient support and a smaller group trapped in repeated failure. The average did not reveal the problem. It sanded it smooth.
The product is built for the composite, the message is written for the composite, the price is set for the composite, and the service system is staffed for the composite.
Then an actual customer arrives.
Design for Variation, Not the Composite
Once you accept that the average customer is a statistical construction, your goal is not to find a more precise average. You should design the business so that meaningful differences among customers can survive.
That does not mean creating a unique product, price, message, and service process for every individual. That way lies madness, or at least a CRM implementation that never quite ends.
It means resisting the urge to force every customer into one standard experience. Where needs differ, offer a small number of useful choices. Where behaviors differ, avoid assuming that the most common behavior is universal. Where customer economics differ, stop treating every account as equally valuable or equally costly to serve.
The Air Force’s answer was not a better version of a standardized cockpit. It was adjustability.
Your answer may be product tiers, different service models, varied onboarding paths, or distinct messages for customers with different priorities. The form will depend on the business. The principle will not.
Design around the variation that matters, not the composite customer who appears in the averages (see the Average Customer Audit instructions at the end of this column).
Segment Only When the Segment Changes a Decision
Not every difference deserves action.
Age, income, geography, and household size are easy to describe, but easy does not mean useful. A good segment changes what you do.
Useful customer differences often involve needs, behaviors, product usage, profitability, price sensitivity, lifecycle stage, or likelihood to leave.
Netflix provides a good example. Its recommendation system does not show everyone the programs that are most popular overall. It tries to predict what each viewer is likely to watch.
A recent study using Netflix viewing data estimated that replacing its current recommendation system with a popularity-based approach would reduce engagement by 12%. It would also narrow the variety of programs people watched. Most of the benefit came from targeting the right titles to the right viewers, not merely placing a title in front of more people.
Popular is not the same as relevant.
You do not need Netflix’s computing infrastructure to apply the lesson. A retailer can recommend different products based on past purchases. A bank can vary financial guidance by lifecycle stage. A software company can alter onboarding based on which features a customer intends to use.
The technology can be simple. The important part is recognizing that the same answer will not serve everyone.
No changed decision means no meaningful segment.
Build Adjustability into the Experience
Prediction is useful. Adjustability is often better.
Instead of trying to infer exactly what every customer wants, let customers choose among a manageable set of options: basic or premium service, digital or human support, monthly or annual billing, frequent or occasional communication.
The Air Force did not need an algorithm to predict each pilot’s leg length. It needed a movable seat.
Businesses tend to overcomplicate personalization because sophisticated systems sound strategic. Sometimes the best answer is a setting, a tier, or a choice presented at the right moment.
Test Whether the Differences Are Real
Segmentation can become its own fiction. The organization replaces one imaginary average customer with six imaginary personas, each given a first name, a stock photograph, and a biography detailed enough to qualify for a modest streaming series.
Personas are hypotheses; behavior is evidence.
Test different messages, offers, onboarding paths, and service levels. Look at whether groups respond differently. An initiative with a modest average lift can be transformative for one segment, irrelevant to another, and damaging to a third.
The aggregate result is not false. It just doesn’t tell you what happened to everyone underneath it.
The Average Customer Audit
You do not need a new segmentation model or a six-month customer transformation program to test whether your average customer exists.
You can do what Lieutenant Daniels did.
This week, ask your data team to run what I’ll call the Average Customer Audit.
- Choose the customer measures that matter most. Select perhaps five to ten metrics your company routinely uses to describe customers: age, annual spending, purchase frequency, tenure, product usage, satisfaction, service contacts, margin, or whatever else appears in your dashboards and customer profiles.
- Define what counts as average on each measure. For every metric, establish a reasonable range around the average. The exact definition is less important than making it explicit. You might use the middle 30%, the middle 50%, or a range based on business judgment.
- Count the customers who are average on every measure. This is your Daniels number. You may find that the customer represented by your collection of averages is rare or does not exist at all.
- Relax the test one measure at a time. Count how many customers are average on all but one measure, all but two, and so on. This shows how quickly real customers begin to appear once you stop demanding that they match the composite on every dimension.
- Show the result to the people designing the experience. Put the findings in front of product, marketing, pricing, and customer-service leaders. Then ask a simple question: How much of our business has been designed around the customer in the first row of this table?
The point is not to prove that averages are bad. It is to reveal what they become when assembled into a person.
Run this analysis before commissioning another persona deck, redesigning the customer journey, or declaring that your organization now understands “the customer.”
You may discover that the average customer is not your largest segment.
You may discover that it is not a segment at all.
No One Sits in the Average Seat
Customers do not experience your company in aggregate. They encounter a specific product, price, message, recommendation, policy, or service interaction.
Averages help you understand the population. Trouble begins when you ask a population-level summary to stand in for the person making the purchase, opening the email, calling support, or walking away.
Do not throw away the averages. Put them in their proper place.
They belong on the dashboard, not in the seat.
Click here for more columns from Michael Bagalman’s Data Science for Decision Makers series.
