Preventing churn with data. A practical guide.

Any successful business must have a working customer acquisition method. But the ones that stay successful are the ones that keep their customers for the long run.

Despite this fact, many succesful companies do not have a customer retention strategy.

Churn is especially relevant for businesses such as services, subscription models, or for lifestyle choices that involve ongoing expenses (diet, fashion, pets etc), for whom the customer lifetime value happens over longer stretches of time.

Targeting churn

There are many reasons people could churn.

  • They did not see the value of a repeat purchase.
  • They could be disengaged.
  • They might have found cheaper alternatives.
  • An event changed their ability to use your product/service
  • Service disruption.

And just as many things we could do about it.

  • We could do customer success campaigns
  • We could do engagement campaigns (or
  • We could make competitively discounted offers.
  • We could recommend a new product/service fitted to their needs
  • We can apologise with a small gift for service disruptions.

But what happens if we give the wrong message to the wrong customer? We would fail at achieving our communication goal and thus waste the opportunity, your time and theirs. Give them a wrong offer, and your communication is at best costly and ineffective. At worst, contacting low-risk users it can lead to more undesirable behaviours such as more churn (if it ain’t broke, don’t fix it)

It is only when our action solves a need of the customer, that they are likely to respond positively to it.

Such, understanding not only the risk to churn but also the reason is the only way to be effective at reducing it.

Collecting the data

Collect the likely churn reasons: Ask your customers, ask your customer support team, ask your common sense. It’s usually one or a handful of reasons that end up covering most of the cases. Once you have this understanding, you are able to define actual datapoints that may be useful in a prediction.

For instance, your data points could look like :

Customer might be price sensitive and looking for a cheaper alternative:

  • % discounted purchase ratio
  • % bottom shelf purchase ratio
  • Bargain/sales participation rate
  • average basket value

Customer might be unsuccessful at extracting value from your proposition:

  • Customer success funnel percent complete
  • Last action (eg, abandoned cart, payment due, purchase, return, cancellation, customer support call)

Customer is disengaged from your proposition:

  • Last action days ago
  • avg time between orders
  • order cycles missed (time since last order / avg time between orders)
  • Subscription cycles unused
  • Depth of action (some actions are more relevant, like referring a friend or sharing your posts)

Customer needs something else now.

  • As this is an external reason for a customer’s need changing (eg, pregnancy ending, moving city), there is probably not much you can do about it. But you can try leading them to a service that suits their new needs, if applicable.

Customer is dissatisfied:

  • Recent actions (eg, returns, support calls, complaints)
  • Customer feedback (calls, complaints, comments)
  • Return rate (% products returned after purchase)
  • Usage rate (where measurable)
  • NPS survey
  • User received crappy products (poor general feedback)
  • Was active user during service outage

After checking the relationship between the features used for prediction and the churn events, predict the likelyhood or churn and the reason. (or let us do it for you)

Assess the business impact

Not all types of churn can be prevented, so focus on the main preventable reasons and create targeted campaigns. Before going overboard, make sure that the expected improvement is large enough to justify the cost of taking action (enough users, enough value).

Make sure you measure performance against your goal correctly.

But they churn even faster!??

– my reaction to a churn prevention campaign after a few days

Users with a high risk of churn are like mimosas – if you have any touchpoints, they might cancel right away.

However, silence is not the answer. If you do nothing, they all cancel. If you reach out, Some will cancel sooner, but some will come back to being an active customer. But 10% of users cancelling 10% sooner means a 1% CLV loss, while a 10% reactivation rate with a 50% repurchase value is a 5% CLV gain.

Despite some of them churning faster, the overall cohort will churn less.

… so actually we reduce churn

Measure what you really care. In some cases companies want to maximize profitability, in which case CLV optimisation would be crucial. In other cases, it’s a market share grab and all that matters are subscribed customers. Make sure your campaigns are performing against goals, and don’t worry about non-goals.

Why do this at all?

On paper it might look like a lot of effort for a small CLV boost, but how does it look in practice? In practice, we have multiple processes we optimise with data, all with compounding results.

Assume a 20% CLV increase due to targeted CRM treatments.

This 20% CLV increase can translate into +50% or more max CAC (what you can spend on a customer while still being profitable, assuming your CAC is 30-60% of Revenue)

This max CAC change now allows you to purchase another cohort of users (at a price between previous maxCAC and current maxCAC), say another 15%.

But your current users being more loyal also translates to an increased (longer-living) user base, say 15%

And all the extra profitable users, and all the extra profit from existing users can easily translate to another 30-50% to your revenues.

So while at first it may seem small, from a modest CRM performance of +20% we end up with +30% user base and +40% profit. This is the difference between a business slowly dwindling, or winning the market. And all the numbers above are rather conservative.

Plug

Of course, if you need help with any of the above, don’t hesitate to get in touch.

Leave a comment

Your email address will not be published.