Product Design · Leadership Transparency Index · 2026

Companies publish the data. It is public information.Nobody could read it.

The Leadership Transparency Index reads the two filings every public company already submits, an SEC proxy and an EEO-1, and turns them into one ladder you can compare: every rung on its own, the sample beside the share, and every figure traced to the document it came from.

The Power Funnel

The industry baseline, every level. LTI measures it per company, from each firm’s filings.

MenWomen

C-Suite

71%

29%

Sr Vice President

68%

32%

Vice President

65%

35%

Sr. Manager

61%

39%

Manager

58%

42%

Entry Level

51%

49%

Source: McKinsey

Role

Product and interface design, the choke-point metric, the classification rules, and the business model

Goal

Turn two unusable public documents into one visual record anyone can read, with every figure traceable to its source

Sector

Financial services · Governance

Timeline

WIN Challenge · one term · 4-person team · shipped live

TL;DR · 2 min readOr read the full study ↓

LTI

Problem

Public companies already publish this data, by law. It’s filed in a format built for regulators, not people, so almost nobody outside a company can actually read it or understand what it says.

Role

Product design end to end: the visualization system for reading a filing, the data-classification rules behind it, and the two-track certification model.

Approach

Turn the raw filings into one visual format anyone can actually read, compare, and make sense of, a single place built to hold many companies side by side rather than one filing at a time.

Outcome

Shown to people cold, with no explanation, about 90% understood the problem and what the index was communicating. Filings that were unreadable became something people could actually make sense of.

4Rungs reported on their own, never averaged into one figure

0Estimated values anywhere in the index. Every figure traces to a filing

A percentage of eleven people is a rumour. The design job was refusing to print it like a fact.

The design thesis

The Leadership Transparency Index landing screen, held on a tablet: the headline 'Power doesn't become invisible. It becomes narrow,' and the Power Funnel chart beside it.

Jump to the decisions

The problem

How might we make the data companies already publish readable, in a format the general public can actually understand?

The brief was the WIN Challenge’s Narrative pillar: change the story about women at work, at a scale that holds. The challenge’s own list of barriers names the broken rung, the level where women stop being promoted. Every public company already files who holds power, for free, by law. Almost nobody opens it, so the figure that circulates instead is a company-wide average, and an average hides the exact rung the brief is about.

That broken rung was never a headcount problem. Gender equity is not about a 50/50 ratio. It is about a 100% chance to grow and hold power, the same chance at every rung once you are already inside a company. A filing can be read to check that claim. A company-wide average cannot, which is exactly why almost nobody reads one.

A filing is a snapshot. It can show exactly where a ladder narrows and it can never say why. Every decision after this one is about holding that line on a screen, where a confident-looking chart will happily imply the explanation the data does not contain.

The reading

The answer is filed in two documents that were never built to agree.

One is written by counsel for investors, the other is a compliance form written to federal job categories. Between them they cover four levels of a company. Nothing maps one onto the other, so which level a person occupies depends on which document you opened first.

SEC proxy statement

Named individuals, one by one.

It names the board and the executive officers it is required to name, which is not always the whole executive team. Gender comes from the company’s own disclosure.

Board · C-suite

EEO-1 report

Aggregate headcounts, never names.

Filed with the EEOC and published by the company. Leadership arrives as totals inside the government’s own job categories, so nothing in it can be read down to a person.

Senior leadership · Managers

So the work is the reconciliation. Take the two filings, put their four levels on one ladder, and draw it, so that what a reader sees is which rung is which and where each figure came from.

Product

Pick a company → read the ladder → switch the unit → compare.

The Leadership Transparency Index turns two filings a company already submits into one ladder anyone can read: every rung reported on its own, the sample kept beside the share, and every figure traced to the document it came from. No company-wide average anywhere on it.

Prototype · not in the deployed buildAppleWomen’s share of seats, by rung · EEO-1 2024 and Proxy 2025

SectorCompany

ManagersEEO-1

30%

Senior leadershipEEO-1 · the break

22%

C-suiteSEC proxy

40%

Board of directorsSEC proxy

50%

Apple prints its own nominee gender matrix, so the board rung is a reading and not an inference. The C-suite is the proxy's stated group of twelve current directors and executive officers, less the eight directors it names.

Campaign film · on an OOH billboard

Who Decides

whodecidespower.lovable.appOpen the full index ↗

What the page does not have: a ranking table, because an ordered list invites a single score and a score restores the averaging this exists to break. A separate methodology route, because a method nobody scrolls to is a method nobody reads. Per-company URLs, because routing implies a directory and the deployed index holds two companies.

Decisions

Three questions, three decisions.

01The unit

A percentage was the obvious unit. It hides its own sample.

The filings hand you shares, so the obvious design prints a share. The first version reported one representation percentage per company, the way every DEI report does, and the way the source deck itself did.

And it failed immediately: two companies with completely different ladders can land on the exact same percentage, because the number carries no trace of the sample it came from. Sixteen percent of fifty-eight and sixteen percent of eleven print as the identical claim, and they are not the same claim.

One ratio per level replaced it, computed identically at every rung, with the shares it came from kept beside it and never averaged across the ladder. For every woman at this level, five and a half men. It is a sentence before it is a statistic. The price is that there is no single headline figure, so the finding is harder to carry in one breath.

A percentage describes a share. A ratio per rung describes a chance, which is the thing actually in question. Pick the string a reader will repeat, because that string is the product.

The full Where the ladder tightens screen: heading, company selector, and all four rungs, each ending on its choke point ratio. Manager 1 to 3.4, Senior leader 1 to 5.4, Board 1 to 1.8, C-suite 1 to 1.3. — The whole screen as it ships. Every rung reported on its own, the ratio carried at the end of each row, and no company-wide figure anywhere on it.

Tried first · one number per company~~A single representation percentage~~✗ Opposite ladders, identical number.

02The toggle

16% and 36% look equally solid. One of them is four people.

The ratio fixed the averaging but not the sample. A percentage of eleven renders at exactly the same weight as a percentage of four hundred, and small populations move violently while looking identical on a screen.

The next version showed percentages only and carried a footnote about sample size. A footnote is not a defence. It sits below the number that has already done the persuading.

So Seats and Percent went under one control, and counts appear wherever the filing publishes them. The second half of the fix, making the track itself carry the headcount, is prototyped above and is not in the deployed build, which still draws every rung to the same width. 16% is a texture. Nine women out of fifty-eight is a room, and a room is what gets repeated in a meeting. The price is a second unit to learn and a chart that changes shape under the reader.

Small-sample percentages are how leadership is made to look better than it is. The toggle is the part of the product that refuses.

The ladder in seats mode, controls included. Managers reads 89 of 392 on a full-width track; senior leadership 9 of 58 on a track roughly a seventh as long; board of directors 4 of 11 and C-suite 4 of 9 on tracks a few pixels wide. — The prototype in seats mode, drawn in the product’s own design. The track is the rung’s headcount, so the eleven-seat board can no longer draw like the four-hundred-person one. Switch it yourself in the section above.

Tried first · percent only~~Percentages only, as DEI reports doA footnote about sample size~~✗ A board of 11 outprints a workforce of 400.

03The comparison

A chart turns sixteen numbers into a shape. The reader wanted two numbers.

Comparison is the one screen where the reader arrives with the question already formed. They are not looking for a pattern. They are looking up a value.

The first version was a grouped bar chart: two companies, four rungs, four bars per rung. The two values a reader actually wants to compare are never adjacent, and sixteen values read as a texture rather than a table.

Rows became rungs and columns became companies, so the compared pair sits side by side, and the choke point is a third unit rather than a fifth column. The price is that a table is harder to feel than a chart. It also shipped a disabled “Multi-year coming soon” chip, which is a promise inside a product surface. That one is a mistake, not a tradeoff.

Precision beats impression once the reader has already decided what to look at.

The compare section: two company dropdowns, unit toggles, and a table of four roles across two companies. — The whole compare screen. Two dropdowns, the unit toggle, and the pair a reader came for on one line. The disabled multi-year chip is the mistake named above.

Tried first · grouped bars~~Grouped bar chart, four bars a rung~~✗ The compared pair is never adjacent.

Three decisions, all of them defensible on the screen. None of them survives if the two filings underneath are read differently twice.

How it’s calculated

Two filings in, one comparable ladder out. No estimates anywhere.

Nothing here is private. The index only reads what companies already file.

The hard part was never the arithmetic. It was deciding what a level is, when the two documents describing it were written by different institutions and agree on almost nothing. Get that mapping wrong and the one claim the index exists to test, whether a person has an equal chance to grow at every rung, stops being checkable at all.

The classification, and where it loses somethingFour rungs, their source, and what each one cannot say

Rung on screen	Source	What it actually gives	Where the classification is lossy
Board of directors	SEC proxy	Named individuals, counted one by one	Gender comes from the company’s own disclosure. Where it is absent the seat is flagged, never inferred from a name.
C-suite	SEC proxy	Named executive officers	The proxy names the officers it is required to name, which is not always the whole executive team. The rung can be narrower than the room.
Senior leadership	EEO-1	Aggregate headcount, “Executive and Senior Officials”	Already bucketed by the government. No names, no titles, nothing an outsider can audit at the level of a person.
Managers	EEO-1	Aggregate headcount, “First and Mid-level Officials”	One category spanning a seniority band no two companies would draw the same way. Comparable across firms only because the government imposed it.

One rung, end to end.

This is the whole conversion, run on the rung where Tradeweb breaks. Nothing between the document and the number is a judgement call we made about an individual.

01 · The documentEEO-1 report

Filed with the EEOC, published by the company. Downloadable by anyone, no request, no login.

Public filing · 2024

02 · The categoryExecutive and Senior Officials

The federal job category the filing already sorts people into. We map it to one rung and re-sort nothing.

As filed · not re-bucketed

03 · The count9 women, 49 men

Fifty-eight seats on this rung. Aggregate headcounts, so no individual is read, identified or assumed.

Counted · never modelled

04 · The number on screen1 : 0.0

The only derived value in the chain, and the shares it came from stay beside it so a reader can redo the division.

Derived · 49 ÷ 9

The arithmetic is now fixed: two filings in, one comparable ladder out. What the tool will not do with that ladder was settled before any of it was drawn.

What it refuses

The rules were fixed before anything was drawn, and every one of them costs something.

A measurement tool is defined by what it will not do.

Four rules, settled before the first screen. Each one made the product smaller.

Never estimate

If a filing does not say it, the product does not show it. No modelled values, no inferred gender, no filling a gap because the layout looks better full.

What it costsCoverage grows one document at a time, and some rungs stay visibly empty.

Never average across levels

Every rung is reported on its own. The company-wide number most DEI reporting leads with is the exact operation this product refuses.

What it costsThere is no single headline figure, so the finding is harder to carry in one breath.

Always name the source

Every figure carries the document and the year it came from, on screen, not in an appendix nobody opens.

What it costsMore chrome on screen than a marketing chart would ever tolerate.

Never rank

Profiles, not a league table. An ordered list invites a score, and a score hides the rung that matters.

What it costsIt will never travel the way a leaderboard would. That is the trade I would make again.

The states the build skipped, designed.

Rule 01 means this product spends most of its life with something missing, so the states where it has nothing to say are not edge cases here, they are the normal condition. The live build does not handle them. These four were drawn for this case study and never shipped. They are the first thing I would build.

Company not in the indexDesigned · not shipped

⚲Goldman Sachs

Not read yet

The index holds two companies today. Every figure is read from filings by hand, so coverage grows slowly and on purpose.

Tradeweb MarketsCitigroup

Today the field accepts anything and returns nothing legible. Stop pretending to be a directory: name the coverage before the reader types, and treat a miss as a fact about the index rather than a failed query.

Seats requested, no headcount filedDesigned · not shipped

SeatsPercent

Senior leader

Mennot published

Womennot published

Citigroup publishes shares without headcounts. Seats is available for 1 of 2 companies.

Today the row silently empties. The missing data is honest; going quiet about it is not.

Gender cannot be verifiedDesigned · not shipped

Board of directors · 11 seats

Women4

Men7

Uncertain0

A seat is counted only where identity is public. Unverifiable seats stay in the total, are reported on their own line, and are never assigned.

Gender is read from what a company discloses, never guessed from a name. The line reads zero here and still has to exist, because the first time it reads two the reader needs to know nothing was rounded away.

The gap has not been explainedDesigned · not shipped

Gap unexplained

Senior leadership, 1 : 5.4. No filed explanation for the drop-off from the rung below.

Awaiting company response

Gap explained

Company filed a specialisation-pipeline rationale with evidence. Assessed against the sector framework.

Evidence on file

Improving · flat · declining is reported separately from explained · unexplained. A company can be improving and still owe an explanation.

The interpretation layer the live build never reached, and the piece that keeps the index non-punitive. An unequal ratio is a finding; whether it is justified is a separate question, and collapsing the two turns a measurement tool into an accusation engine.

Four rules, each one making the product smaller. What survives them is a number a reader can check.

Outcomes

The whole point of this index is the broken rung: the specific level in a company’s leadership pipeline where women stop advancing at the same rate as men, the level a single company-wide average is built to hide. That is what these four rungs, and the number attached to each one, exist to show.

None of it is about forcing a 50/50 split. It is about whether a person has an equal chance to grow and hold power at every rung, once they are already inside a company. A filing can be read to check that chance, rung by rung. A company-wide average cannot, which is why one never appears anywhere on this page.

Shipped

Estimated values anywhere in the index

Observed · informal

~90%

Of people shown the product understood the problem and what it was communicating

Shipped

100%

Of the four leadership levels reconciled between two filing types never designed to agree

08 / What I'm still unsure aboutAn honest note

Readable is not the same as consequential. I only ever tested the first one.

The ~90% comprehension figure says people can look at this and understand the problem. It says nothing about whether understanding changes what anyone does next, whether a candidate picks a different employer, an investor asks a harder question, a company moves a number before the next filing. Rule 04 in what this refuses to do exists on purpose: no ranking, no league table, no accusation engine, because a tool built to pressure a company into moving is a different, less honest tool than one built to show a reader what a filing actually says. I chose the honest one without knowing whether it is strong enough to matter to anyone with the power to change the number.

✎ Margin note · to self

a leaderboard would travel further and pressure harder than a set of profiles ever will, and i chose profiles anyway.

i don’t know yet if honest and effective are the same product here.

Next case study

Qlarc.

Product DesignAI GovernanceB2B Platform

Designed every screen for the human who owns what the AI produces.

Read Qlarc

← All work Let’s talk →