Analytics creates value in three places: better management decisions, sharper strategy, and the product itself.
Most organisations already hold the data, generated in their own systems or accessible to them. Holding it is the easy part. The harder part is deciding which question it is supposed to answer, and that decision comes from the business model rather than from the data team.
The KPI set falls out of the business model
Four questions on the core dimensions of the business model set the frame. They come before any tooling decision, because they determine which numbers are worth instrumenting.
- Customers: Who are the target customers of the products and services?
- Value proposition: How is the value add generated for those customers?
- Monetisation: What are the revenue-generating mechanisms?
- Value chain: What are the key activities to deliver value to customers?
The answers narrow the KPI set to the numbers that actually reflect the growth model. That set is what a Business Intelligence and Analytics Strategy gets built on.
Questions that come up in that process:
- “Who are our most profitable customers and why is that the case?”
- “What is the value-maximizing pricing strategy?”
- “What should be the focus of our retention measures?”
Three types of analytics, each built on the one below
The order matters: a forecast is only as good as the reconciled history underneath it.
- Descriptive analytics: “What has happened?” The focus is data aggregation and evaluation, identifying patterns and dependencies and generating a basis for business decisions. This can be a one-time exercise as well as an ongoing one.
- Predictive analytics: “What may happen?” Based on different models, an extrapolation of the future is generated through algorithms. The interpretation of data is done through human-machine interaction. This can be applied for example to forecasting customer churn.
- Prescriptive analytics: “What should be done?” The in-depth understanding of actual data reflects back on strategy in order to defend competitive advantages or create new ones.
Two views on churn that change the LTV number
Churn sets the ceiling on customer lifetime value (LTV). A churn-prevention, win-back and re-activation strategy therefore does two things at once: the customer base grows faster, and the base available for upselling grows with it.
Cohort analysis: de-averaging what changed and when
Build up a detailed view of your customer cohorts and their development over time. Cohort analysis de-averages changes that were applied in the past, a pricing move or a shift in marketing channel mix or products sold, and thereby generates insights into the true drivers of business performance. It also allows general patterns of user behaviour to be derived, which can be used for forecasting future revenues and profits.
Plot monthly retention by monthly cohort and the spread around the average becomes visible: some cohorts perform materially better than others. The differences usually trace back to something concrete, a different product mix or a different usage pattern. Build the same view not only for retention rates but for average revenue of the remaining customers, ARPU (Average Revenue per User), the number and type of products used, and COGS (Cost of Goods Sold), depending on the business model.
Combining two cohort views is where the analysis earns its cost. In one example, cohort sizes rose while retention stayed stable, and the outsized cohorts retained worse because they had been bought with aggressive discounts. The general lift across all retention curves traced back to added product features.
LTV driver analysis: where contribution margin stops working
This breaks the lifetime value of a cohort, a customer segment or a product segment into the drivers that produce it. Contribution margin gives one number at one point in time. The driver view gives the development over time, which is what a decision needs.
The results are frequently counter-intuitive. A segment that looks unprofitable on direct margin can be LTV positive once indirect effects are counted, for example the new customers its members recommend.
Drivers cluster along six dimensions:
- All acquisition costs and efforts: Marketing and sales activities as driver
- Revenue: Product mix and pricing as drivers
- COGS: Direct variable costs of delivering the product or services
- Upselling and downgrading: Account management and CRM activities
- Churn: Active churn prevention and win-back activities
- Indirect effects: Recommendation of users and other viral effects that drive new customers
Analytics: the questions we get asked most
What is the difference between descriptive, predictive and prescriptive analytics?
Descriptive analytics reports what happened, from reconciled actual data. Predictive analytics estimates what may happen, by extrapolating with models. Prescriptive analytics says what to do about it, by feeding the conclusion back into strategy and pricing. The three stack, and a prediction inherits every error in the history it was trained on.
What is a cohort analysis?
A cohort analysis groups customers by the period they were acquired in and tracks each group separately over time. It de-averages the effect of past changes, a price move or a shift in channel mix, which an aggregate figure hides. Build it for retention first, then for ARPU, product mix and COGS.
What is LTV driver analysis?
It breaks the lifetime value of a segment into the components that produce it: acquisition cost, revenue, COGS, upselling and downgrading, churn, and indirect effects such as recommendations. That replaces a single contribution margin at one point in time with a view of how value develops.
Where should an analytics programme start?
Start with the business model rather than the tooling. Work through the four dimensions, customers, value creation, monetisation and delivery, and the KPI set follows from the answers. A stack chosen first ends up measuring whatever was easiest to connect.
Why is churn the metric to start with?
Because it caps everything above it. Retention decides how many periods the margin is earned over, so an improvement compounds across the whole relationship. It is also where a cohort view changes the answer most: an aggregate rate hides which cohorts are leaving and when.
Authors
Stefan Benndorf
stefan.benndorf@scaleon.com
Dr. Philipp Engelhardt
philipp.engelhardt@scaleon.com
