WELCOME DAVE GIROUARD, CO-FOUNDER & CEO The Opportunity AGENDA PAUL GU, CO-FOUNDER & CTO Our AI Advantage Break CHANTAL RAPPORT, CMO Our Category-of-One Business SANJAY DATTA, CFO Profitability and Resilience Executive Q&A During the course of AI…
WELCOME
DAVE GIROUARD, CO-FOUNDER & CEO
The Opportunity
AGENDA
PAUL GU, CO-FOUNDER & CTO
Our AI Advantage
Break
CHANTAL RAPPORT, CMO
Our Category-of-One Business
SANJAY DATTA, CFO
Profitability and Resilience
Executive Q&A
During the course of AI Day, and throughout this presentation, the executive team of Upstart Holdings, Inc. (the “Company”, “we”, or “our”) may make “forward-looking” statements that are based on our
management’s beliefs and assumptions and on information currently available to management. Forward-looking statements include all statements, other than statements of historical fact discussed during AI Day
and contained in this presentation, including but not limited to, information or predictions concerning our future financial performance, including our financial outlook for Q1 2025 and the full fiscal year 2025,
projected growth and other strategies, business plans and objectives, potential market and growth opportunities, competitive position, technological or market trends and industry environment. Forward-looking
statements are inherently subject to risks and uncertainties, some of which cannot be predicted or quantified. In some cases, you can identify forward-looking statements by terminology such as “may,” “will,”
“should,” “could,” “expect,” “plan,” anticipate,” “believe,” “estimate,” “predict,” “intend,” “potential,” “would,” “continue,” “ongoing” or the negative of these terms or other comparable terminology. You should not put
undue reliance on any forward-looking statements. Forward-looking statements should not be read as a guarantee of future performance or results, and will not necessarily be accurate indications of the times at, or
by, which such performance or results will be achieved, if at all. This presentation should be reviewed in conjunction with the accompanying audio-visual recording available on Upstart’s Investor Relations website
(ir.upstart.com). Important context or clarifications provided during the live AI Day discussion may not be fully captured here. Upstart assumes no responsibility for any misunderstanding or incomplete interpretation
resulting f rom reviewing these materials without the benefit of the live AI Day discussion.
Forward-looking statements are based on information available at the time those statements are made or management’s good faith beliefs and assumptions as of that time with respect to future events, and are
subject to risks and uncertainties that could cause actual performance or results to differ materially f rom those expressed in, or suggested by, the forward-looking statements. In light of these risks and uncertainties,
the events and circumstances contemplated by the forward-looking statements made in this presentation may not occur and actual results could differ materially f rom those anticipated or implied in the forward-
looking statements. These risks and uncertainties are more fully described in our filings and reports that we may file f rom time to time with the Securities and Exchange Commission (the “SEC”), including “Risk
Factors” in our most recent Annual Report on Form 10-K and Quarterly Reports on Form 10-Q. These risks and uncertainties include, but are not limited to: our ability to manage the adverse effects of macroeconomic
conditions and disruptions in the banking sector and credit markets, including inflation and related changes in interest rates and monetary policy; our ability to access sufficient loan funding, including through
securitizations, committed capital and other co-investment arrangements, whole loan sales and warehouse credit facilities; the effectiveness of our credit decisioning models and risk management efforts, including
reflecting the impact of macroeconomic conditions on borrowers’ credit risk; our ability to retain existing, and attract new, lending partners; our future growth prospects and financial performance; our ability to
manage risks associated with the loans on our balance sheet; our ability to improve and expand our platform and products; and our ability to operate successfully in a highly-regulated industry. Moreover, we operate
in very competitive and rapidly changing environments, and new risks may emerge f rom time to time. It is not possible for us to predict all risks, nor can we assess the impact of all factors on our business or the
extent to which any factor, or combination of factors, may cause actual results to differ materially f rom those contained in any forward-looking statements we may make. Additional information will be available in
other future reports that we file with the SEC f rom time to time, which could cause actual results to vary f rom expectations. Except as required by law, Upstart does not undertake any obligation to publicly update or
revise any forward-looking statement, whether as a result of new information, future developments or otherwise.
This presentation contains statistical data, estimates and forecasts that are based on independent industry publications or other publicly available information, as well as other information based on our internal
sources. This information involves many assumptions and limitations, and you are cautioned not to give undue weight to these estimates. We have not independently verified the accuracy or completeness of the
data contained in these industry publications and other publicly available information. Accordingly, we make no representations as to the accuracy or completeness of that data nor do we undertake to update such
data after the date of this presentation. Some charts in this presentation display metrics over differing periods. These variations are intentional and reflect our effort to present data in the most relevant and efficient
way possible and for you to appreciate the nuances of our business.
This presentation and the live AI Day discussion may include references to certain non-GAAP financial measures, including, but not limited to, contribution profit, contribution margin adjusted EBITDA, adjusted
EBITDA margin, adjusted net income (loss), and adjusted net income (loss) per share. These non-GAAP financial measures are in addition to, and not a substitute for or superior to measures of financial performance
prepared in accordance with GAAP. There are a number of limitations related to the use of these non-GAAP financial measures. For example, other companies may calculate similarly-titled non-GAAP financial
measures differently. Please refer to the appendix for a reconciliation of the non-GAAP financial measures included in this presentation to the most directly comparable GAAP measures and please see our most
recent Annual Report on Form 10-K and Quarterly Reports on Form 10-Q for a description of non-GAAP financial measures that may be referenced in this presentation and the live AI Day discussion.
FORWARD LOOKING STATEMENTS & NON-GAAP FINANCIAL MEASURES
DAVE
GIROUARD
Co-founder & Chief Executive Officer
WELCOME
TODAY’S GOALS
Help you understand the size
of the AI lending opportunity
Help you understand why
Upstart will win
Convince you to buy our
stock and hold it forever
WHY WE DO THIS
WE DID NOT START AS AN
AI COMPANY
UPSTART’S SUCCESS IS A WIN
FOR CONSUMERS, LENDERS
AND THE ECONOMY
MISSION-DRIVEN
COMPANIES STOP AT NOTHING
TO ACHIEVE THEIR GOALS
is unimaginably large
THE MARKET
in credit originated globally each year
>$25 TRILLION
PERSONAL
COMMERCIAL
AND
INDUSTRIAL
AUTO
CARD
REAL
ESTATE
SMALL
BUSINESS
HOME
Based on internal estimates. Categories listed are sample lending opportunities.
AT L E A S T $1 TRILLION
in revenue is available to the platform that
originates and services that credit
Based on internal estimates.
AI WILL BRING UNPRECEDENTED CHANGE
AND DISRUPTION TO MOST MARKETS
EducationHealthcareTransportationMedia & EntertainmentFinancial Services
Defense & WarfareRetail & E-commerceEnergy & UtilitiesProfessional ServicesManufacturing
AI WILL BRING UNPRECEDENTED CHANGE
AND DISRUPTION TO MOST MARKETS
EducationHealthcareTransportationMedia & EntertainmentFinancial Services
Defense & WarfareRetail & E-commerceEnergy & UtilitiesProfessional ServicesManufacturing
CREDIT IS A
FIRST-TIER CANDIDATE
What can I help with?
AttachSearch
Reason
Voice
List the global industries that are likely to be most impacted by AI
List the global industries that are likely to be most impacted by AI
Here’s the list of global industries expected to be most impacted by AI, based on projected
economic value, automation potential, and transformative use cases:
1. Healthcare
•
Impact Areas: Diagnostics, drug discovery, personalized medicine, radiology,
operational efficiency.
•
Why: High data volume + high stakes = perfect fit for AI augmentation (not just automation).
2. Financial Services
•
Impact Areas: Fraud detection, underwriting, trading algorithms, robo-advisors,
customer service.
•
Why: Data-rich, regulated, and already tech-heavy-Al boosts speed and precision.
3. Retail & E-Commerce
•
Impact Areas: Recommendation engines, inventory forecasting, customer service bots,
dynamic pricing.
•
Why: Huge customer datasets + real-time decisions = massive AI value.
List the global industries that are likely to be most impacted by AI
2. Financial Services
•
Impact Areas: Fraud detection, underwriting, trading algorithms, robo-advisors,
customer service.
•
Why: Data-rich, regulated, and already tech-heavy-Al boosts speed and precision.
WHY DO LOANS
COST SO MUCH?
HOW DOES AI IMPACT
THE PRICE OF A LOAN?
Profit
Credit losses
People & data
Customer acquisition
Profit
HOW DOES AI IMPACT THE PRICE OF A LOAN?
50% REDUCTION
IN CAC DESPITE >5X GROWTH
1
Conversion improvements reduce CAC
Automation eliminates hard cost
(especially people)
66% REDUCTION
IN ONBOARDING PEOPLE COST/
LOAN
2
Economic wins can be shared between borrowers and lenders
Model accuracy drives more approvals
and lower rates
36% REDUCTION
IN APRS AT SAME APPROVAL RATES
3
RESULTS TO DATE
Pro
Credit losses
People & data
Customer acquisition
Profit
(1) Q1'25 vs. Q1’19.
(2) Q1'25 vs. CY22.
(3) Current vs. traditional model.
WHY DOES GETTING A
LOAN SUCK SO MUCH?
In-pe
r
son
c
losing
W
a
i
ting
f
or
a de
c
ision
Empl
o
y
er
v
e
r
i
fication calls
Documen
t
ation
r
equests
Endless
applications
HOW DOES AI IMPACT
THE BORROWING EXPERIENCE?
In-pe
r
son
c
losing
W
a
i
ting
f
or
a de
c
ision
Empl
o
y
er
v
e
r
i
fication calls
Documen
t
ation
r
equests
Endless
applications
In-pe
r
son
c
losing
W
a
i
ting
f
or
a de
c
ision
Empl
o
y
er
v
e
r
i
fication calls
Documen
t
ation
r
equests
Endless
applications
~90% OF UPSTART LOANS HAVE
NONE OF THESE
UPSTART IS BUILDING THE
FOUNDATION MODEL FOR CREDIT
SEPARATION
UPSTART IS BUILDING THE
FOUNDATION MODEL FOR CREDIT
AUTOMATION
SEPARATION
UPSTART IS BUILDING THE
FOUNDATION MODEL FOR CREDIT
CALIBRATION
AUTOMATION
SEPARATION
UPSTART IS BUILDING THE
FOUNDATION MODEL FOR CREDIT
GENERALIZATION
CALIBRATION
AUTOMATION
SEPARATION
UPSTART IS BUILDING THE
FOUNDATION MODEL FOR CREDIT
PERSONALIZATION
GENERALIZATION
CALIBRATION
AUTOMATION
SEPARATION
UPSTART IS BUILDING THE
FOUNDATION MODEL FOR CREDIT
2025
PRIORITIES
10x our advantage in AI
Prepare our funding supply for rapid growth
Return to GAAP profitability in 2H
Giant leaps toward best rates, best process for all
INCREDIBLE PROGRESS TOWARD “BEST RATES”
15
30
45
60
720+660-720<660
45%
47%
51%
UPSTART WIN RATES IN PERSONAL LOANS VERSUS DIGITAL PEERS
Closest peerUpstart
Source: Based on internal analysis of third-party data f rom 4/9-5/6.
51%
47%
45%
PAUL GU
Co-founder & Chief Technology Officer
OUR AI ADVANTAGE
PROFITCREDITGROWTH
Logistic Regression
Logistic regression is a common technique used to develop scorecards in most financial industry
applications, where the predicted variable is binary. In cases where the predicted variable is
continuous, such as modeling loss given default, linear regression is used. The rest of this section
will deal with using multiple logistic regression to predict a binary outcome (good/ bad).
Preliminary Scorecard
In general, the final scorecards produced in this stage consist of between 8 and 15 characteristics,
although in larger banks with a lot of data, I have seen those with 18 to 30 characteristics. This is
done to ensure a comprehensive risk profile that includes all the reasonable variables, as well as a
more stable scorecard whose predictive powers will be strong even if the profile of one or two
characteristics were to change. Scorecards with too few characteristics are generally unable to
withstand the test of time, as they are susceptible to minor changes in the applicant profile.
Logistic Regression
Logistic regression is a common technique used to develop scorecards in most financial industry
applications, where the predicted variable is binary. In cases where the predicted variable is
continuous, such as modeling loss given default, linear regression is used. The rest of this section
will deal with using multiple logistic regression to predict a binary outcome (good/ bad).
Logistic regression, like most other predictive modeling methods, uses a set of predictor
characteristics to predict the likelihood (or probability) of a specific outcome (the target). The
equation for the logit transformation of a probability of an event is shown by:
Logit (p
i
) = β
0
- β
I
x
I
+....+ β
k
x
k
In general, the final scorecards produced in this stage consist of between 8 and 15 characteristics,
although in larger banks with a lot of data, I have seen those with 18 to 30 characteristics. This is
done to ensure a comprehensive risk profile that includes all the reasonable variables, as well as a
more stable scorecard whose predictive powers will be strong even if the profile of one or two
characteristics were to change. Scorecards with too few characteristics are generally unable to
withstand the test of time, as they are susceptible to minor changes in the applicant profile.
Designing a Scorecard
While it is possible to build a scorecard by putting all the characteristics into the regression
model and generating a statistically optimal outcome, this method may not produce results that
Designing a Scorecard
While it is possible to build a scorecard by putting all the characteristics into the regression
model and generating a statistically optimal outcome, this method may not produce results that
are operationally ideal. The scorecard developer would typically rely on some statistical measures
such as p-values, Gini, chi-square, R-squared, and others to determine the quality of the outcome.
There are, however, some business goals that need to be considered when developing
scorecards.
continuous, such as modeling loss given default, linear regression is used. The rest of this section
will deal with using multiple logistic regression to predict a binary outcome (good/ bad).
Logistic regression, like most other predictive modeling methods, uses a set of predictor
characteristics to predict the likelihood (or probability) of a specific outcome (the target). The
equation for the logit transformation of a probability of an event is shown by:
Logit (p
i
) = β
0
- β
I
x
I
+....+ β
k
x
k
ScoreOdds
30030
30131
30233
30334
••
••
AgeDelq at BureauTime at Res
18-24200550-612
25-29321-2407-1825
30-37463-51219-3638
38-45566+537+50
45+80
such as p-values, Gini, chi-square, R-squared, and others to determine the quality of the outcome.
There are, however, some business goals that need to be considered when developing
scorecards.
ScoreOdds
30030
30131
30233
30334
••
••
32060
••
340120
AgeDelq at BureauTime at Res
18-24200550-612
25-29321-2407-1825
30-37463-51219-3638
38-45566+537+50
45+80
R/O BurdenRegionInq 6 mth
015Major Urban20050
1-940Minor Urban251-330
10-2530Rural154-515
26-5025
50-6920
70-8515
86-9910
100+5
0
0
0
0
0
0
31
TRAINING DATAPOINTS
20132014
0
0
0
0
0
0
31
TRAINING DATAPOINTS
Education variables
20132014
5,111
7,536
10,718
14,595
19,276
24,936
31,531
1,835
3,209
5,111
2015
TRAINING DATAPOINTS
2014
0.06M
Ensemble of algorithms
2013
101,986
120,666
141,220
163,642
187,828
213,385
239,906
70,990
85,396
101,986
201420152016
TRAINING DATAPOINTS
0.06M
0.12M
Instant approval
2013
429,261
466,560
505,552
546,207
588,407
633,623
682,486
359,729
393,690
429,261
2014201520162017
TRAINING DATAPOINTS
0.25M
0.6M
XGBoost with censoring
2013
1,100,430
1,192,974
1,293,367
1,398,110
1,506,118
1,617,930
1,735,513
935,682
1,014,655
1,100,430
20142015201620172018
TRAINING DATAPOINTS
1.2M
0.6M
Revenue science
2013
2,603,130
2,774,643
2,954,635
3,140,165
3,331,248
3,529,927
3,737,959
2,278,397
2,437,208
2,603,130
201420152016201720182019
TRAINING DATAPOINTS
3M
1.5M
2013
Loan-month
BUT LOANS ARE A
“SURVIVAL” PROBLEM
Cumulative default probability
0%
2%
4%
6%
8%
10%
12%
14%
Months
0102030405060
DEFAULT
CDFS
POST-LOAN MONTH
BORROWER A
12.65% Probability of default
AVERAGE TIMING CURVE
BORROWER B
12.65% Probability of default
OPTIMAL APR 16.70%
OPTIMAL APR 13.09%
2,603,130
2,774,643
2,954,635
3,140,165
3,331,248
3,529,927
3,737,959
2,278,397
2,437,208
2,603,130
201420152016201720182019
TRAINING DATAPOINTS
3M
1.5M
2013
Loan-month
5,286,411
5,610,899
5,962,012
6,337,137
6,727,755
7,112,578
7, 4 9 1 , 5 2 6
4,700,563
4,983,465
5,286,411
2014201520162017201820192020
TRAINING DATAPOINTS
3M
6M
Neural networks
2013
10,042,546
10,566,872
11,130,093
11,736,692
12,384,768
13,100,301
13,897,531
9, 0 9 5 , 2 74
9, 5 5 5 , 59 0
10,042,546
20142015201620172018201920202021
TRAINING DATAPOINTS
6M
12M
LDAs and SHAP
2013
20,701,366
22,259,813
23,966,263
25,777,893
27,670,579
29,645,983
31,674,562
18,003,683
19,289,044
20,701,366
201420152016201720182019202020212022
TRAINING DATAPOINTS
15M
30M
Proprietary loss function
2013
GENERIC LOSS FUNCTION
Loan that defaults in month 60 Loan that defaults in month 1
Loan that defaultsLoan that prepays
Loan predicted to default in
month 60 and defaults in month 59
Loan predicted to default in
month 1 and defaults in month 59
0.0900
0.0925
0.0950
0.0975
0.1000
0.1025
0.1050
0.1075
0.1100
0.040.050.060.070.080.090.100.110.12
Predicted default in month 1
Predicted default in month 60
Proprietary loss function
Generic loss function
0. 90
0. 94
0. 98
1.02
1.06
1.10
1.14
1.18
Accuracy on
fi
nancial objective
20,701,366
22,259,813
23,966,263
25,777,893
27,670,579
29,645,983
31,674,562
18,003,683
19,289,044
20,701,366
201420152016201720182019202020212022
TRAINING DATAPOINTS
15M
30M
Proprietary loss function
2013
43,831,287
45,815,630
47,764,837
49,700,997
51,602,936
53,472,105
55,317,835
39,798,572
41,823,890
43,831,287
2014201520162017201820192020202120222023
TRAINING DATAPOINTS
25 M
50M
Dynamic macro
2013
65,795,540
67,481,671
69,141,906
70,793,166
72,406,909
74 , 0 1 2 , 0 8 0
75,605,953
62,376,516
64,092,180
65,795,540
20142015201620172018201920202021202220232024
TRAINING DATAPOINTS
35M
70M
APR as a feature
2013
Borrower IDCredit scoreEducation
Bank
balance
Loan
amount
Time
Loan month
index
Outcome
10000714BA$4000$100002023-0110
10000714BA$4000$100002023-0220
10000714BA$4000$100002023-0330
10000714BA$4000$100002023-0440
........................
10000714BA$4000$100002025-04280
10000680HS$5000$75002024-0110
10000680HS$5000$75002024-0220
........................
10000680HS$5000$75002025-01131
UPSTART MATRIX COMPRESSION
Data repeats
Data is unique
UPSTARTXGB REDUCED
MEMORY CONSUMPTION 30-60X
Borrower IDCredit scoreEducation
Bank
balance
Loan
amount
Time
Loan month
index
10000714BA$4000$1000000
1000000002023-011
1000000002023-022
1000000002023-033
.....................
1000000002025-0428
10000680HS$5000$750000
1000000002024-011
.....................
1000000002025-0113
UPSTARTXGB CAN REDUCE
MEMORY CONSUMPTION 30-60X
Novel
“block sparse
matrix” format
ML algorithms
rewritten to
support this
data structure
65,795,540
67,481,671
69,141,906
70,793,166
72,406,909
74 , 0 1 2 , 0 8 0
75,605,953
62,376,516
64,092,180
65,795,540
20142015201620172018201920202021202220232024
TRAINING DATAPOINTS
35M
70M
APR as a feature
2013
87,686,453
89,396,912
91,124,212
201420152016201720182019202020212022202320242025
TRAINING DATAPOINTS
50M
100M
Embeddings
2013
PERSONALIZATION
GENERALIZATION
CALIBRATION
AUTOMATION
SEPARATION
WHY THIS MATTERS
SEPARATION
WHY THIS MATTERS
Loss variance
5
10
15
20
25
30
35
40
Credit score
625649664669674679685691696702708715720725732739750763777786798825
UpstartIndustry range (Min. to max.)
UPSTART VS INDUSTRY PRICING FOR RANDOM LOAN SAMPLE
Source: Based on internal analysis of third-party data f rom 4/9-5/6.
•
FICO 666
•
5.8 years of credit history
•
$61k annual income
5 YEARS
AGO
THE SCENARIO
TRADITIONAL MODEL
REJECT
Probability of default is 32%
Decision:
•
FICO 666
•
5.8 years of credit history
•
$61k annual income
5 YEARS
AGO
THE SCENARIO
•
JD from University of Pittsburgh
•
Lawyer for the City of Philadelphia
•
Low debt use relative to others
in similar circumstances
UPSTART MODEL
TRADITIONAL MODEL
REJECT
Probability of default is 32%
Decision:
APPROVE
Decision:
Probability of default is 2%
OUTCOME: 21 ON-TIME PAYMENTS
BEFORE FULLY PAYING OFF LOAN
Predicted default %
0.0000
0.0150
0.0300
0.0450
0.0600
Payment month
2019-06
2020-02
2020-10
2021-06
2022-02
2022-10
2023-06
2024-02
Predicted Prepayment %
0.0000
0.5000
1.0000
1.5000
2.0000
2.5000
OUTCOME: 21 ON-TIME PAYMENTS
BEFORE FULLY PAYING OFF LOAN
PREPAID HERE
Predicted prepaymentPredicted default
TRADITIONAL
MODEL
UPSTART
MODEL
TRADITIONAL
MODEL
UPSTART
MODEL
9%
DEFAULT RATE
5%
DEFAULT RATE
50%
APPROVAL RATE
50%
APPROVAL RATE
TRADITIONAL
MODEL
UPSTART
MODEL
27%
APPROVAL RATE
7.5%
DEFAULT RATE
7.5%
DEFAULT RATE
67%
APPROVAL RATE
AUTOMATION
SEPARATION
WHY THIS MATTERS
FULLY AUTOMATED LOANS
25%
50%
75%
100%
Q1'18
Q2'18
Q3'18
Q4'18
Q1'19
Q2'19
Q3'19
Q4'19
Q1'20
Q2'20
Q3'20
Q4'20
Q1'21
Q2'21
Q3'21
Q4'21
Q1'22
Q2'22
Q3'22
Q4'22
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
Q1'25
PERCENTAGE OF LOANS FULLY AUTOMATED
FULLY AUTOMATED LOANS
CONVERSION TO LOANS
25%
50%
75%
100%
Q4'20
Q1'21
Q2'21
Q3'21
Q4'21
Q1'22
Q2'22
Q3'22
Q4'22
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
Q1'25
Q2'25
% APPLICATIONS
AUTOMATICALLY APPROVED
25%
50%
75%
100%
Q4'20
Q1'21
Q2'21
Q3'21
Q4'21
Q1'22
Q2'22
Q3'22
Q4'22
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
Q1'25
Q2'25
% Applications automatically approved
% Applications manually reviewed
CALIBRATION
AUTOMATION
SEPARATION
WHY THIS MATTERS
CONSUMER CREDIT POST-COVID
Personal loanCardAuto
60-90 DAY DELINQUENCIES (TRANSUNION DATA)
0.4%
0.8%
1.2%
1.6%
2021-04
2021-06
2021-08
2021-10
2021-12
2022-02
2022-04
2022-06
2022-08
2022-10
2022-12
2023-02
2023-04
2023-06
2023-08
2023-10
2023-12
2024-02
2024-04
2024-06
2024-08
2024-10
2024-12
2025-02
1.5x
1.7x
1.4x
-0.3
-0.15
0
0.15
0.3
0.45
0.6
-7%
-16%
-15%
-8%-8%
0%
-7%
-5%
12%
28%
22%
-3%
-5%
16%
13%
21%
38%
30%
41%
55%
58%
59%
HOW OUR MODEL DID
2022-Q12022-Q22022-Q32022-Q42023-Q12023-Q22023-Q32023-Q42024-Q12024-Q32024-Q2
LOSS VARIANCE
59%58%
55%
41%
30%
38%
21%
13%
16%
-5%
-3%
-0.3
-0.15
0
0.15
0.3
0.45
0.6
-7%
-16%
-15%
-8%-8%
0%
-7%
-5%
12%
28%
22%
-3%
-5%
16%
13%
21%
38%
30%
41%
55%
58%
59%
TODAY’S MODEL
IN COMPARISON
2022-Q12022-Q22022-Q32022-Q42023-Q12023-Q22023-Q32023-Q42024-Q12024-Q32024-Q2
LOSS VARIANCE
59%58%
55%
41%
30%
38%
21%
13%
16%
-5%
-3%
22%
28%
12%
-5%
-7%
0%
-8%-8%
-15%
-16%
-7%
GENERALIZATION
CALIBRATION
AUTOMATION
SEPARATION
WHY THIS MATTERS
HELOC
AUTO
RETAIL
SDL
Education
variables
Ensemble of
algorithms
Instant
approval
Revenue
science
Loan
month
Neural
networks
LDAs and
SHAP
Proprietary
loss function
APR as a
feature
EmbeddingsDynamic
macro
MODEL FEATURES BY PRODUCT
AUTO
REFI
PERSONALIZATION
GENERALIZATION
CALIBRATION
AUTOMATION
SEPARATION
WHY THIS MATTERS
TARGETED LOAN
MODIFICATIONS
TARGETED
HIGH-TOUCH
INTERVENTIONS
TARGETED
LOW-COST
INTERVENTIONS
TARGETED DEBT
SETTLEMENT
OPTIMIZED
PAYMENT DUE
DATES
AGENT &
BORROWER
MATCHING
AI AGENT
SERVICING
*Approximate
AT C O N S TA N T C O S T
*
REDUCTION IN
CHARGE-OFF
1%
INCREASE
IN REVENUE
1%
Short-term credit
Unsecured installmentAuto
Home
CREDITGROWTHPROFIT
CHANTAL
RAPPORT
Chief Marketing Officer & SVP of Growth
OUR CATEGORY-OF-ONE
BUSINESS
CREDIT
MATTERS
Banks losing market share
NEW REALITIES OF LENDING
Banks losing market share
NEW REALITIES OF LENDING
Loans will be shopped
Banks losing market share
NEW REALITIES OF LENDING
Loans will be shopped
AI is a commitment
Banks losing market share
NEW REALITIES OF LENDING
Loans will be shopped
AI is a commitment
Private credit emerges
THE BUSINESS MODEL THAT WINS
Intelligent
marketplace
Winning
brand
Leading
foundation model
LEADING
FOUNDATION
MODEL
UPSTART’S WINNING BUSINESS MODEL
WHAT ARE THE PROOF POINTS
OF LEADING AI?
Accelerated learnings
Positive selection
Pricing power
Superior product
Personal loan take rate
3%
5%
8%
10%
Period
20212024
Fee%
Accelerated learnings
Positive selection
Pricing power
Superior product
WHAT ARE THE PROOF POINTS
OF LEADING AI?
Loss variance
5
10
15
20
25
30
35
40
Credit score
625649664669674679685691696702708715720725732739750763777786798825
WHAT ARE THE PROOF POINTS
OF LEADING AI?
Accelerated learnings
Positive selection
Pricing power
Superior product
Source: Based on internal analysis of third-party data f rom 4/9-5/6.
Industry range (Min. to max.)Upstart
Loss variance
5
10
15
20
25
30
35
40
Credit score
625649664669674679685691696702708715720725732739750763777786798825
Upstart
Loss variance
5
10
15
20
25
30
35
40
Credit score
625649664669674679685691696702708715720725732739750763777786798825
Upstart
WHAT ARE THE PROOF POINTS
OF LEADING AI?
Accelerated learnings
Positive selection
Pricing power
Superior product
Source: Based on internal analysis of third-party data f rom 4/9-5/6.
Upstart
Loss variance
5
10
15
20
25
30
35
40
Credit score
625649664669674679685691696702708715720725732739750763777786798825
WHAT ARE THE PROOF POINTS
OF LEADING AI?
Accelerated learnings
Positive selection
Pricing power
Superior product
Source: Based on internal analysis of third-party data f rom 4/9-5/6.
Loss variance
5
10
15
20
25
30
35
40
Credit score
625649664669674679685691696702708715720725732739750763777786798825
Upstart
WHAT ARE THE PROOF POINTS
OF LEADING AI?
Accelerated learnings
Positive selection
Pricing power
Superior product
1
% Increase
in accuracy
WHAT ARE THE PROOF POINTS
OF LEADING AI?
:1
% Increase
in accuracy
13
% Increase
in conversion
Accelerated learnings
Positive selection
Pricing power
Superior product
WHAT ARE THE PROOF POINTS
OF LEADING AI?
.
More
consumers
Better
product
Increased
conversion
Faster learnings
Accelerated learnings
Positive selection
Pricing power
Superior product
WHAT ARE THE PROOF POINTS
OF LEADING AI?
BEST RATES,
BEST PROCESS
FOR ALL
Accelerated learnings
Positive selection
Pricing power
Superior product
MARKETPLACE
CAPITAL
UPSTART’S WINNING BUSINESS MODEL
LENDER
CONSUMER
COMPETITION
AMONG
DIVERSE
CAPITAL
SOURCES
INTELLIGENT MARKETPLACE
MARKETPLACE AUCTION CREATES
STRUCTURAL ADVANTAGES
SCALABLE
Serve more
borrowers across
more credit tiers
FLEXIBLE
Deploy more
adaptable capital
sources
EFFICIENT
Optimize rates
and returns
dynamically
THE WINNING
BRAND
UPSTART’S WINNING BUSINESS MODEL
CURRENT PRODUCT COVERAGE
Super
prime
Prime
Non
prime
Home
Relief loans
Auto refiPersonal
SOMETHING
FOR
EVERYONE
BRAND ADVANTAGE
76%
of applicants have cars
40%
of applicants own homes
30%
of borrowers are likely to take
out another loan within 5 years
OF OUR CUSTOMERS...
LENDING FOR
A LIFETIME
BRAND ADVANTAGE
Source: Based on internal analysis.
BRAND ADVANTAGE
WE LIVE
FOR SPEED
>90% OF
PERSONAL LOANS
APPROVED
AUTOMATICALLY
BRAND ADVANTAGE
Source: Based on internal analysis.
THE FIRST INSTANT
AUTO REFINANCE
BRAND ADVANTAGE
Source: Based on internal analysis.
HELOC
>2X FASTER THAN
THE INDUSTRY
BRAND ADVANTAGE
Source: Based on internal analysis.
THE REVIEWS
SPEAK FOR
THEMSELVES
BRAND ADVANTAGE
The process was very easy, quick, & legit!
The process was very easy & quick...[and] the interest rate!!!
I haven't seen an interest rate offer this good in YEARS. I
will be able to pay off debts without my payment being
eaten up by interest. They have won me over as a customer
& I will definitely tell others.
The Upstart loan process for credit...
The Upstart loan process for credit card consolidation was
super quick and easy. I received a better rate than another
UNLIMITED POTENTIAL
AUTOMOBILE
REFINANCING
BRINGING IT TOGETHER
A $0 BILLION MARKET?
MEET
FRAN
11:53 AM
Applies for a personal loan
11:53 AM
Applies for a personal loan
12:03 PM
Loan processed
11:53 AM
Applies for a personal loan
12:03 PM
Loan processed
12:05 PM
Applies for an auto refi loan
11:53 AM
Applies for a personal loan
12:03 PM
Loan processed
12:05 PM
Applies for an auto refi loan
12:14 PM
Refinances her car
FEEDBACK
I am extremely grateful to you. You can't
imagine how much help you have given me.
I would recommend you to all my family
and friends without a doubt the best.
Thank you.
ACCELERATING
EARLY
PRODUCT
GROWTH
Q1 2024
1x
Q1 2025
19x
Auto refi originations
WHERE WE ARE TODAY
TRUSTED BY MILLIONS
Over 3M borrowers served
CATEGORY-LEADING AI
8x better risk separation
MARKETPLACE AUCTION
100+ partners funding full credit spectrum
EXPANDING PRODUCT BREADTH
4+ lending products
GROWING LIFETIME VALUE
0% CAC second loans
WHERE WE ARE TODAY
TRUSTED BY MILLIONS
Over 3M borrowers served
CATEGORY-LEADING AI
8x better risk separation
MARKETPLACE AUCTION
100+ partners funding full credit spectrum
EXPANDING PRODUCT BREADTH
4+ lending products
GROWING LIFETIME VALUE
0% CAC second loans
OUR VISION
ALWAYS-ON CREDIT
For 100% of Americans
EVERYTHING STORE FOR CREDIT
To borrow, manage, and optimize all credit
GENERATIONAL BRAND
From first loan to life-long relationship
A CATEGORY OF ONE
We are peerless
THE FUTURE OF CREDIT
IS HERE
SANJAY
DATTA
Chief Financial Officer
PROFITABILITY AND
RESILIENCE
THE MARKET OPPORTUNITY IS VAS T
WHY DO LOANS COST SO MUCH?
WHY DOES GETTING A LOAN
SUCK SO MUCH?
In-pe
r
son
c
losing
W
a
i
ting
f
or
a de
c
ision
Empl
o
y
er
v
e
r
i
fication calls
Documen
t
ation
r
equests
Endless
applications
Credit losses
People & data
Customer acquisition
Profit
UPSTART IS BUILDING THE
FOUNDATION MODEL FOR CREDIT
PERSONALIZATION
GENERALIZATION
CALIBRATION
AUTOMATION
SEPARATION
BEST RATES, BEST PROCESS FOR ALL
OUR CATEGORY-OF-ONE BUSINESS
WHERE WE ARE TODAY
TRUSTED BY MILLIONS
Over 3M borrowers served
CATEGORY-LEADING AI
8x better risk separation
MARKETPLACE AUCTION
100+ partners funding full credit spectrum
EXPANDING PRODUCT BREADTH
4+ lending products
GROWING LIFETIME VALUE
0% CAC second loans
OUR VISION
ALWAYS-ON CREDIT
For 100% of Americans
EVERYTHING STORE FOR CREDIT
To borrow, manage, and optimize all credit
GENERATIONAL BRAND
From first loan to life-long relationship
A CATEGORY OF ONE
We are peerless
Growth
Operating leverage
Profits
Pricing power
Margins
UPSTART AI + BUSINESS MODEL
= UNIQUE FINANCIAL PROFILE
UPSTART AI + BUSINESS MODEL
= UNIQUE FINANCIAL PROFILE
0%
5%
10%
15%
20%
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
Q1'25
CONVERSION RATE
Growth
Operating leverage
Profits
Pricing power
Margins
UPSTART AI + BUSINESS MODEL
= UNIQUE FINANCIAL PROFILE
0%
2.5%
5%
7.5%
10%
Q1'22
Q2'22
Q3'22
Q4'22
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
PERSONAL LOAN TAKE RATE
Growth
Operating leverage
Profits
Pricing power
Margins
UPSTART AI + BUSINESS MODEL
= UNIQUE FINANCIAL PROFILE
Growth
Operating leverage
Profits
Pricing power
Margins
CAC
1%
1.5%
2%
2.5%
3%
CAC AND % FULLY AUTOMATED
% Fully automated
60%
70%
80%
90%
100%
Q1'21
Q2'21
Q3'21
Q4'21
Q1'22
Q2'22
Q3'22
Q4'22
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
% Fully automatedCAC %
0%
20%
40%
60%
80%
Q1'21
Q2'21
Q3'21
Q4'21
Q1'22
Q2'22
Q3'22
Q4'22
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
CONTRIBUTION MARGIN
Growth
Operating leverage
Profits
Pricing power
Margins
CAC
1%
1.5%
2%
2.5%
3%
CAC AND % FULLY AUTOMATED
% Fully automated
60%
70%
80%
90%
100%
Q1'21
Q2'21
Q3'21
Q4'21
Q1'22
Q2'22
Q3'22
Q4'22
Q1'23
Q2'23
Q3'23
Q4'23
Q1'24
Q2'24
Q3'24
Q4'24
% Fully automated
CAC %
UPSTART AI + BUSINESS MODEL
= UNIQUE FINANCIAL PROFILE
Growth
Operating leverage
Profits
Pricing power
Margins
MONTHLY FEE REVENUE VS. FIXED COST
$ in millions
$20
$40
$60
$80
2024-01
2024-02
2024-03
2024-04
2024-05
2024-06
2024-07
2024-08
2024-09
2024-10
2024-11
2024-12
2025-01
2025-02
2025-03
Fixed costRevenue from fees
Fixed costs capture headcount-related costs along with recurring business expenses that tend to remain constant regardless of origination volume.
UPSTART AI + BUSINESS MODEL
= UNIQUE FINANCIAL PROFILE
Growth
Operating leverage
Profits
Pricing power
Margins
,
NET INCOME (LOSS): STEADY PROGRESS LAST 5 QUARTERS
.
$-70M
$0M
Q1'24
Q2'24
Q3'24
Q4'24
Q1'25
OUR TECHNOLOGY + BUSINESS MODEL
= UNIQUE FINANCIAL PROFILE
Growth
Operating leverage
Profits
Pricing power
Margins
Low capital
intensity
BUSINESS MODEL RESILIENCE
Borrower diversity
Originations ($M)
$0
$60
$120
$180
$240
Monthly from 2024-03 to 2025-03
2024-03
2024-04
2024-05
2024-06
2024-07
2024-08
2024-09
2024-10
2024-11
2024-12
2025-01
2025-02
2025-03
Product diversity
Originations ($M)
$0
$18
$35
$53
$70
Q1 '24Q2 '24Q3 '24Q4 '24Q1 '25
$61
$43
$26
$18
$13
AUTO LOANS
Originations ($M)
$0
$13
$25
$38
$50
Q1 '24Q2 '24Q3 '24Q4 '24Q1 '25
$41
$27
$17
$8
$7
HOME LOANS
Funding resilience
0%
2022
35%
2023
65%
2024
Macro-tooling
CUMULATIVE LOSS VARIANCE BY ORIGINATION HALF
-30%
-15%
0%
15%
30%
45%
60%
H1'22H2'22H1'23H2'23H1'24H2'24
-2%
-16%
-8%
-4%
4%
25%
5%5%
17%
34%
48%
59%
ActualWith current tools
$7
$8
$17
$27
$41
$61
$43
$26
$13
$18
$13
59%
48%
34%
17%
5%5%
25%
4%
-4%
-8%
-16%
-2%
59%
48%
34%
17%
5%5%
25%
4%
-4%
-8%
-16%
-2%
Originations ($M)
$0
$60
$120
$180
$240
Monthly from 2024-03 to 2025-03
2024-03
2024-04
2024-05
2024-06
2024-07
2024-08
2024-09
2024-10
2024-11
2024-12
2025-01
2025-02
2025-03
SUPER-PRIME ORIGINATIONS (FICO >720)
BORROWER DIVERSITY
T-Prime launched
Originations refers to Transaction Volume, Dollars. Chart refers to Personal Loans.
Originations ($M)
$0
$18
$35
$53
$70
Q1 '24Q2 '24Q3 '24Q4 '24Q1 '25
$61
$43
$26
$18
$13
AUTO LOANS
Originations ($M)
$0
$13
$25
$38
$50
Q1 '24Q2 '24Q3 '24Q4 '24Q1 '25
$41
$27
$17
$8
$7
HOME LOANS
PRODUCT DIVERSITY
$7
$8
$17
$27
$41
$61
$43
$26
$13
$18
FUNDING RESILIENCE
35%
2023
65%
2024
0%
2022
>50%
of the credit
underperformance
experienced would
have been avoided
MACRO-TOOLING
CUMULATIVE LOSS VARIANCE BY ORIGINATION HALF
Loss variance
-30%
-15%
0%
15%
30%
45%
60%
H1'22H2'22H1'23H2'23H1'24H2'24
-2%
-16%
-8%
-4%
4%
25%
5%5%
17%
34%
48%
59%
ActualWith current tools
59%
48%
34%
17%
5%5%
25%
4%
-4%
-8%
-16%
-2%
FUTURE
REVENUE
MODELS
RATABLE FEE
REVENUE
SUBSCRIPTIONS
SERVICING
FOR ALL
REVOLVING
THANK YOU
RECONCILIATION OF NON-GAAP FINANCIAL MEASURES
Fiscal 2021Fiscal 2022Fiscal 2023Fiscal 2024
Q1 2021Q2 2021Q3 2021Q4 2021Q1 2022Q2 2022Q3 2022Q4 2022Q1 2023Q2 2023Q3 2023Q4 2023Q1 2024Q2 2024Q3 2024Q4 2024
Revenue from fees, net
$116.2$187.3$210.4$287.4$314.0$258.3$179.3$155.6$117.1$143.7$146.8$152.8$138.1$130.5$167.6$199.3
Loss from operations
$15.6$36.3$28.6$60.4$34.8-$32.1-$58.1-$58.5-$131.8-$33.3-$43.8-$47.5-$67.5-$55.5-$45.2-$4.8
Operating margin
13%19%14%21%11%-12%-32%-38%-113%-23%-30%-31%-49%-43%-27%-2%
•Sales and marketing, net of borrower acquisition costs
$3.3$5.0$6.2$11.4$9.6$11.9$10.3$11.2$11.7$4.8$9.4$10.6$10.3$9.7$10.5$11.2
•Customer operations, net of borrower verification and servicing costs
$3.1$4.5$7.6$6.6$6.1$6.9$7.7$9.5$10.8$8.1$7.9$7.0$7.3$7.5$6.8$7.5
•Engineering and product development
$19.0$31.4$37.1$46.5$50.0$57.0$66.2$64.0$110.1$58.0$54.9$57.2$63.1$58.5$64.9$67.2
•General, administrative, and other
$20.0$26.1$34.4$42.1$43.5$46.9$47.8$47.1$52.7$50.4$53.5$55.8$57.6$53.0$59.9$60.4
•Interest income, interest expense, and fair value adjustments, net
-$5.2-$6.6-$18.0-$17.5$3.8$30.2$22.1$8.7$14.2$7.9$12.2$12.5$10.3$2.9$5.5-$19.7
Contribution profit
$55.8$96.7$95.9$149.5$147.8$120.9$96.0$82.0$67.6$95.9$94.2$95.6$81.1$76.1$102.4$121.9
Contribution margin
48%52%46%52%47%47%54%53%58%67%64%63%59%58%61%61%
($ IN MILLIONS)