Customer Segmentation

Pricing is complicated. Today we make it simple: with segmentation. And, lemonade stands.


lemonade stand.jpg

You have experienced segmentation before; in fact, if you are a pricing professional you are probably acutely aware of it. Especially if you drive a Porsche (because don't all pricing professionals drive Porsches? I digress.)

What is segmentation? How can we simplify something so complicated? Read on, and learn, young entrepreneur.

 

Get the Ultimate Pricing Book

 

Segmentation made simple

Segmentation means grouping your customers together based on characteristics ("features" in model-speak). For example, you have some millennial customers, and some baby boomers. This very simplistic, one dimensional grouping, can be useful: boomers are rich, but don't like trying new things; millennials are not rich, but are constantly trying new things (as long as those new things are free). 

You could also group by geography, say people who live in that most famous of zipcodes, 90210, and those who live in (adjacent) 90211. Or you could do both: after all, those 90210 millennials have some spending money. And they like to try new things.  

There–we're done with segmentation. Easy, right?

What do you do with segmentation?

Perhaps we're not quite done. Ideally, that similarity will also reveal different willingness to pay, and be actionable.

For example, if we were selling lemonade, at a lemonade stand, and we saw that people from 90210 were willing to pay more for it than people in 90211, we could do little until we opened a second lemonade stand in 90211. (You may ask how we'd know which zipcode our customers come from: there are many, many ways to figure that out, but the easiest would simply be to ask customers where they're coming from). 

Once we open the second stand, however, we can charge more for lemonade at our stand in 90210, but less at the stand one zip code over, capturing the most profit possible. 

What about the age dimension? Let's say we observed that boomers prefer buying lemonade at lunch break, but millennials buy after school gets out. We could increase the price at lunch to capture higher willingness to pay at both stands, and then adjust it back down for the millennials afterwards. And yes, I realize that millennials are older than students today but the name for the next generation remains unsettled (Generation Z? Screeners?).

Did you notice how many features we're dealing with now? 

  • Geography
  • Time of day
  • Day of week 
  • Age

And did you notice that all of these hinge on the behavior of your customer (when they are in the market, what they are willing to pay, whether they bought at a given price). 

If this same example were for an online seller of lemonade, we would have fewer dimensions to work with: it would be hard (and possibly illegal) to charge different prices for 90210 and 90211 residents as they visited your website. You could–and leading ecommerce companies like Staples and Walmart do this today–but let's just say Shopify doesn't support that feature, just yet.

 

Watch the webinar recording for an in depth explanation >>

 

For SaaS businesses–like Shopify–there are new, added dimensions, which can capture this. For example, versioning: you make a slightly different product for a segment of customers who are willing to pay more (or less). Like, for enterprise ecommerce companies, I might enable a custom per-zipcode dynamic pricing model, and charge them more for that version than for the basic version. Or, perhaps I charge less for the mobile version and more to buy via phone, knowing millennials buy things on their phones and boomers buy via phone. Yeah, I don't get it either, but the data doesn't lie. Did you know that people still pay for AOL? In 2016? 

I thought this was complicated

This was a lemonade stand example, and even that had 4 features after one paragraph. If you consider more complex businesses–say, a rental car company with many different cars, in different locations, rented for different lengths–it does in fact get quite complex. However, there are two approaches we find that can extract useful segments from nearly any dataset. 

The human-led approach

People have a good sense for their customers. Like the barista who noticed that people in suits buy more extras, more often, than customers in Birkenstocks do. If you give people the right tools–maybe even just Excel and pivot tables–they can usually follow their intuition and find a few segments. And our lemonade example hit the major features to look for–geography, time of day, day of week, and age. Gender is also incredibly important (usually, the most important). 

Gender segmentation is why the men's department always gets short shrift in department stores: men aren't the valuable customers. 

The machine learning approach

For complex businesses, the variables involved can be overwhelming and valuable patterns can be missed. For industries like rental car, hotel, airline, ride sharing, and marketplace, where the ability to set differentiated pricing is quite possibly limitless (think: the opposite of an online lemonade store) extracting all of those segments can be worth millions of dollars. 

One approach we recommend is to train your model with each purchase, properly tagged, to understand willingness to pay across each bucket without prejudice. The resulting model will reveal patterns, especially around certain product/time of day or day of week combinations, that will surprise you. We have seen human errors in mispricing, when fixed, result in dramatic increases in utilization and revenue (over 50% in both cases).

Alternatively, using cluster models separate from your demand model enables you to have more stable, consistent segmentation and to learn about each segment, while allowing for highly dynamic demand. 

The further advantage to using machine learning is, unlike a hand-built model, you can turn it more easily into a scalable, adapting production service. Human-driven models usually make their way into production as a rules engine, which can provide significant gains early on but becomes very, very difficult to manage (Think: increase prices for hotels booked on Tuesday mornings before noon, but not in Las Vegas, or New York, and not if it's a Deluxe room, and not...).

Conclusion

Segmentation is easy. Seriously. However, getting it right takes skill, thought, and real work. And getting it right for a large, complex enterprise is a serious undertaking that may require sophisticated tools and serious big data firepower. 

But the rewards can be significant–life changing, even–so the investment is well worth it, and maybe one day you can be Jack Bonneau–and have a chain of lemonade stands by age 10. 

Related Posts