# UPST: forecasting Q3 revenue

I was asked to post this to the board.
I do actually think it would be good to crowdsource everybody’s input.

Before I begin, however, major obvious caveat:

These projections could be totally wrong. The last thing I want is for a reader to make investment decisions based off this. Everyone needs to do their own analyses of the business fundamentals of a company and NOT buy/sell stock just because some random guy on the internet spouted off numbers that sounded appealing.

Plus, the last mathematics course I ever took was in high school, and I have zero tech background. That should not instill much confidence in my ‘forecasts’.

Let’s treat this as a purely ‘fun’ exercise.

With that out of the way, my opinion is that UPST’s management is sandbagging their Q3 estimates.
Management is guiding for 205M to 215M for Q3.

In Q2 conference call, it was stated “June was our first month with more than 100,000 loans”
So let’s assume exactly 100000 loans were transacted in June.
We know they did 286864 total loans in Q2.
This leaves 186864 for the two other months.
If we assume a “proportional” growth in loan numbers each month (5% growth each month), then we can estimate roughly 95550 in May and 91000 in April.

If we assume identical growth into Q3, at 5% increase each further month.
This would be: 105000 in July, 110250 in August, 115762 in September.
That’s 331012 total loans for Q3.
At roughly \$670 revenue per loan (\$670 comes from 194M of Q2 rev divided by Q2 loans of 286864), that comes to \$221 million rev. That’s a 2.8% beat on the top end of their guidance.

Here are my projections via other estimations.

Semrush data is broken down into their estimate of upstart.com organic/paid traffic via desktop and mobile for the month of August. I have checked each day, and the numbers fluctuate, so Semrush is performing real time estimations. The month is almost over, so surely it will have a final estimation of August by Sept 1.
For today, here is what it is showing: 284864 desktop + 254235 mobile = 539099 total traffic for August.
This is in comparison to Semrush’s estimation of: June 390688, July 509122

Let us assume 100000 loans in June were transacted.

July had 30.3% increase in traffic. Let us assume the conversion rates and application rates are identical. Then we can estimate July had 130314 loans.

August would then be 37.98% increase in traffic and thus loans over June, so we estimate 137987 loans in August.

Let’s assume September is identical to June, to be extremely conservative. So assume 100000 loans in September.

If we add this up, we project 368301 loans. That’s revenue of 246.76M in Q3, at \$670 per loan.

If we were euphoric and estimate Septmeber is identical to August at 137987 loans, then we’re looking at 406288 in Q3 with revenue of 272M. That’s 40% sequential growth from Q2. It seems too ridiculous so I’m going to go with the above conservative estimate.
We especially have to keep in mind a lot of traffic in August may be noise from investors like us visiting the upstart website after their earnings report

Also, hats off to Pavlos21 for introducing Semrush to the board, as I didn’t even know they existed previously, or what SEO even was. Without that post https://discussion.fool.com/semrush-semr-q2-earningsthoughts-349… I wouldn’t know to use their platform.

Moving on to Google Trends data.

“Upstart” appears to be the best predicting search term. I have tried other terms such as “upstart login”, “upstart loan” or “upstart loans” and I found the all-encompassing “upstart” was still the best at correlating with prior quarterly results.

And I’ve so far found the time series data set that seems to fit well with prior results is Q1 2020 to present (up to 8/22/2021 data).

Since Google Trends only gives relative search term popularity, I take the area under the curve to estimate that as directly proportional to absolute search volume which should correlate strongly to actual traffic and loan application numbers. From there, I adjust for conversion rate numbers depending on how I want to analyze it (I’ve created ‘coefficients’ to estimate application numbers and stuff, but I didn’t include those results in this post as they weren’t conservative appearing).

The area under curve numbers I get are:
Q1 20 323
Q2 20 230
Q3 20 331.5
Q4 20 469
Q1 21 495
Q2 21 766
Q3 21 569.5

A quick back of the napkin math shows:

The 569.5 number is then divided by (53 / 92), because 53 days had elapsed of the 92 days in Q3 (as the data was up to date to 8/22/2021). That estimates 988 for Q3.
988 is about 29% bigger than Q2’s search figure of 766.
Assuming identical conversion and application rates, then Q3 would have 1.29 x 286864 = 368940 loans

That’s 247.19M revenue estimate, similar to the Semrush “conservative estimation” above.

Again, all of this should be with a HUGE grain of salt.

There’s a lot of noise in Google Trends data for the term “upstart”, as we can see it correlating to terms “UPST”, “upstart stock” which is traffic coming from investors like us. So we definitely should remain conservative on Q3 estimates. (See https://i.imgur.com/A9vHsqX.png )

But the underlying increase in popularity for searching “upstart” from actual borrowers is very much real, based on “upstart login” or “upstart loan” search trends (see https://i.imgur.com/EQ65OCW.png )

Google trends also gives me a huge range of estimations depending on the time series of data pulled and depending on how I try to account for conversion rate.

And I could not get last year 2020 Semrush data to fit very well with previous quarters data, so I really don’t know how accurate Semrush is in the long run.

I do have some other ideas, including using the exact URL that goes to Upstart’s application. (That would remove extraneous traffic from investors like us, but I think I need a full Semrush account to get that level of detail.)
Also I did try to use Trustpilot data but that was not predictive at all when ‘backtested’ on previous quarterly results.

Finally here’s a fun graph of Upstart’s google trend popularity over time, using the ‘R’ statistic program: https://i.imgur.com/Z8kajYV.png
I don’t know how to code at all, so credit goes to a friend.

If anyone has any criticism, suggestions, tweaks, or other fun ideas to try to predict Q3’s revenue, please post! I hope it will be an entertaining discussion.

134 Likes

Hi jonwayne235!

Thank you for all of your good work on Upstart.

UPST will sandbag expectations, but likely not when the facts are in.

I was fortunate to find the company in early January and have been heavily invested since.

Regarding sandbagging, the annual 2021 revenue expectations was only \$279M before March. The 2020 Q4 report was not released until March 17, wherein UPST revised 2021 revenue up to \$500M causing the stock to double the next day.

But in that 3/17 report, management also offered very accurate guidance for Q1. With only 14 days remaining in the quarter. UPST guided to just 4 million under actual. IIRC, guide was for \$118M, while actual of \$122M was reported for Q1 on 5/11 when, as we all know, revenue projection was raised another \$100M to \$600M.

Point is, when they can justify the uncertainty, management goes quite conservative. But UPST reports their quarterly results rather late, which might be one reason most sandbagging is divulged in upped annual revenue, rather than for the subsequent quarter.

Tidbit on UPST/Hochfeld: The 3/17 Q report that raised earnings guidance from \$279M to \$500M so astounded the market that the stock exploded the next day from \$60 with follow through to \$165 in less than a week. That pop induced Bert Hochfeld to withhold his impending recommendation on UPST. “What’s the point” he said to me. Finally, on April 20, after the stock dropped back under \$100, Bert released his well received “Upstart Holdings: The Name Says A Lot About This Fintech Disruptor.”

Thanks again jon. It is apparent that your excellent UPST research and contributions here have provided reassurance and increased conviction among many of us.

33 Likes

addedupon, Thank you for the kind words.

I agree they are more likely to be sandbagging guidance at 51 days remaining in a quarter versus 14 days when they have better visibility on the quarter. Of course, there’s only been 3 quarterly reports since IPO, but it makes sense. I believe they will ultimately do 850M to 900M in 2021 revenue.

I think much of the guidance revisions each quarter is a mix of a surprise to them - their AI model updates are just working better than expected.
They’ve mentioned in earnings calls/fireside chats they always aim to be ‘accurate’ in predicting the effect size of each update, but sometimes the performance delivered is greater than anticipated.
Of course, there’s always a risk that tweaking the model actually leads to lower approval rates too, which the CEO had said happened one time in the past.
The updates since IPO fortunately have been wins.

From the Q2 call:
"In the second quarter, we continued to drive separation between our AI-powered platform and more conventional lending systems. We eliminated a rules-based constraint in our model that handled situations related to the size of loan requested.

With more powerful algorithms and growth in training data, our models can now handle that issue natively and with more precision. This led to a boost in approval rates and a more accurate system overall.

We also recalibrated our acquisition models to harmonize them with the funnel improvements we experienced earlier in the year. In other words, our models for digital and off-line acquisition caught up to the most recent funnel wins that drove our earlier growth and began to target applicants they would have previously ignored or misprioritized.

They were also recalibrated to broader bank partner eligibility criteria, enabling us to market to more consumers."

8 Likes

jonwayne235,

Really great post! I use Google search to navigate to my companies’ investor relations webpages. Specifically, I use “[TICKER] investor relations”. This doesn’t work for UPST because when I type UPST it gets changed to UPS. So for UPST I must use “Upstart investor relations”.

If the oat of a Semrush subscription is keeping you from accessing that data, I’ll happily contribute to the cost.

GR

3 Likes

GauchoRico,

Thanks! I did actually trial a ‘pro’ account at Semrush but that didn’t seem to provide any extra ability to estimate traffic towards exact URL links within upstart such as the borrower applications page.

When I overlay “UPST”, “upstart stock”, “Upstart IR” “upstart investor relations” etc in Google Trends versus “Upstart” it certainly does show that a lot of “upstart” search traffic is likely due to investors like us.
I’m just not sure how to account for the investor search traffic since the data Google Trends gives us is all ‘relative’ popularity and not absolute search volumes.

The only term that truly filters out all of that is “Upstart login”, I think we just have to wait until after a few more quarters to really become the best at correlating to loan transaction volumes.

I suspect the reason “upstart login” has not been very predictive when backtested on past quarterly data while the term “upstart” is predictive (so far), is because Upstart didn’t IPO until Dec 2020.
So a lot of search for the term “upstart” last year was truly all coming from borrowers, since investors like us couldn’t really buy UPST back when it wasn’t public and thus had no interest in searching for upstart’s website.

And since upstart wasn’t as popular (relatively) among borrowers until this year, the search volume for “upstart login” in 2020 was just too little and too volatile to correlate with past quarters

6 Likes

According to Similarweb, Upstart website traffic increased from 2.2 to 2.5 million visits in August:
https://www.similarweb.com/website/upstart.com/#overview
Also, in August they got more visitors from nerdwallet.com than from creditkarma (26.6% vs 24.9%).

In July they got 33.7% of visitors from creditcarma and 22.3% from nerdwallet.

So, it looks like they’ve got more traffic in August and solved a possible problem with creditkarma being the biggest marketing channel.

However, the number of reviews on TrustPilot (https://www.trustpilot.com/review/www.upstart.com/transparen…) decreased from 2176 in July to 1945 in August - interesting, why. Maybe they ask some borrowers to leave a review on some other service.

6 Likes

jonwayne235 : Moving on to Google Trends data.

“Upstart” appears to be the best predicting search term. I have tried other terms such as “upstart login”, “upstart loan” or “upstart loans” and I found the all-encompassing “upstart” was still the best at correlating with prior quarterly results.

And I’ve so far found the time series data set that seems to fit well with prior results is Q1 2020 to present (up to 8/22/2021 data).

Since Google Trends only gives relative search term popularity, I take the area under the curve to estimate that as directly proportional to absolute search volume which should correlate strongly to actual traffic and loan application numbers. From there, I adjust for conversion rate numbers depending on how I want to analyze it (I’ve created ‘coefficients’ to estimate application numbers and stuff, but I didn’t include those results in this post as they weren’t conservative appearing).

The area under curve numbers I get are:
Q1 20 323
Q2 20 230
Q3 20 331.5
Q4 20 469
Q1 21 495
Q2 21 766
Q3 21 569.5

A quick back of the napkin math shows:

The 569.5 number is then divided by (53 / 92), because 53 days had elapsed of the 92 days in Q3 (as the data was up to date to 8/22/2021). That estimates 988 for Q3.
988 is about 29% bigger than Q2’s search figure of 766.
Assuming identical conversion and application rates, then Q3 would have 1.29 x 286864 = 368940 loans

This back of napkin math came out to be quite close the actual number of loans at 362,780

Kudos to jonwayne235 for sharing his thoughts and tons of analysis about Upstart all through.
Greatly appreciated!!

17 Likes