The unit must reflect how value is created

A unit is a description of what is being counted, not a convenient divisor chosen after a cost total is known. An order can contain many items, a contract can serve many locations and a customer can maintain several accounts. Dividing cost by whichever count is largest makes the result look efficient while changing the object of the decision. Before comparing commercial alternatives, the analyst must say which object receives the service and which object gives rise to the next obligation.

Meridian’s illustrative unit is an expected active customer receiving a recurring monthly service. Its acquisition unit is a newly acquired customer, and its price is a monthly amount for a surviving customer. Those units connect the acquisition process to the service process through an explicit cohort relationship. They do not imply that every customer uses equal support time in practice. Equal service cost is a supplied simplification, suitable for explaining the mechanism and unsuitable for silently representing a heterogeneous operating base.

The physical unit in an industrial business can be very different. A tonne delivered, a loaded vehicle movement or a completed maintenance task may drive much of the work. The commercial contract might charge per shipment or bundle access for a period. In that case a commercial denominator and an operating denominator both matter. Revenue per customer could rise while cost per task also rises because the service intensity changes. A customer ratio cannot reveal that mechanism unless the unit bridge is retained.

A useful unit bridge lists the commercial object, service object, cash object and capacity object. The commercial object is what the buyer agrees to purchase. The service object is what staff and equipment must perform. The cash object is the invoice, milestone or deposit whose payment creates liquidity. The capacity object is the constrained work package, slot or qualified resource. They can coincide, but the analyst should show that coincidence rather than assume it.

The danger of a mixed denominator becomes clear when sales staff call a new location a customer while finance counts the legal billing entity. Acquisition cost per customer then changes even though campaign spending and the underlying relationships are unchanged. A multi-location customer can look expensive to acquire in one report and cheap in another. Neither number is meaningful until the customer identifier and activation rule are stable. The correction is a reconciliation of counts, not a discussion about which team has the better metric.

Time is another part of the unit. Monthly contribution cannot be compared with annual acquisition spending without a stated allocation or timing bridge. A customer acquired late in an appraisal window has fewer service months inside that window than a customer acquired at its beginning. Combining the totals and dividing by all acquired customers can therefore produce a value that describes the calendar window rather than a new customer’s expected contribution path. The distinction matters when acquisition is accelerating.

The canonical engine preserves both calendar and age dimensions. Monthly rows describe the organisation at a calendar month; the one-customer curve describes a new customer at successive ages; cohort rows connect the two. This architecture makes it possible to ask what is missing from an average. It does not make a scalar automatically comparable to another scalar. The analyst still needs the unit, timing and cost perimeter printed beside the value.

The correct denominator depends on the proposed action. A question about acquiring a relationship needs an acquisition unit that corresponds to the relationship the campaign is intended to create. A question about serving an existing base needs a unit linked to service intensity. A question about a constrained support team may need task demand rather than customer count. It is possible to use several units well; it is dangerous to merge them into a single score whose denominator changes between alternatives.

Meridian’s completed unit record treats an active customer as a service account with a formally continuing subscription-like relationship. It excludes dormant prospects, duplicate accounts and organisations that have not begun the promised service. The illustrative engine starts that service in the acquisition month. If a real onboarding process takes time before activation, the modelling convention has to change or a dated service-start schedule has to be built outside this tool. The label should not be preserved while the meaning changes.

Acquisition cost needs a corresponding boundary. The numerator might include campaign purchases, sales commissions, staff time and partner payments. The denominator might count signed contracts, activated accounts or paying accounts. A number created with signed contracts cannot be read as cost per successfully activated paying customer unless the intervening attrition is accounted for. The paper’s fixed supplied CAC is a literal model assumption. It is not a recommendation to ignore those transitions when collecting evidence.

A dispute about unit choice should be settled by tracing an individual record through the proposed definitions. Take a fictional organisation that signs a contract, activates several accounts and pays a consolidated invoice. Ask which event enters the acquisition count, how many service units appear, which entity owes payment and what capacity obligation is created. This trace often exposes ambiguity more quickly than averaging a large dataset. It also gives data engineers a concrete acceptance test for the eventual metric pipeline.

An important falsifier is a change in the relationship between units. A package revision can add service tasks without changing customer count. A channel can aggregate several end customers into one reseller account. A new billing system can split one legal customer into several identifiers. When these events occur, a stable looking denominator can conceal a changed operating object. The unit record should require a bridge across the change, rather than pretending a continuous chart proves comparability.

The unit record also states what is not being represented. An expected fractional customer is a mathematical expectation, not a partial person receiving an actual contract. It can be useful in a deterministic cohort calculation because expected demand and contribution need not be integers. It cannot allocate an individual staffing slot or guarantee enough resources for a cluster of simultaneous arrivals. If the decision is about discrete staffing, the service owner must supply a discrete workload plan and contingency arrangement.

The foundation acquires 75.00 expected customers in its first month. The first acquired cohort contributes 80,441.21 USD through the illustrative horizon, while the final acquired cohort contributes 4,575.00 USD through that same calendar window. Both cohorts start with the same acquired count and input assumptions. Their totals differ because their exposure inside the finite calendar window differs. This comparison supplies no evidence that the later cohort is intrinsically inferior.

This distinction prevents a common allocation error. If an analyst divides total business contribution over the window by all acquired customers, the result combines older and newer cohorts with different service ages. It may be useful as a calendar-window management statistic if labelled accurately. It is not the same quantity as the finite-horizon value of one newly acquired customer followed over its own complete age window. The model provides both kinds of building blocks, but it does not excuse mixing their interpretations.

Customer economics also needs a distinction between entity and household, buyer and user, payer and recipient. A business-to-business account can be commercially retained while individual users stop using the service. A user can remain active while the legal payer stops paying. Each event has a different implication for revenue, service cost and cash. The illustrative homogeneous account simplifies those relationships. A real metric specification should preserve the links needed for the decision rather than assuming that one active flag settles all of them.

The practical test for a proposed unit is whether two independent owners can reconstruct the same count from the same underlying events. The commercial owner states the business meaning; the data steward implements it; the finance controller reconciles it to monetary records where relevant. Disagreement is evidence about the metric, not a rounding nuisance. Until it is resolved, a precise acquisition ratio should be treated as a hypothesis about how records map to relationships.

There is no universal requirement to use the smallest possible unit. A contract may be the useful decision unit when obligations are negotiated and delivered as a package. Splitting it into artificial customer fragments can remove the very feature that drives value. The unit choice should preserve the mechanism under review, remain consistent across alternatives and expose its bridge to cash and capacity. That standard is stronger than an attractive denominator and more practical than insisting that every management report use identical counts.