50 THENEWYORKER, MARCH 9, 2020
Kenshoo, based in Tel Aviv—to send sub-
contractors to work for him in San An-
tonio. Sprinklr also assigned remote em-
ployees, stationed in various time zones,
to crunch numbers at all hours. In addi-
tion to data provided by the R.N.C. and
traditional voter files, the Trump cam-
paign had access to a repository of infor-
mation provided by the Data Trust, a pri-
vate company that Karl Rove and other
conservative bigwigs had established in
- There are restrictions that prevent
certain kinds of data sharing among non-
profit political entities, but those don’t
apply to for-profit companies.
According to a source familiar with
the campaign, Parscale pitted Sprinklr
and Kenshoo against each other, hoping
to inspire a “trading-floor mentality.” The
idea came from Jared Kushner, who got it
from his friend Gabriel Leydon, a mobile-
gaming entrepreneur from Palo Alto.
(Leydon founded Machine Zone, now
called M.Z., a multibillion-dollar com-
pany known for the popular games Mo-
bile Strike and Game of War.) Parscale,
hoping to turn his operation into “a mer-
itocracy,” told the ad-tech firms that they
would have to compete, and that the win-
ner would earn the campaign’s business.
Over time, Kenshoo’s performance started
to lag, and the Sprinklr employees were
kept on. Sprinklr, which was valued at
$1.8 billion in 2016, now lists several of
its prominent past clients on its Web
site—Nike, NASA, Nasdaq—but makes
no mention of the Trump campaign.
One Project Alamo staffer took sev-
eral pages of notes during the campaign,
recounting the operation in remarkable
detail. (The notes have never been pub-
lished, but they have been shared pri-
vately with U.S. government officials.) I
reviewed the notes and spoke at length
with the person who wrote them, who
asked to remain anonymous. “I would
draw your attention, first of all, to what’s
not in them,” the staffer said. “They’re
not, for instance, about how I sat next
to some guy named Vlad who had a di-
rect line to the Kremlin.” If the Trump
campaign was accepting foreign inter-
vention, the staffer was implying, then
it wasn’t evident in the San Antonio office.
Parscale relied on Facebook to help
him accomplish several campaign objec-
tives, including persuasion, fund-raising,
and G.O.T.V., or “get out the vote.” Find-
ing and motivating likely voters through
traditional means, such as TV ads or
door-to-door canvassing, is expensive
and time-consuming compared with so-
cial media. The notes refer to a study
conducted on Facebook in which likely
Republican voters in early-voting swing
states were split into two groups: an
experimental group, which was given
information about early voting, and a
control group, which was not. The ex-
perimental group was more likely to be
aware that early voting was an option—
significantly more likely, for example, in
Florida, a state that Trump won by a sin-
gle percentage point.
Some of Parscale’s subcontractors in
San Antonio, including a couple of his
most trusted advisers, were employees of
Cambridge Analytica, the firm best
known for acquiring the data of eighty-
seven million Facebook users in 2014.
Asked about the firm’s impact on the
2016 election, the staffer said, “That’s an-
other story line that gets blown out of
proportion”: although some of the Cam-
bridge data was acquired under dubious
circumstances, “what they actually did
with it was pretty standard data science.”
When the data breach became interna-
tional news, in 2018, it incited a wave of
public outrage, not least because Cam-
bridge Analytica had a hand in some of
the most misleading political campaigns
in recent memory: Brexit; Uhuru Ken-
yatta’s disinformation-heavy campaigns,
in Kenya; the despotic propaganda tac-
tics of Rodrigo Duterte, in the Philip-pines. Still, the company’s central sales
pitch—that its proprietary “psychographic
modelling” allowed it to predict each
user’s deepest fears and desires—is now
widely dismissed as snake oil. “There’s
never been any public evidence that Cam-
bridge Analytica brought anything to
the table beyond what was standard cam-
paign practice,” Daniel Kreiss, a political-
communications professor at the Uni-
versity of North Carolina, told me. In
2016, the Trump campaign paid Cam-bridge Analytica slightly more than
six million dollars. Giles-Parscale was
awarded fifteen times more, making it
one of the most highly paid venders in
political history.
The office culture within Project
Alamo was one of brash experimenta-
tion—not unlike that of a successful but
amoral startup. According to the staffer’s
notes, all employees, from executives to
interns, were encouraged to voice their
ideas, no matter how ridiculous. For ex-
ample, one junior designer, responding
to the popularity of Pokémon Go, made
a video in which a Hillary-themed
Pokémon was being chased—it didn’t
make much sense, and it had little to do
with politics, but it turned out to be a
viral hit. Mark Zuckerberg once wrote a
manifesto of sorts, in which he encour-
aged his employees to follow what he
called “the hacker way”: “Instead of de-
bating for days whether a new idea is
possible or what the best way to build
something is, hackers would rather just
prototype something and see what works.”
In San Antonio, Parscale seemed to as-
pire to a similar ethos.
One of Parscale’s favorite Facebook
marketing tools was called Lookalike
Audiences. “I mean, it’s why the plat-
form’s great,” he said in an interview with
“Frontline,” in 2018. The tool works like
this: an advertiser uploads a “Custom
List,” an Excel spreadsheet of people the
advertiser wants to target. Even if the
spreadsheet comprises only scraps of in-
formation—an e-mail address here, a
mobile advertising I.D. there—Facebook,
with its unparalleled accretions of con-
sumer data, can usually fill in the gaps.
Lookalike Audiences then multiplies the
power of Custom Lists, using Facebook’s
proprietary software to replicate the tar-
get audience. If you have a Custom List
of three hundred thousand people, Par-
scale explained to “Frontline,” you can
use Lookalike Audiences to find another
three hundred thousand Facebook users
with attributes similar to those in the
first group. One of the most difficult
tasks of a political campaign—distin-
guishing likely supporters from the
undifferentiated mass of the American
electorate—can now be accomplished
instantly through artificial intelligence.
When the “Frontline” interviewer asked
how accurate Lookalike Audiences was,
Parscale called it “pretty amazing.”