How to Build a Data-Driven Customer Management System | by Hans Christian Ekne | Nov, 2024

Though a primary CBM system will supply some stable advantages and insights, to get the utmost worth out of a CBM system, extra superior elements are wanted. Beneath we talk about a couple of of a very powerful elements, corresponding to having churn fashions with a number of time horizons, including value optimization, utilizing simulation-based forecasting and including competitor pricing information.

A number of Horizon Churn Fashions

Generally it is smart to take a look at churn from totally different views, and a type of angles is the time horizon — or consequence interval — you permit the mannequin to have. For some enterprise eventualities, it is smart to have a mannequin with a brief consequence interval, whereas for others it could possibly make sense to have a mannequin with a 1-year consequence interval.

To higher clarify this idea, assume you construct a churn mannequin with 10-week consequence interval. This mannequin can then be used to present a prediction whether or not a given buyer will churn inside a 10-week interval. Nevertheless, assume now that you’ve remoted a particular occasion that you already know causes churn and that you’ve a brief window of maybe 3 weeks to implement any preventative measure. On this case it is smart to coach a churn mannequin with a 3-week horizon, conditional on the particular occasion you already know causes churn. This manner you’ll be able to focus any retention actions on the purchasers most liable to churning.

This type of differentiated strategy permits for a extra strategic allocation of assets, specializing in high-impact interventions the place they’re wanted most. By adapting the mannequin’s time horizon to particular conditions, firms can optimize their retention efforts, in the end bettering buyer lifetime worth and lowering pointless churn.

Pricing Optimization & Buyer Value Elasticity

Value is in lots of instances the ultimate a part of technique execution, and the winners are those who can successfully translate a technique into an efficient value regime. That is precisely what a CBM system with prize optimization permit firms to do. Whereas the subject of value optimization simply warrants its personal article, we attempt to briefly summarize the important thing concepts under.

The very first thing wanted to get began is to get information on historic costs. Ideally totally different ranges of value throughout time and different explanatory variables. This lets you develop an estimate for value elasticity. As soon as that’s in place, you’ll be able to develop anticipated values for churn at varied value factors and use that to forecast anticipated values for income. Aggregating up from a buyer stage offers the anticipated worth and anticipated churn on a product foundation and you could find optimum costs per product. In additional complicated instances it’s also possible to have a number of cohorts per product that every have their optimum value factors.

For instance, assume an organization has two totally different merchandise, product A and product B. For product A, the corporate needs to develop its person base and are solely prepared to just accept a set quantity of churn, whereas additionally being aggressive out there. Nevertheless, for product B they’re prepared to just accept a certain quantity of churn in return for having an optimum value with respect to anticipated revenues. A CBM system permits for the roll out of such a technique and offers the management a forecast for the long run anticipated revenues of the technique.

Simulation-Based mostly Forecasting

Simulation primarily based forecasting offers a extra sturdy approach producing forecast estimates quite than simply doing level estimation primarily based on anticipated values. Through the use of strategies like Monte Carlo simulation, we’re ready generate likelihood densities for outcomes, and thus present determination makers with ranges for our predictions. That is extra highly effective than simply level estimates as a result of we’re capable of quantify the uncertainty.

To grasp how simulation primarily based forecasting can be utilized, we are able to illustrate with an instance. Suppose now we have 10 clients with given churn chances, and that every of those clients have a yearly anticipated income. (In actuality we usually have a multivariate churn perform that predicts churn for every of the purchasers.) For simplicity, assume that if the client churns we find yourself with 0 income and in the event that they don’t churn we preserve all of the income. We are able to use python to make this instance concrete:

import random
# Set the seed for reproducibility
random.seed(42)
# Generate the lists once more with the required modifications
churn_rates = [round(random.uniform(0.4, 0.8), 2) for _ in range(10)]
yearly_revenue = [random.randint(1000, 4000) for _ in range(10)]
churn_rates, yearly_revenue

This offers us the next values for churn_rates and yearly_revenue:

churn_rates: [0.66, 0.41, 0.51, 0.49, 0.69, 0.67, 0.76, 0.43, 0.57, 0.41]
yearly_revenue: [1895, 1952, 3069, 3465, 1108, 3298, 1814, 3932, 3661, 3872]

Utilizing the numbers above, and assuming the churn occasions are impartial, we are able to simply calculate the typical churn charge and likewise the full anticipated income.

# Calculate the full anticipated income utilizing (1 - churn_rate) * yearly_revenue for every buyer
adjusted_revenue = [(1 - churn_rate) * revenue for churn_rate, revenue in zip(churn_rates, yearly_revenue)]
total_adjusted_revenue = sum(adjusted_revenue)
# Recalculate the anticipated common churn charge primarily based on the unique information
average_churn_rate = sum(churn_rates) / len(churn_rates)
average_churn_rate, total_adjusted_revenue

With the next numbers for average_churn_rate and total_adjusted_revenue:

average_churn_rate:0.56, 
total_adjusted_revenue: 13034.07

So, we are able to anticipate to have about 56% churn and a complete income of 13034, however this doesn’t inform us something concerning the variation we are able to anticipate to see. To get a deeper understanding of the vary of doable outcomes we are able to anticipate, we flip to Monte Carlo simulation. As an alternative of taking the anticipated worth of the churn charge and complete income, we as a substitute let the scenario play out 10000 occasions (10000 is right here chosen arbitrarily; the quantity needs to be chosen in order to attain the specified granularity of the ensuing distribution), and for every occasion of the simulation clients both churn with likelihood churn_rate or they stick with likelihood 1- churn_rate.

import pandas as pd
simulations = pd.DataFrame({
'churn_rate': churn_rates * 10000,
'yearly_revenue': yearly_revenue * 10000
})
# Add a column with random numbers between 0 and 1
simulations['random_number'] = (
[random.uniform(0, 1) for _ in range(len(simulations))])
# Add a column 'not_churned' and set it to 1, then replace it to 0 primarily based on the random quantity
simulations['not_churned'] = (
simulations['random_number'] >= simulations['churn_rate']).astype(int)
# Add an 'iteration' column ranging from 1 to 10000
simulations['iteration'] = (simulations.index // 10) + 1

This offers a desk just like the one under:

We are able to summarize our outcomes utilizing the next code:

# Group by 'iteration' and calculate the required values
abstract = simulations.groupby('iteration').agg(
total_revenue=('yearly_revenue', 
lambda x: sum(x * simulations.loc[x.index, 'not_churned'])),
total_churners=('not_churned', lambda x: 10 - sum(x))
).reset_index()

And eventually, plotting this with plotly yields:

The graphs above inform a a lot richer story than the 2 level estimates of 0.56 and 13034 we began with. We now perceive way more concerning the doable outcomes we are able to anticipate to see, and we are able to have an knowledgeable dialogue about what ranges of churn and income we we discover acceptable.

Persevering with with the instance above we might for instance say that we’d solely be ready to just accept a 0.1 % likelihood of 8 or extra churn occasions. Utilizing particular person buyer value elasticities and simulation primarily based forecasting, we might tweak the anticipated churn_rates for purchasers in order that we might precisely obtain this consequence. This type of buyer base management is simply achievable with a complicated CBM system.

The Significance of Competitor Pricing

One of the essential components in pricing is the competitor value. How aggressive rivals are will to a big diploma decide how versatile an organization could be in its personal pricing. That is very true for commoditized companies corresponding to utilities or telcos the place it’s laborious for suppliers to distinguish. Nevertheless, regardless of the significance of competitor pricing, many enterprise select to not combine this information into their very own value optimization algorithms.

The explanations for not together with competitor pricing in value algorithms are diversified. Some firms declare that it’s too tough and time consuming to gather the information, and even when they began now, they nonetheless wouldn’t have all of the historical past they should practice all the value elasticity fashions. Others say the costs of competitor merchandise will not be instantly akin to their very own and that accumulating them could be tough. Lastly, most firms additionally declare that they’ve value managers who manually monitor the market and when rivals make strikes, they will modify their very own costs in response, so that they don’t must have this information of their algorithms.

The primary argument can more and more be mitigated by good internet scraping and different intelligence gathering strategies. If that isn’t sufficient, there are additionally typically businesses that may present historic market information on costs for varied industries and sectors. Relating to the second argument about not having comparable merchandise, one also can use machine studying methods to tease out the precise price of particular person product elements. One other methodology can be to make use of totally different person personas that can be utilized to estimate the full month-to-month prices of a particular set of merchandise or product.

In the end, not together with competitor costs leaves the pricing algorithms and optimization engines at a drawback. In industries the place value calculators and comparability web sites make it more and more simple for purchasers to get a grasp of the market, firms run a danger of being out-competed on value by extra superior rivals.