“I spent exactly enough.”
She sat down.
“What do you recommend?”
“Preserve.”
“Then independent audit.”
“Before public conclusions.”
Naomi looked at the viral screenshot.
“People already concluding.”
“People aren't governance.”
That sounded so much like Marcus she nearly laughed.
“Fine.”
The Arden & Vale CEO, Priya Desai, agreed to an emergency review.
Priya was fifty-four, Indian American, had joined the company after Marcus stepped down.
She had no sentimental attachment to old stories.
That made Naomi trust her more.
At the first meeting, Priya did something Marcus would have appreciated.
She began with:
“I approved expansion of this system.”
No deflection.
No “the company.”
“I did.”
Naomi looked at her.
“Why?”
“Revenue efficiency.”
“Client coverage.”
“Associates were spending disproportionate time on low-conversion traffic while known clients complained about access.”
Jordan joined remotely.
“Did you evaluate demographic proxies?”
“Yes.”
“And?”
“No protected-class fields.”
Jordan closed his eyes.
“That isn’t what I asked.”
Priya frowned.
Jordan continued.
“You don’t need race field to reproduce racial or socioeconomic structure.”
“We tested zip-code effects.”
“How?”
“Legal reviewed.”
“That is not fairness testing.”
The company’s chief data officer, Daniel Cho, interrupted.
“We did disparate-impact simulations.”
Jordan looked at him.
“Show me.”
They did.
The model appeared statistically neutral across broad census racial categories at national level.
That seemed reassuring.
Jordan asked for subgroup performance.
City.
Age.
Data sparsity.
Privacy-trust customers.
People with common names.
People without social media.
People who used cash.
People whose occupations did not map to high-income databases.
Patterns appeared.
Not simple Black-versus-white.
Something more interesting.
Customers with thin commercial data profiles were routinely scored lower.
Who had thin profiles?
Privacy-conscious wealthy people.
Recent immigrants.
People using family trusts.
Older customers with little online activity.
Younger professionals without established luxury history.
Cash users.
Some public-sector employees.
Some doctors and lawyers whose wealth sat outside consumer-data networks.
And disproportionately, in several urban markets, Black and Latino high-income professionals with fewer connections to luxury-industry data ecosystems.
The model had not learned race.
It had learned legibility.
Who looked rich to data brokers.
Naomi stared at chart.
Jordan said quietly:
“There it is.”
Priya looked sick.
Daniel leaned forward.
“This does not prove discrimination.”
“No.”
Jordan agreed.
“It proves the model rewards people the data ecosystem understands.”
“What’s wrong with that?”
asked the CFO.
Jordan looked at him.
“Nothing, if you use it to decide who gets a birthday email.”
“Potentially a lot, if you use it to decide who gets to see merchandise.”
Silence.