Change the Denominator and the Success Disappears

Numbers don’t just measure the world; we use them to steer it. And the steering wheel is almost always the denominator.

Change the per what (per user, per dollar, per quarter, etc.) and the same dataset redraws reality.

We treat that choice like it’s plumbing, a technicality someone else settled in a meeting we missed. It isn’t.

The numerator can keep telling its story. Whoever picks the denominator decides whose world that story is true in.

The denominator is where judgment lives: what counts as a unit of life, how far harm is allowed to travel before it stops counting, where responsibility ends, and whose time remains invisible enough to leave off the ledger.

Diagram

Denominator frame

View source

%% title: Denominator frame
%% caption: The numerator can stay constant while the denominator changes whose world the metric describes. The doctrine lives in the frame, not just the figure.
flowchart TD
  classDef actor stroke-width:1.6px;
  classDef system stroke-width:1.2px;
  classDef gate stroke-width:1.6px;
  classDef binding stroke-width:3px,font-weight:bold;
  classDef status stroke-dasharray:4 3;
  classDef burden stroke-width:2px;
  classDef repair stroke-width:2.4px,font-weight:bold;
  classDef failure stroke-width:2.2px,stroke-dasharray:6 3;
  classDef concept stroke-dasharray:2 2;

  D["Same underlying data"]:::system --> N["Numerator claim"]:::system
  N --> Q["Per quarter / per transaction"]:::system
  N --> L["Per lifecycle / per person harmed"]:::system
  Q --> W([Reported win]):::status
  L --> T([Transfer exposed]):::failure

You can see this quiet distortion anywhere efficiency sells:

• A hospital trims length-of-stay; two weeks later, readmissions spike and families take the unpaid night shift.
• A university celebrates its graduation rate while debt and underemployment fall just outside the reporting window.
• A platform touts engagement per user; subtract the bots, coerced taps, and dark-pattern clicks, and the “growth” dissolves.
• An AI lab leads with overall accuracy; break results by subgroup, and the concentrated harm reappears exactly where it matters most.

Precision is cheap when the boundary lies

Most accountability reporting runs on confidence—crisp lines, precise numbers. But precision is cheap when the boundary does the lying.

If your window stops where harm starts, you didn’t improve; you exported. If your scope ends at the firm boundary, pollution becomes someone else’s weather.

An institution that picks its own denominator never has to report a false number. Hold the ruler still while the world moves, change the per what, and disaster becomes success. Institutions invite it when they mistake measurement for engineering (make the figure go up) rather than governance (justify the frame).

The change is both mundane and radical: make the frame visible. Say the doctrine out loud, then flip the claim. Don’t argue about the numerator until you’ve named the per what.

• “Cost per visit” is meaningless without “complications per patient within 90 days.”
• “Accuracy overall” is a performance until it sits next to “false negatives per high-risk subgroup.”
• “ROI per quarter” is a vibe; “net return per lifecycle, including cleanup” is a position.
• “Utilization per employee” flatters exhaustion; “error-plus-rework minutes per thousand cases” presents the bill.

If the headline survives its mirror, the success is real. If it collapses, you had transfer, and the next job is to find who carried it.

Wells Fargo built the mirror and then declined to show it. For years the bank reported “cross-sell,” the average number of products per retail household, as a measure of its relationships with customers. The figure sat a little above six against a long-advertised goal of eight, and every open account counted, whether or not the customer used it or had asked for it. According to the Securities and Exchange Commission, the retail bank’s own team built an “active cross-sell” that counted only products customers used. For the third quarter of 2015 the reported figure was 6.13 and the active one was 5.17, and more than half of retail households held at least one inactive account. Early drafts of the investor presentation meant to introduce the new measure said that as much as 20 percent of the accounts in the reported figure were inactive; by the end of February 2016 that reference was gone.1 The transfer had been carried by customers holding accounts they never wanted and by the more than 5,300 employees fired between 2011 and 2016 for the sales misconduct the goals rewarded. The bank stopped reporting the ratio in 2017.2

To find these transfers, ask of any metric:

Over what time? Per what unit? At what scope? Including whose labor?

If answering those flips the sign on the win, your success was compression. This requires habits, not heroics:

1. Put the frame (unit, horizon, scope) in the headline instead of burying it in a footnote.
2. Always show the counter-denominator most likely to invert the claim.
3. Rotate ownership of the math so the people who benefit from a metric are not the ones who define it.
4. Publish minority reports when a claim of success is only true under one convenient lens.
5. Reward the disclosure of uncertainty and off-ledger work, like moderation and exception handling.

What this slows down is mostly expensive speed. Velocity flatters until rework, churn, grievances, and remediation come due, and those bills arrive on the lifecycle’s schedule long after the quarter has been reported.

When denominators get daylight, debates stop impersonating metaphysics and return to honest politics: which horizon we privilege, which labor we refuse to steal, and which risks we decline to export.

The frame was chosen by somebody

Disclosure makes a frame readable and does nothing for its legitimacy. Most argument about metrics runs on the assumption that readability and legitimacy are one property rather than two. Naming the per what out loud tells you what the institution is for. Whether it was entitled to decide that is a separate question, and no amount of visibility answers it.

“Make the frame visible” settles which costs belong in the frame. It leaves alone who had the standing to set the frame, and the authority survives intact, because the honest version of a throughput objective is still a throughput objective.

Visibility lets the affected party argue. Which arguments they get to make is a different matter. If a metric treats patients or bank customers as inputs to a throughput number, publishing its denominator does not change what it allows.

Rotating ownership of the math helps, and it belongs in the discipline. Rotation is still a procedure, though, and if the people who benefit from a frame run the rotation, they can revise the frame without ever limiting it. The test is whether the people the frame consumes can change it by exercising an ordinary right, without first winning a contest the arrangement was not built to let them enter.

The heavier version of the discipline stops treating the important things as costs and starts treating them as limits. With dignity as a term to be traded, the optimizer has a price for it, and a large enough benefit on the other side buys any quantity of it. With dignity as a limit, there is no price, because the search stops at the floor, and a run that returns no feasible answer reports a fact about the world rather than a failure of ambition.

Some things belong on that side of the inequality. Burden that cannot be substituted, since a person cannot be paid instead of slept and no hourly rate repairs the difference. Agency that holds up once its cost is counted, since a right that costs more to use than the person can bear exists only on paper. And the right of the people inside the frame to challenge it, since a frame they cannot challenge has no claim to be optimized, however good its results look.

None of them is a metric anyone currently reports, and none can be satisfied by disclosure. Asking who gets to set a limit is a political question wearing an engineering costume. It also arrives last, which is not an accident. The denominator is much easier to win.

This logic applies beyond spreadsheets. A “frictionless” interface that offers a single, tidy answer is just shaving off context. This is flattening. Consolidation is convenient, but it’s also how scope creep turns into moral erasure.

In any system, never amputate the pathways people need to appeal, reverse, or recover.

You’ll know the culture is shifting when the language does. Teams start saying per relationship, not just per transaction. Investors ask about return on recovery, not just return per quarter.

The numerator can keep telling its story. Before believing it, ask who picked the denominator, and whose world that story is true in.