In our last post, we presented an introduction to Customer Lifetime Value (CLV).  Here, we'll take a deeper dive into how CLV is actually modeled and calculated.  Understanding the concepts behind CLV can help you use it properly.

## What's Needed?

To figure out CLV, at a high level, you need to know:

• Which customers are "active" and will purchase again in the future
• How often each customer purchases from you
• How much each customer will spend when they purchase
• How long each customer will remain active for, before they cancel (or stop purchasing)

## Contractual vs. Non-Contractual Relationships

If you run a gym, counting your active customers is simple: because customers sign a contract with their gym membership, you can just count the number of active contracts.  And you know when a customer cancels, because they either cancel their active contract, or fail to renew it.

However, a restaurant can't count their active customers quite as easily.  Customers don't sign a contract to buy cheeseburgers or pizza, they just order when they want.  If a customer decides to "cancel" their relationship with a restaurant, they simply don't return.

In other words, a gym has contractual relationships with their customers, because they sign a contract, whereas a restaurant has non-contractual relationships with their customers.

## Counting Active Customers

So, without a contract to indicate that a customer is active, how can we figure out which customers are active, and which aren't?

The simplest approach uses time duration; for example, we could say that any customer that hasn't ordered for 30 days is considered inactive.  Any customer that's ordered within 30 days is active.  But we can do better.

### Recency

Obviously, a customer that has just ordered is active.  A customer that hasn't ordered in 10 years is probably not active, so we know intuitively that customers "decay" over time.  The longer it's been since a customer has ordered, the less likely they are to be active.

This is key: our improved model does not just say that a customer is active or inactive.  We assign a probability to the likelihood that a given customer is active.  For example, we could say that Joe Smith is 75% likely to order again, since he just ordered two weeks ago.

Our model should incorporate the recency with which a customer has ordered.

### Frequency

Next, we should consider the rate of decay for a given customer.  If all customers behaved the same, then we could simply build out probabilities based on time, and be done.  But, not all customers behave the same.

If a regular customer orders once a week, like clockwork, then doesn't order for 2-3 weeks, alarm bells should be ringing!  This customer is slipping away rapidly.  But another customer might only order once every few months, so for them, a 2-3 week absence is no big deal: that's part of their normal pattern.

So, our model should incorporate the frequency with which customers place orders.

## Putting it Together

Now our model that tells us the probability that a given customer is active, from 0-100%.  We can use this to figure out how long our customer is likely to remain active, which could range from days to years, depending on the individual customer's activity and your overall customer retention rates.

Combining all of this information, we could model the following information about a few hypothetical customers:

• Joe just ordered yesterday, so he's likely to be active.  He usually orders twice a month, and spends about \$22 per order.  Projecting 36 months out, we assign Joe a 97% probability of being active, and spend about \$1,600 in total.
• Bill hasn't ordered for a few weeks.  He usually orders twice a week, so he's not likely to be active.  He normally spends about \$17 per order.  Projecting 36 months out, we assign Bill a 34% probability of being active, spending only about \$30 in total.
• Mary usually orders once every few months.  She hasn't ordered for a few weeks, but that's okay, given her prior ordering patterns, so she's still probably active.  She normally spends about \$72 per order.  Projecting 36 months out, we assign Mary an 81% probability of being active, spending about \$1,100 in total.

These are just a few examples of how customer lifetime value modeling can provide insight into your restaurant's customer health.  In future posts, we'll cover specific ways to use CLV information to drive business results.