Chapter 1 of 8

Framing the Decision and Choosing KPIs

3 sections · ~8 min read · 3 practice questions · part of Business Analytics

Learning objectives

  • Convert a vague business goal into a decision with named options and a measurable outcome
  • Build a KPI tree that decomposes a headline metric into drivers a team can actually move
  • Compare candidate initiatives on net effect rather than on the size of the driver they touch

Analysis begins with a decision, not with data

An analysis can be technically sound and still leave the decision maker unsure what to do. Before pulling a single row of data, write down the decision in a single sentence with three parts: who decides, what the options are, and what would have to be true for each option to win. If you cannot name at least two options, there is no decision, only a report. If you cannot name the evidence that would change the decider's mind, the analysis has no stopping point and will expand until the deadline. A useful discipline is to write the conclusion sentence with the numbers left blank, for example, we should choose option A over option B because it adds roughly blank dollars per month at a cost of blank dollars per month. Filling in the blanks is then the entire scope of work. This framing also fixes the unit of analysis early. Revenue per month, contribution per unit, and lifetime value per customer are three different currencies, and mixing them mid-analysis is a source of misleading recommendations. State the currency, the time period, and whether the figure is gross or net before you begin.

KPI trees and the difference between drivers and outcomes

A key performance indicator is useful when it is both diagnostic and actionable. Headline metrics such as monthly recurring revenue or gross margin are diagnostic but rarely actionable, because no single team controls them. The fix is a KPI tree: decompose the headline metric multiplicatively or additively until you reach quantities a named team can change within a quarter. Monthly recurring revenue equals active subscribers times average revenue per user. Active subscribers next month equals active subscribers this month, times one minus the churn rate, plus gross additions. Gross additions equal trial starts times trial-to-paid conversion. Now marketing owns trial starts, product and onboarding own conversion, and customer success owns churn, and each owns a number that visibly rolls up. Distinguish leading from lagging indicators as you build the tree. Churn is lagging; it confirms what already happened. Product usage in the first fourteen days is leading; it moves first and gives you time to intervene. Finally, guard against the incentive problem: any metric that becomes a target invites gaming, so pair each target metric with a guardrail metric that would visibly deteriorate if the target were hit the wrong way. Pair conversion targets with refund rate, and volume targets with margin.

Sizing an initiative in steady state

Subscription and repeat-purchase businesses behave like a tank with an inflow and a leak. If gross additions per month are a and the monthly churn rate is c, then the subscriber count settles at the level where the leak exactly matches the inflow. Setting subscribers next month equal to subscribers this month in the recursion gives a steady state of a divided by c. This one expression does most of the work in initiative sizing, and it carries an important warning: a percentage point of churn and a percentage point of conversion do not have equal value, because churn sits in the denominator. Cutting churn from four percent to three percent raises the steady state by a third; cutting it from two percent to one percent doubles it. Two caveats matter in practice. First, steady state is a destination, not a date. The gap to steady state closes by roughly the churn rate each month, so at four percent monthly churn about half the gap closes in seventeen months, and quoting a steady-state number as though it arrives next quarter overstates the near-term case. Second, always subtract the cost of the initiative. A larger gross effect at a larger cost can lose to a smaller effect that is nearly free.

Worked example

Problem

A hypothetical subscription tool has 8,000 active subscribers, average revenue per user of 25 dollars per month, 1,600 trial starts per month, a trial-to-paid conversion rate of 20 percent, and monthly churn of 4 percent of beginning subscribers. Two proposals compete for one quarter of engineering time. Proposal A rebuilds onboarding to lift conversion from 20 percent to 23 percent and costs 12,000 dollars per month to run. Proposal B adds a save-offer flow that cuts monthly churn from 4.0 percent to 3.5 percent and costs 10,000 dollars per month. Compare them on steady-state monthly recurring revenue, net of running cost.

Step by step

  1. Confirm the current state is internally consistent. Gross additions a = 1,600 x 0.20 = 320 subscribers per month. Steady state = a / c = 320 / 0.04 = 8,000 subscribers, which matches the stated base. Current MRR = 8,000 x 25 = 200,000 dollars per month.
  2. Proposal A. New gross additions a = 1,600 x 0.23 = 368 subscribers per month. Churn is unchanged at 0.04.
  3. Proposal A steady state = 368 / 0.04 = 9,200 subscribers. MRR = 9,200 x 25 = 230,000 dollars per month. Gross gain = 230,000 - 200,000 = 30,000 dollars per month.
  4. Proposal A net gain = 30,000 - 12,000 = 18,000 dollars per month.
  5. Proposal B. Gross additions stay at 320 per month. New churn c = 0.035.
  6. Proposal B steady state = 320 / 0.035 = 9,142.86 subscribers. MRR = 9,142.86 x 25 = 228,571.43 dollars per month. Gross gain = 228,571.43 - 200,000 = 28,571.43 dollars per month.
  7. Proposal B net gain = 28,571.43 - 10,000 = 18,571.43 dollars per month.
  8. Compare. A wins on gross effect by 30,000 - 28,571.43 = 1,428.57 dollars per month, but B wins on net by 18,571.43 - 18,000 = 571.43 dollars per month, because B costs 2,000 dollars per month less to run.
  9. Check the timing caveat. Under Proposal B the gap to the new steady state closes at about 3.5 percent per month, so after 12 months roughly 1 - (1 - 0.035) raised to the 12th power = 1 - 0.652 = 35 percent of the 1,142.86 subscriber gain has arrived, about 400 subscribers, or 10,000 dollars per month of the eventual 28,571.

Answer. Steady state under A is 9,200 subscribers and 230,000 dollars per month of MRR; under B it is 9,142.86 subscribers and 228,571.43 dollars per month. Net of running cost, A adds 18,000 dollars per month and B adds 18,571.43 dollars per month, so B is marginally better. The two are within about 3 percent of each other, which is well inside the uncertainty on the conversion and churn estimates, so the estimates do not support a clear preference on value and the choice should be made on secondary grounds such as implementation risk, reversibility, or which team has capacity. Both figures are steady-state values in dollars per month and neither is reached within the first year.

Practice

Work each question before opening the solution.

  1. A retention team is given a target of cutting monthly churn from 5 percent to 4 percent. Name a guardrail metric that should be tracked alongside it, and explain the specific way the target could be hit while making the business worse.

    Show solution for question 1

    Track discount depth or realised average revenue per user, and separately track the churn rate of customers who never received a save offer. Churn can be cut by handing out steep permanent discounts to anyone who tries to cancel, which reduces the churn rate while cutting revenue per user and, once the offer becomes known, teaching healthy customers to threaten cancellation. The headline metric improves and contribution falls.

  2. A business has 450 gross additions per month and 5 percent monthly churn. Compute its steady-state subscriber count. Then compute the steady state if churn is halved to 2.5 percent, and state in one sentence why the second change is worth more than a 450 to 500 increase in gross additions.

    Show solution for question 2

    Steady state = 450 / 0.05 = 9,000 subscribers. With churn at 2.5 percent, steady state = 450 / 0.025 = 18,000 subscribers, a gain of 9,000. Raising gross additions from 450 to 500 at 5 percent churn gives 500 / 0.05 = 10,000, a gain of only 1,000. Churn sits in the denominator, so halving it doubles the steady state, while additions in the numerator scale it only proportionally.

  3. An executive asks for a dashboard showing everything about customer behaviour. Rewrite the request as a decision framing with two options and one deciding number.

    Show solution for question 3

    A workable reframing: the decision is whether to move the next quarter of support headcount into proactive onboarding calls or leave it in reactive ticket handling; the deciding number is the difference in ninety-day retention between customers who did and did not receive an onboarding call, expressed in retained dollars per month per support hour. Two options, one currency, one number that flips the choice, and a scope that ends when that number is estimated.