One of the truisms of business is that it is easier to keep a customer than to get a new one. Actually it may be it’s cheaper to keep a customer. In a lot of modern marketing, especially for startups there are so many channels available that it is often easier to just keep pumping money into them and get an endless stream of new customers. So long as the costs aren’t too high and you are getting growth…. what’s the problem with that? This article is about sing cohort analysis to help with this, but first some reasons why churning customers is a bad idea.
Well there are a number of problems.
- Wasting money
- Lower marketing ROI
- Poorer customer satisfaction
- Limited customer pool
Shallow Pools of Customers – However Large the Market
I’m not going to spend any time on the first three. The last one is interesting. When you start off and define your niche it’s fairly easy to define who your ideal customer is. They are X, with Y job and have these values and these hobbies. You may do some demographic research and estimate that there are 100,000 of them in your city. When you are on customer zero, 100,000 customers is a pretty huge pool for you to swim in and catch fish.
Move forward six months or a year and if you are even moderately successful you are going to have reached a lot of those potential customers. There will be awareness even if they haven’t tried your product. If your churn is high – 50, 60, 70% or even more then it actually becomes fairly easy to start exhausting that pool of customers. What do you do then? You have to start moving into new niches and extending your product line. That carries a bunch of it’s own risks. Will people like your product or service as much? Will it be as good a market? How many more extensions can you make before you have the same problem again and again?
You see if your retention rates are poor you are a bis like a slash and burn farmer. You are going through a beautiful verdant jungle and chopping it all down for short term profit. What you leave behind is a wasteland of customers who will never return
Cohort Analysis & Retention Rates
What cohort analysis does is to look at which customers come back and how often. The first time I used it I was working on a mobile app. I’d just come onto the job and had had some amazing passionate stuff told to me by the entrepreneur. I like data and so I jumped onto App Annie to get a sense of how good the app was at retaining customers. Despite multiple product improvements 95% of customers never opened the app twice. Close to 100% never opened it after the first day. Oooh! That made my role kind of pointless as there was little point pouring good customers into a bad sewage filled bucket with more holes than plastic in it.
Restaurants and Cohort Analysis
I was recently working with a restaurant and crazily they actually had some useful data that we could do cohort analysis on to look at their retention rates. Why do I say crazily? Well I asked a few restaurant owners and managers
Even the largest chains struggle with this for a number of reasons, starting with the lack of sophistication of the talent they employ.
First, they don’t know what and how and why part of the question.
I know of no restaurant owners in New Orleans that use anything close to cohort analysis to assist in converting new customers to regulars. I can tell you with certainty we do not do it.
Could the industry improve through such analytical methods? I think not.
most don’t have a way to gather the data and analyze it.
Gulp. Stuff that is standard practice in the tech industry because it helps to deliver consistent scalable results is seen as too difficult and lacking in value. This is in an industry that has an exceptionally high rate of business failure and notoriously low wages and margins.
In this case we used data from a delivery company similar to Deliveroo or FoodPanda that allowed us to see when and how frequently customers ordered. The original data came as a simple excel list. Person X ordered Y at Z date/time. For a more traditional restaurant wanting to analyse walk in customers one approach would be to access the CCTV feed and use facial recognition software to give each customer a unique anonymised identity. So everyone who comes through the door (and says for long than 20 minutes) is given a unique ID. There is no need to know anything else about them – sex, age, name etc – and this makes it easier in many countries to comply with data protection laws.
The first step was to turn this into cohorts. A cohort is a group of people who you can treat similarly for the purposes of analysis. In this case we decided that given the nature of the restaurant and it’s menu we’d treat everyone coming into the business in the same period as a chort. So we split them into months. First of all you convert your order/time data into a date reference. So if someone ordered at 15/06/2013 at 17:30 then we tag them as ordering in June 2013. Then using excel we work out when was the first month that they came in and ordered – their start cohort.
In this case if this was the first time they had ordered we would lump them with all the other people who ordered for the first time in June 2013. From then on the individal disappears into the data. All you are interested in are the June 2013 cohort and how they differ from the July, August 2013 cohorts.
How do you see how they differ? Well the easiest way is to see how many times they come in and visit. You go back to excel and count how many times they visited in June, July, August etc. The you do the same for all the other cohorts.
Simple Tip: This takes ages if you do it by hand. So it’s normally simplest to use pivot table. This have Cohort as the column, Age as the row and count of visits/orders as the value.
Then you can start doing a bit of comparison.
Example Cohort Analysis
This cohort analysis is some random data that I pulled off the internet (so I don’t accidentally breach any NDAs)
If we assume that the column is the cohort, the rows are the number of months since first order and the colours are the retention rates with green being good and red being bad we can see some pretty interesting things with this data.
It is a triangle shape because the first customers have been around for 20 months, the next cohort for 19 months, the next for 18 etc etc
For the first 2 or three months people like the service. Each month you lose a few more people and over time this adds up until no one from that cohort returns any longer. If we look at the top we can see that big red blotch. That’s some sort of service failure, if we are talking about a restaurant. Why did people suddenly stop coming after the fifth month? Was there a change in a customer loyalty programme?
It certainly seems related to how long people had been a customer rather than an event in the restaurant. In one case a customer had had a major flood and as that impacted all customers at once it appeared as a diagonal band of bad retention across the whole chart. In this case because it always hits after a certain period it is far more likley to be the way that customers are being treated.
And this is the strength of cohort analysis. By comparing the behaviour of similar customers, or prospects or leads over time you have the ability to get feedback on what you are doing right or wrong. Designing a new menu may result in increased sales – but if it is too narrow and does not have enough variety to keep customers coming back then this will be obvious in the cohort analysis far earlier than it will be in the sales.
This is just starting to scratch the service as all we are looking at is frequency of customer visits at this stage. If we start mashing it with sales or marketing data or using cohorts segmented by persona we can then start asking interesting questions about the impact of different changes on customer spend, the impact on long term customers and how easy it is to convert someone to a long term customer
As a final point. As a business process consultant data is absolutely critical to helping me make informed analysis. Without it I am waving my hands in the air and who knows how valuable that is. I spoke to one CEO about this and he said ‘We can’t do it. We’re not sophisticated enough’. He gave me access to the POS data and within just a few hours we had a fully functioning cohort analysis for his business that he was then able to use to make powerful decisions that had a significant impact on his success